Vaccine Hyporesponse Induced by Individual Antibiotic Treatment in Mice and Non-Human Primates Is Diminished upon Recovery of the Gut Microbiome

Abstract

Emerging evidence demonstrates a connection between microbiome composition and suboptimal response to vaccines (vaccine hyporesponse). Harnessing the interaction between microbes and the immune system could provide novel therapeutic strategies for improving vaccine response. Currently we do not fully understand the mechanisms and dynamics by which the microbiome influences vaccine response. Using both mouse and non-human primate models, we report that short-term oral treatment with a single antibiotic (vancomycin) results in the disruption of the gut microbiome and this correlates with a decrease in systemic levels of antigen-specific IgG upon subsequent parenteral vaccination. We further show that recovery of microbial diversity before vaccination prevents antibiotic-induced vaccine hyporesponse, and that the antigen specific IgG response correlates with the recovery of microbiome diversity. RNA sequencing analysis of small intestine, spleen, whole blood, and secondary lymphoid organs from antibiotic treated mice revealed a dramatic impact on the immune system, and a muted inflammatory signature is correlated with loss of bacteria from Lachnospiraceae, Ruminococcaceae, and Clostridiaceae. These results suggest that microbially modulated immune pathways may be leveraged to promote vaccine response and will inform future vaccine design and development strategies.

1. Introduction

Vaccination plays an essential role in global health by contributing to prevention and eradication of infectious diseases. When individuals are immunized with vaccine antigens, the body mounts an immune response that leads to long-term immune memory and protection against future infection. However, many vaccines still suffer from relatively low response rates in some populations or geographical locations [1,2,3,4,5]. The key immunological mechanisms underlying this lack of response are not well understood, which limits progress in improving vaccine outcomes. Investigating the pathways involved in such vaccine immune hyporesponsiveness may inform new solutions to increasing vaccine efficacy and facilitate greater protection in at-risk populations [6,7].

Emerging evidence in clinical and preclinical studies suggests that vaccine hyporesponse may be linked to changes in the human microbiome—the community of microbes that live in and on the human body [8]. Found in areas such as the gut, the skin, and respiratory and genitourinary mucosal surfaces, the microbiota are a rich source of nutrients, biologically active metabolites, and immunoregulatory molecules [9,10]. The microbiome plays a key role in numerous aspects of human biology [11,12,13], and perturbation of the gut microbiome has a significant influence on various disease outcomes [14,15,16]. In particular, natural or antibiotic-induced dysbiosis of the gut microbiome has been associated with vaccine hyporesponse in humans [2,17,18,19,20,21]. Preclinical studies in antibiotic-treated or germ-free mice also support a connection between the microbiota and vaccine response [22,23,24,25].

Although multiple lines of evidence demonstrate a link between the microbiome and response to a variety of vaccines, we do not fully understand the underlying microbial and immunological mechanisms that dictate the breadth and longevity of vaccine-induced antigen-specific immune responses, nor a direct influence of the microbiome on such responses. Studies have suggested some correlation between microbiome and vaccine responses (e.g., inflammation and microbial secondary bile acids) [2,22,25,26,27,28], but they have yet to focus on microbiome perturbation with recovery dynamics in mice or non-human primates (NHPs). Here, we begin to address this knowledge gap by characterizing the dynamics of vaccine hyporesponse associated with antibiotic-mediated microbiome perturbation and recovery in mice as well as non-human primates.

We report that short-term oral treatment of individual antibiotics prior to parenteral vaccination impacted antigen-specific immune responses, and that the extent of vaccine hyporesponse correlated with the extent of microbiome disruption across different antibiotics. Specifically, we found that vancomycin, a glycopeptide antibiotic that specifically targets Gram-positive bacteria, robustly impacted vaccine-induced antigen-specific IgG responses. Furthermore, we observed that vancomycin treatment diminished IgG responses against multiple vaccines including an adjuvanted model sub-unit vaccine antigen (i.e., ovalbumin), as well two clinically relevant vaccines (i.e., hepatitis B virus vaccine and influenza vaccine). We report data to suggest that natural recovery of microbiome diversity over time after antibiotic treatment was sufficient to mitigate vancomycin-induced vaccine hyporesponse in both murine and non-human primates. We further identified immune signatures associated with potential commensal bacteria belonging to the taxa Lachnospiraceae, Ruminococcaceae, and Clostridiaceae. These results further support the link between the microbiome and immune system and also provide an improved understanding of the dynamic connection between recovery of the microbiome and vaccine response. Such an improved understanding of host and microbiome-derived factors is critical for optimal outcomes after vaccination and has the potential to influence the future design of vaccines and immunotherapies.

2. Materials and Methods

2.1. Mice

All animal work was approved by the Institutional Animal Care and Use Committee (IACUC) of Merck & Co., Inc., Kenilworth, NJ 07033, USA in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals. Specific pathogen-free (SPF) female C57BL/6 and BALB/c mice, aged 5–9 weeks, were purchased from Taconic Biosciences (Rensselaer, NY, USA). Mice were acclimated for 1–2 weeks before experimental treatments.

2.2. Rhesus Macaques

Animal studies involving rhesus macaques were performed at the New Iberia Primate Research Center (NIRC-New Iberia, LA 70503, USA) in accordance with relevant guidelines using protocols approved by NIRC and the Institutional Animal Care and Use Committee of Merck & Co., Inc., Kenilworth, NJ 07033, USA. Female Indian rhesus macaques were weight and age matched for each group. All animals were influenza naïve prior to study start. For each antibiotic course, vancomycin (15 mg/kg) was delivered orally once daily for 14 days. Fluzone quadrivalent influenza vaccine (Sanofi Pasteur, Lyon, France; 2018–2019 season) was administered as a single injection per dose, intramuscularly on the right deltoid, in a total volume of 500 µL. Blood for serum and stool were collected at various timepoints and frozen at −80 °C.

2.3. Antibiotic Treatments

Mice were treated with a single antibiotic in drinking water for 1–7 days as indicated in the figure legends. The following antibiotics were used: ampicillin [1 g/L]; clindamycin [1 g/L]; enrofloxacin/Baytril [0.27 g/L]; metronidazole [1 g/L]; neomycin [1 g/L]; nystatin [0.5 g/L]; tobramycin [1 g/L]; vancomycin [0.5 g/L]. Antibiotic-containing drinking water was replaced every 2–3 days. All antibiotics were of pharmaceutical grade and were purchased from United States Pharmacopeia (USP, Rockville, MD, USA) or Sigma-Aldrich (St. Louis, MO, USA). Prior to vaccination, mice were supplied drinking water without antibiotic for 1–21 days, as indicated in the figure legends.

2.4. Immunizations, Vaccine Antigens and Adjuvants

Mice were immunized by IM injection of 50 µL into each hind limb quadriceps (100 µL total dose per animal per immunization). The following antigens and adjuvants were resuspended in endotoxin-free PBS (Life Technologies, Carlsbad, CA, USA) and used for vaccination studies as indicated in figure legends: endotoxin-free ovalbumin (OVA) from Hyglos GmbH [10 µg/100 µL dose]; recombinant hepatitis B surface antigen (HBsAg) manufactured by Merck & Co., Inc. in West Point, PA, USA [0.2 µg/100 µL dose]; Imject© Alum from ThermoFisher according to manufacturer’s specifications; lipid nanoparticles based adjuvant VA-510 (LNPs) prepared at Merck & Co., Inc. in West Point, PA 19486, USA as described [29,30,31] [125 µg/100 µL dose].

2.5. Collection of Stool and Tissues

Whole blood was collected and mixed with sodium citrate. For serum, blood was collected and allowed to clot. For mice, fresh stool pellets were collected as they were excreted and flash frozen without any buffers/stabilizers. For rhesus macaques, stool samples were collected using manual evacuation method and frozen immediately without any buffers/stabilizers. Spleens, mesenteric lymph nodes, and small intestines without contents, were excised and flash-frozen in liquid nitrogen. All tissues were stored at −80 °C until used for analysis.

2.6. Antigen-Specific Antibody Titers

Sera was collected at the timepoints indicated in the figures and text above. Antigen-specific antibody titers were measured in duplicate via endpoint titer enzyme-linked immunosorbent assay (ELISA), following the Mouse Anti-OVA IgG Antibody Assay Kit (Chondrex, Woodinville, WA, USA), or as previously described [29].

2.7. Pharmacokinetics

Tissues were homogenized in 3–9 volumes DPS buffer (w/v) and extracted by protein precipitation using 3 volumes methanol (v/v) containing a generic internal standard mixture. Extracted samples were quantified by LC-MS/MS on a Waters Acquity UPLC system (Milford, MA, USA) with mobile phase A consisting of 0.1% formic acid in water, and mobile phase B consisting of 0.1% formic acid in acetonitrile using a Waters HSS T3 (100 Å, 1.8 µm, 2.1 mm × 50 mm) column maintained at 30 °C with gradient elution. The analyte and internal standards were detected via tandem mass spectrometry employing an AB/Sciex 5000+^TM triple quadrupole system (Concord, ON, Canada) using Multiple Reaction Monitoring (MRM).

2.8. T-cell Activity Quantification

Spleens from vaccinated mice were collected at the indicated timepoint, mechanically disrupted, and treated with ACK (Ammonium-Chloride-Potassium) buffer to lyse erythrocytes. Splenocytes were cultured at 37 °C, 5% CO₂ in RPMI media with 10% FBS and 1.25 µg/mL each α-CD28 (clone 37.51, BD) and α-CD49d antibodies (clone R1–2, BD). MHCI-specific ovalbumin SIINFEKL/OVA_257–264 or MHCII-specific ovalbumin peptide OVA_323–339 were added at 2 µg/mL (Invivogen, San Diego, CA, USA). Phorbol 12-myristate 13-acetate (Promega) and ionomycin (Tocris, Bristol, UK) were added at 1.25 µg/mL and 0.02 µg/mL, respectively. After 24 h, interferon-gamma (IFN-γ) in the supernatant was quantified using the mouse IFN gamma ELISA (AbCam, Cambridge, UK).

2.9. DNA Extraction and 16S rRNA Gene Sequencing

DNA from 1–2 stool pellets was extracted using the QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany) as per the manufacturer’s instructions. The V4 region of the 16S rRNA gene of the stool samples was amplified on the Illumina MiSeq platform using their standard 2 × 250 bp paired-end protocol and the 515F and 806R 16S rRNA gene sequence primers [32]. Resulting DNA sequences were processed as previously described [33]. Briefly, we used Mothur (v1.39.5) to assemble the paired-end reads (requiring complete overlap) and perform quality filtering. Sequences were aligned to the SILVA 16S rRNA gene sequence database (v132) and filtered to remove chimeras and lineages of archaea, chloroplasts, and mitochondria. OTUs were clustered at 97% similarity using the OptiClust algorithm, and reads were aligned to OTUs for quantification. OTU tables and taxonomic assignments were exported for downstream analysis in R.

The rhesus macaque stool sample 16S rRNA gene sequenced were bioinformatically processed using the standard CosmosID protocol. Briefly, OTUs were identified using a closed-reference approach and 97% sequence similarity using Qiime [34].

2.10. Shotgun Metagenomic Sequencing and KEGG Pathway Enrichment Analysis

Reads were randomly amplified for sequencing using the Illumina TruSeq library kit. The paired-end sequencing was performed on the Illumina HiSeq platform using a universal primer in one direction, and a set of standard index primers in the reverse direction. Sequences were processed for quality using bbduk for adapter removal and quality trimming (ktrim = 1, mink = 17, maq = 30, minlen = 75), and fastqc for general quality visualization. Bowtie2 was used to remove reads that aligned with similarity to the PhiX and human genome. Details can be found in the supplemental analysis code.

Metagenomic sequences were aligned to the KEGG database through the public API (accessed on 12 August 2020) and using the diamond alignment algorithm [35,36,37,38]. KEGG orthology (KO) counts were generated by compressing gene counts and equally splitting counts when genes were present in multiple KO groups, or vice versa. KO counts were normalized using the MUSICC algorithm [39]. KEGG pathway enrichment was calculated using the Generally Applicable Gene-set/Pathway (GAGE) analysis package, available through R Bioconductor. The enrichment calculation for each antibiotic treatment group was performed in reference to the water control group, using the “unpaired” comparison and “same direction” parameters [40].

2.11. RNA Extraction, Sequencing, and Differential Gene Expression Analysis

Total RNA was extracted from spleens, mesenteric lymph nodes, and small intestine without contents using the RNeasy Mini Kit (Qiagen) as per manufacturer’s instructions. Whole-blood was collected using RNAprotect Animal Blood Tubes (Qiagen) and processed as per manufacturer’s instructions. RNA quality and concentration was measured using an Agilent 4200 TapeStation (Agilent Technologies, Santa Clara, CA, USA) as per manufacturer’s instructions and an RNA integrity number (RIN) cutoff of 6 was used to progress samples to sequencing. RNA was sequenced by BGI Genomics (Shenzhen, China). FASTQ files were trimmed and mapped with OmicSoft (v10.1) [OmicSoft Corporation, Cary, NC, USA]. Trimming was performed with a 5bp window starting from the 3′ end. Using this sliding window, if 2 or more bases in this 5bp window were below a PHRED quality score of 15, those bases and any base pair to the right were removed.

For analysis of differential gene expression, genes were included if the read count was above 1 read count per million in at least 5 samples of the group. Differential expression was calculated using EdgeR (v3.20.9) which leverages Limma dependencies (v3.34.9) in the R statistical computing environment (R version 3.4.3) [41]. EdgeR moderates the degree of overdispersion using an overdispersed Poisson model of biological and technical variability coupled with an empirical Bayes method. Library size normalization factors were generated using the trimmed mean of M-values (TMM) method (trimmed mean of M-values) from the “calcNormFactors” function in the edgeR library and used as covariates in modeling. Data was then fit to a generalized linear model (GLM) and a GLM likelihood ratio test was performed to determine if the coefficient representing the contrast between the conditions of interest was equal to 0 indicating no differential expression. Significance for differential expression was set at an FDR (Flase Discovery Rate) corrected p-value < 0.05.

2.12. GO Term Analysis and Clustering

Gene Ontology (GO) terms were analyzed for enrichment using the ToppFun program from ToppGene [42]. Differentially expressed genes (FDR corrected p-values < 0.05) were provided to this program to assess enrichment in the following GO categories: Biological Process, Molecular Function, and Cell Component. A term was considered significantly enriched with FDR (Benjamini and Hochberg) corrected p-value < 0.05. The mesenteric lymph node GO term network was constructed using the Cytoscape application EnrichmentMap [43], using FDR (BH) cutoff of 0.035 and overall edge cutoff of 0.8, and clusters were identified using ClusterMaker—MCL Cluster [44] with granularity parameter of 1.5. Clusters were manually annotated. A new network was constructed from GO terms in immune-related clusters using EnrichmentMap with an overall edge cutoff of 0.6, and clusters were manually identified and annotated.

2.13. Statistical Analyses

Statistical analyses were performed in the R statistical computing environment [41]. The statistical tests, multiple hypotheses corrections, and other details are identified above, in association with the results, and the details can be found in the associated analysis code (see link in Data Availability Statement).

3. Results

3.1. Microbiome Perturbation by Different Antibiotics Differentially Modulates Vaccine Immune Response

Recent studies have demonstrated that treatment with a cocktail of antibiotics prior to vaccination alter vaccine response by disrupting the microbiome [2,22,24,26,28,45]. However, the majority of these studies utilized chronic treatment (~4 weeks) with a combination of antibiotics. We therefore hypothesized that short-term (i.e., 7 days) oral treatment of individual antibiotics from different classes may differentially impact vaccine response, due to differences in their impact on the microbiome. To test this hypothesis, we treated BALB/c mice for 7 days with each of seven antibiotics in drinking water with no artificial sweeteners or additives, followed by one day of recovery without antibiotics, prior to the first dose of intramuscular vaccination with alum-adjuvanted ovalbumin (OVA) antigen (Figure 1A). A booster dose with no preceding antibiotic treatment was delivered 30 days after the first dose (Figure 1A). To ensure that the antibiotics did not adversely affect the health of mice during treatment, we confirmed acceptable and consistent water intake, food intake, estimated dosage, and body weight throughout the antibiotic treatment period (Figure S1). Twelve days following the booster dose, we found that each antibiotic had a unique reductive impact on serum OVA-specific IgG titer compared to water-treated control mice (Figure 1B).

We validated the pharmacodynamic impact of oral antibiotic treatment by characterizing the gut microbiome of each mouse at time of vaccination using 16S rRNA gene sequencing of fecal pellets. As expected, each antibiotic treatment uniquely altered the microbiome, which resulted in distinct microbial communities at the time of vaccination (Figure 1C). Microbial communities from mice treated with vancomycin and clindamycin, the two antibiotics with the strongest impact on vaccine response, visibly clustered together because of their shared community composition (Figure 1D), suggesting the presence of some common community elements in these most disruptive antibiotic treatments (e.g., Salmonella and Enterobactericeae).

Because we observed some community structure commonalities between our treatments, we performed a correlation analysis across all mice to determine whether specific community signatures were associated with resulting titer. We observed no strong correlations between the relative abundance of individual bacterial taxa and antibody titers (Figure S2A). We also investigated whether reduced antibody titer was associated with total community disruption (beta diversity compared to water) and observed an overall correlation between antibody titers and the of antibiotic-mediated microbiome disruption (Figure 1E). Together, these results suggest that the extent of microbiome community disruption at the time of vaccination correlated with the extent of vaccine hyporesponse and that this phenomenon may have been driven by changes in complex community interactions rather than by disruption of specific individual bacterial taxa.

3.2. Antibiotic-Mediated Vaccine Hyporesponse Is Associated with Altered Microbiome Functionality

The correlation between antibody titers and bacteria at the community scale, rather than the individual genus level, suggested this complex dynamic could be informed by a function-based analysis. Microbes from different taxonomic groups can execute redundant functions and may sometimes be interchangeable [46], which could mask common functional alterations across antibiotic treatments in our data. We therefore supplemented our taxonomic analyses with shotgun metagenomic sequencing and identified correlations between the enrichment of microbiome functional potential and antibody titers (Figure S2B–E). The four pathways that significantly correlated with antigen specific titers were metabolic pathways such as terpenoid-quinone biosynthesis, sulfur metabolism, tryptophan metabolism, and selenocompound metabolism. All of the observed significant correlations were negative, meaning that enrichment of the metabolic pathways was associated with reduced antigen-specific IgG titers. The role of these microbial metabolic pathways in immune function and vaccine response is unclear, though they have previously been associated with microbial dysbiosis [47,48,49,50] and may result from expansion of normally rare microbes in antibiotic-treated mice [51].

In addition to correlating sample pathway enrichment with resulting vaccine titer, we also identified the metagenomic pathways depleted and enriched in each antibiotic treatment group relative to water treated controls (Tables S1 and S2, FDR corrected p-values < 0.01). Four pathways were depleted in the antibiotic treatment groups, and while three of the pathways were only depleted in a single antibiotic group, the aminoacyl-tRNA biosynthesis pathway was depleted in five antibiotic treatment groups including clindamycin, metronidazole, neomycin, tobramycin, and vancomycin. This signal could reflect depletion of Ruminococceae and Lactobacillales across those antibiotics [52,53]. Conversely, there was no overlap of enriched pathways across antibiotic treatments, though many were associated with core metabolism and growth. The majority of enriched pathways (12 of 15) were in the vancomycin group, including lipopolysaccharide biosynthesis, which mirrors the enrichment of Gram-negative genera after vancomycin treatment (Figure 1B). Together, these data likely reflect alterations in representation of specific pathways due to reduced bacterial diversity after antibiotic treatment, especially in the vancomycin group, but future investigation may reveal functional links between altered microbial biosynthetic activity and vaccine response.

3.3. Vancomycin-Induced Vaccine Hyporesponse Is Minimized by Allowing Sufficient Time for the Microbiome to Recover

Our data above suggest that microbiome disruption at the time of immunization was associated with suboptimal antigen-specific IgG responses. To further test association of the microbiome with vaccine specific IgG titers, we asked whether the converse relationship was true: Can vaccine response be improved by allowing the disrupted microbiome to recover? We hypothesized that extending the time after antibiotic treatment and before immunization would allow for microbiome recovery and prevent vaccine hyporesponse in antibiotic-treated animals.

We tested our hypothesis using vancomycin as a model antibiotic due to its robust effect on humoral vaccine responses observed above (Figure 1). Vancomycin also demonstrated consistent vaccine-modulatory effects of subsequent studies wherein: (i) vancomycin, but not ampicillin, robustly impacted alum-adjuvant OVA vaccine-specific titers in a different mouse strain (C57Bl/6; Figure S3A); (ii) vancomycin decreased antigen-specific IgG titers and T cell responses when mice were vaccinated with OVA adjuvanted with a novel lipid nanoparticle (LNP)-based adjuvant (Figure S4A–D); and (iii) vancomycin reduced antigen-specific IgG titers against a clinically relevant hepatitis B vaccine conducted in two geographically distinct vivaria (Figure S4E–F). Vancomycin treatment has previously been shown to disrupt response to nonadjuvanted inactivated influenza vaccine [22] and rabies vaccine [54], and is included in many antibiotic cocktails used to perturb the microbiome for vaccination studies with variable effects on individual vaccine antigens [22,26,28,54]. Vancomycin treatment has also been shown to modulate several immune-related pathways [55,56,57,58], and reduces the abundance of bacteria belonging to Clostridiales and Bacteroidales, while allowing increases in Proteobacteria [55,56,57,58,59], in agreement with our data (Figure 1C).

To test our hypothesis using vancomycin, we treated mice with vancomycin and extended the number of days between antibiotic cessation and immunization with LNP adjuvanted-OVA vaccine from 1 to 7, 14, or 21 days (Figure 2A). We observed a visible trend between the length of recovery time after vancomycin treatment and antigen-specific antibody titers, in which animals with shorter recovery times had statistically significantly decreased antibody titers, whereas mice given 21 days of recovery before immunization had similar antibody titers to water-treated animals (Figure 2B).

Microbiome profiling by 16S rRNA gene sequencing revealed that microbial community recovery after vancomycin treatment followed defined dynamics, as measured by community dissimilarity (beta-diversity) from water (Figure 2A) and OTU richness (Figure S5A). Prior to vaccination, each treatment group had recovered to a different extent, and the extent of microbiome disruption was significantly negatively correlated with resulting antigen-specific antibody titers (Figure 2C). Of note, this result mirrors the correlation observed between antibody titers and microbiome disruption mediated by different classes of individual antibiotics (Figure 1E), demonstrating that microbiome disruption is associated with vaccine hyporesponse in both disruption and recovery models.

To confirm that these results are not due to differences in vancomycin treatment conditions between groups or direct effects of vancomycin on the systemic immune system, we validated that vancomycin exposure was comparable across mice, that >98% of vancomycin was excreted in the feces by the time of immunization relative to peak levels, and that vancomycin remained undetectable systemically in the blood at any timepoint (Figure S5E,F).

Because we found that greater disruptions of the microbiome were associated with greater reductions in the antigen-specific IgG response (Figure 1E), we hypothesized that reducing microbiome disruption by decreasing vancomycin treatment time to one or three days would also result in a reduced impact on vaccine hyporesponse. Indeed, we found that three days of vancomycin treatment moderately impacted the antigen-specific IgG titers following vaccination, while one day of treatment did not alter response (Figure S5C). We also confirmed that a single day of vancomycin treatment only moderately reduced bacterial richness compared to three or seven days of treatment (Figure S5B). The findings support our previously observed trends toward reduced vancomycin exposure in the gut (Figure S5E,F) and that the extent of microbiome disruption correlated with the effect on antigen-specific antibody response (Figure 1). Together, these findings suggest that mitigating the effects of antibiotic treatment on the gut microbiome by delaying immunization or decreasing treatment duration could prevent antibiotic-induced vaccine hyporesponse.

We further explored whether the reduction in antigen-specific IgG titers were correlated with individual bacterial taxa prior to immunization. We found no taxa strongly associated with antibody responses (Figure S5D). Altogether, our data points to a fine-tuned system by which the diversity of the microbiome composition differentially influences the outcome of vaccine-induced antigen-specific IgG titers: e.g., reduced antigen-specific IgG titers when the microbiome is disrupted by antibiotic treatment, but restoration of optimal responses when the vaccine is administered after the microbiome is allowed sufficient time to recover.

3.4. Vancomycin Disrupts Vaccine Outcome in Rhesus Macaques

In addition to studying mouse models, we further evaluated whether our observations of vancomycin-induced microbiome disruption and subsequent vaccine hyporesponse were replicated in an NHP model (i.e., rhesus macaque). Rhesus macaques are an important translational animal model with a human-like immune system and microbiome composition. We orally treated groups of 4 vaccine-naïve animals with vancomycin or a vehicle control for 14 days, followed by 1 day of no treatment, prior to each of three doses of intramuscular immunization with the clinically relevant quadrivalent influenza vaccine (Fluzone 2018–2019 season; chosen for its clinical relevance over ovalbumin), given on Days 0, 28, and 84 (Figure 3A). While we observed a robust increase in influenza-specific IgG titers in vehicle-treated control animals (peaking at 7 days after the second and third doses), the vancomycin-treated animals had a reduced antibody response (Figure 3A). To validate the effect of vancomycin treatment on the microbiome, we longitudinally evaluated the gut microbiome composition by 16 s rRNA gene sequencing of stool samples from all animals. As expected, after each 14-day vancomycin treatment regimen (Days 0, 28, 84), the microbial richness was reduced approximately 2-fold compared to vehicle-treated animals or other timepoints (Figure 3B) and the microbial communities clustered separately from all other samples (Figure 3C).

Because in mice we found that microbiome recovery before vaccination prevents hyporesponse, we administered a fourth booster dose without vancomycin pre-treatment on Day 173 to investigate whether microbiome recovery in macaques previously treated with vancomycin was sufficient to restore vaccine responsiveness. In contrast to the previous vaccination timepoints, the microbiome was similar in both cohorts (Figure 3). Strikingly, the vancomycin cohort produced influenza vaccine-specific IgG titers approaching the levels of the NHPs in the vehicle cohort. This crucial observation suggests that microbial dysbiosis at the time of vaccination disrupts vaccine response, but that recovery of the microbiome after antibiotic treatment prior to immunization may lead to successful vaccination responses.

3.5. Vancomycin Impacts Key Innate Immune Pathways in the Mesenteric Lymph Nodes

Together our data suggested that vaccine response is associated with the state of the microbiome at the time of immunization. While measurement of antigen-specific IgG titers served as a surrogate for immune tone in our studies, the specific immune pathways altered upon antibiotic treatment in systemic immune cells or the gastrointestinal mucosal compartment remained unclear. To this end, we treated mice with either vancomycin or a water control for seven days, followed by one day of antibiotic recovery before harvesting the small intestines, mesenteric lymph nodes, spleens, and whole blood. RNA sequencing analysis revealed changes in host gene expression in the small intestine (8 genes), spleen (11 genes), blood (8 genes), and mesenteric lymph nodes (367 genes) due to the perturbation of the gut microbiome. It should be noted that the cell-types that make up these tissues, particularly the spleen, blood, and lymph nodes, are highly dynamic during an immune response. Due to this, the gene expression changes that are observed may be due in part to the cell type frequencies in the tissues changing, and this should be considered in interpreting the results.

Within the mouse mesenteric lymph node dataset, we identified 753 Gene Ontology (GO) biological processes that were enriched using gene sent enrichment analysis implemented in ToppGene [42]. Clustering of GO terms revealed that vancomycin mediated perturbation of the gut microbiome led to changes in modulated metabolism, signaling, trafficking, differentiation, and cell death pathways, as well as several immune-related pathways (Figure S6). Sub-clustering of the immune-related GO terms highlighted the following immune functions: cytokine response and production, response to bacteria, T-cell adhesion and activation, reactive oxygen and nitrogen species, and inflammatory signaling (Figure 4A).

Although our previous analyses did not identify individual bacterial taxa correlated with antibody titer after vaccination, we hypothesized that some taxa could be associated with individual immune pathways. We therefore examined whether vancomycin-induced changes in immune-related gene expression were correlated with the abundance of particular bacterial taxa. We found that specific OTUs within Lachnospiraceae, Ruminococcus, Clostridia, and Enterobacteriaceae were associated with genes involved in innate and adaptive cell polarization and cytokine/chemokine production. More specifically, we found that several genes associated with productive activation of the immune system were decreased [Asb2 [60], Cxcl5 [61], Tnfrsf9 [62], Bcl2a1b [63], Ccdc88b [64], Flt3 [65]], while genes associated with dampening the immune response were upregulated [Mmp28 [66,67], Crispld2 [68]] (Figure 4B). This important finding implicates particular bacterial taxa as key immunomodulatory agents and suggests these taxa may work in concert to set the immune “tone” for subsequent immune responses including vaccination. The correlation of multiple bacteria with multiple diverse immune functions also highlights the complexity of microbially modulated immunity, and the need to continue exploring these underlying dynamics at the systems biology level.

4. Discussion

Interactions between the gut microbiota and host play a fundamental role in local and systemic immunity and homeostasis [69,70,71]. Recent reports have employed antibiotics to link microbiome perturbation and the host response to viral infections [72,73], immune checkpoint inhibitor therapy [74,75,76], and autoimmune/allergic diseases [55,58]. Several groups have already demonstrated an effect of antibiotic-mediated microbiome perturbation on response to vaccination [22,24,26,28,45,77,78], although specific mechanisms and dynamics underlying the vaccine-modulatory effects of antibiotics remain an area of active investigation.

Here, we report that differentially perturbing the microbiome, either by using increasingly disruptive antibiotics, decreasing the interval between vancomycin treatment and immunization, or by increasing the duration of vancomycin treatment, proportionally decreased the response to vaccination. Conversely, recovery of the microbiome allowed for recovery of vaccine response. We did not observe relevant associations between vaccine response and specific taxa or microbial metagenomic pathways, which supports the idea that multiple, possibly redundant, microbial factors may modulate vaccine response, rather than specific microbes, in agreement with other preclinical studies of vaccine response [22], tumorigenesis [79] and pathogenesis [80]. Indeed, clinical studies linking the microbiome and vaccine response do not reveal a consistent link to a specific microbial signature supporting the existence of redundant overlapping species and pathways.

While previous publications in the microbiome-vaccine research field have employed chronic treatment of combination antibiotics [22,23,24,26,28,54,77,78], we took a broader approach in which we demonstrated individual antibiotics have differential levels of impact on two different strains of mice (BALB/c and C57BL/6). We also note that while the majority of research in this space has focused on aluminum-based adjuvants and inactivated/attenuated viral vaccines [81], we included subunit vaccines adjuvanted with aluminum or novel adjuvants such as lipid nanoparticles, as well as inactivated influenza vaccine. Additionally, some antibiotics such as Vancomycin are often administered intravenously (IV) in the clinic, and because we use oral administration here in our models, further studies warrant investigation into IV administration. However, it is important to point out that our pre-clinical studies pertaining to a single antibiotic use was largely used as an exploratory tool to uncover novel commensal microbial communities that are critical for optimal vaccine responses; not necessarily to mirror clinical prescriptions. Furthermore, we demonstrate vancomycin’s effect on both the model subunit antigen ovalbumin and the clinically relevant hepatitis B surface antigen. Additionally, although the majority of our studies measured IgG titers as a proxy for vaccine response, we also evaluated antigen-specific T cell responses and found appreciable decreases in vancomycin-treated mice. Thus, our data robustly support the conclusion that antibiotic treatment disrupts vaccine response across a variety of formulation measurements and animal models.

Our data support the hypothesis that microbial communities affect immune outcomes by modulating the underlying immune tone, or state of responsiveness Vancomycin depletion of the gut microbiome diversity modulated expression of key host genes which likely contributed to the immune tone and inflammatory response, which could lead to subsequent vaccine hyporesponsiveness. We also found correlations between altered immune gene expression signatures and certain bacterial taxa. This observation suggests diverse bacteria may work in concert to affect immune tone, which could explain the lack of consistent microbial signatures associated with vaccine response [82,83]. Although most studies in the microbiome-vaccine field rely on 16S rRNA sequencing to identify perturbations in bacterial communities, we additionally performed shot-gun metagenomic sequencing to reveal potential pathways in the microbiome that correlate with antigen-specific IgG titers. This method provides a unique advantage to identify microbial pathways that are potentially indicative of optimal health across broad measures of immune tone. To further dissect the role(s) of the heterogeneous microbiota during response to vaccines, it will be critical to embrace functional and systems biology approaches.

Emerging technologies will play a key role in future investigations of microbiome dynamics and vaccine response. While our approach here using antibiotic treatment facilitated generation of distinct and diverse microbial communities and revealed a spectrum of effects on vaccine response, our current data cannot completely rule out direct effects of antibiotic treatment on the immune system [75,84,85]. An antibiotic-free model, such as fecal transplant of microbiota into germ-free mice, has been used to demonstrate microbiome-dependent changes in vaccine response and will also be an important next step in our current research program. In addition, observations derived from mouse models must be more extended to human clinical studies [2] to verify that conclusions are relevant for human health.

A unique component of this work is the use of rhesus macaques, an important preclinical model for human vaccines. We found that vancomycin treatment negatively impacted response to inactivated influenza vaccine in rhesus macaques, which mirrors a recent publication that reported a cocktail of antibiotics including vancomycin impacted influenza vaccine outcome in healthy adults with low pre-existing immunity [2]. In our study, a booster immunization without prior antibiotic treatment appeared to improve antibody titers. Together, these data highlight the importance of understanding the temporal dynamics of antibiotic treatment and the microbial community when administering vaccines.

5. Conclusions

Altogether we report evidence that antibiotic-mediated microbiome perturbation can greatly impact vaccine outcome, but that recovery of the microbiome after cessation of treatment may prevent antibiotic-induced vaccine hyporesponse in adult animals; it remains unclear whether or not this is the case in neonates [23]. In addition, our data uncovers important host pathways and immune mediators that may be critical for optimal vaccine outcome. Our results provide support for the importance of understanding antibiotic usage and the microbiome in the context of vaccine discovery. Our results suggest that further investigations may identify novel host and microbial factors that are critical for host immunity, and result in the development of more effective vaccines or vaccine co-therapies to address global unmet needs.