Bifidobacteria support optimal infant vaccine responses

Abstract

Accumulating evidence indicates that antibiotic exposure may lead to impaired vaccine responses^1,2,3,4; however, the mechanisms underlying this association remain poorly understood. Here we prospectively followed 191 healthy, vaginally born, term infants from birth to 15 months, using a systems vaccinology approach to assess the effects of antibiotic exposure on immune responses to vaccination. Exposure to direct neonatal but not intrapartum antibiotics was associated with significantly lower antibody titres against various polysaccharides in the 13-valent pneumococcal conjugate vaccine and the Haemophilus influenzae type b polyribosylribitol phosphate and diphtheria toxoid antigens in the combined 6-in-1 Infanrix Hexa vaccine at 7 months of age. Blood from infants exposed to neonatal antibiotics had an inflammatory transcriptional profile before vaccination; in addition, faecal metagenomics showed reduced abundance of Bifidobacterium species in these infants at the time of vaccination, which was correlated with reduced vaccine antibody titres 6 months later. In preclinical models, responses to the 13-valent pneumococcal conjugate vaccine were strongly dependent on an intact microbiota but could be restored in germ-free mice by administering a consortium of Bifidobacterium species or a probiotic already widely used in neonatal units. Our data suggest that microbiota-targeted interventions could mitigate the detrimental effects of early-life antibiotics on vaccine immunogenicity.

Main

Global immunization programmes prevent millions of deaths per year^5,6; however, for reasons that are unclear, immune responses to vaccination exhibit substantial interindividual variation and are frequently suboptimal in the populations at most risk from infectious disease⁷. Evidence increasingly indicates that the gut microbiota is an important and targetable factor that may, at least in part, explain these interindividual and interpopulation differences in immune responses to vaccination^7,8. In mice, antibiotic exposure in early life has been shown to lead to impaired antibody responses to five different adjuvanted and live-attenuated parenteral vaccines¹. Antibiotic-treated and germ-free (GF) adult mice have also been shown to have impaired antibody responses to an unadjuvanted influenza vaccine but not to the live yellow fever vaccine⁹ or a COVID-19 messenger RNA (mRNA) vaccine¹⁰. Furthermore, adult participants randomized to antibiotics before immunization with unadjuvanted trivalent influenza vaccine had significantly impaired neutralizing and binding antibody titres; however, this was not observed in adults with pre-existing immunity to influenza¹¹. On the other hand, oral rotavirus vaccine immunogenicity was not found to be reduced in adults randomized to antibiotics¹². Consistent with our previous preclinical data suggesting that antibiotics may have a more profound effect on antibody responses to adjuvanted vaccines in early life¹, a recent retrospective study found that vaccine antibody titres were lower in antibiotic-exposed infants². Cumulative antibiotic exposure during the first year of life has also been associated with reduced vaccine response at 12 months^3,4.

To prospectively assess whether antibiotic exposure in early life affects antibody responses to vaccination and to better understand the mechanisms involved, we established a prospective observational study that recruited healthy, full-term, vaginally delivered infants. Infants were defined as not exposed to antibiotics (no-ABX, n = 80) or exposed to antibiotics (ABX, n = 111) in the peripartum period (Fig. 1a and Extended Data Fig. 1). ABX infants were further subdivided (Fig. 1b) into three antibiotic exposure groups defined a priori (Methods). Thirty of 32 infants (94%) in the direct neonatal antibiotic exposure group (neo-ABX) started antibiotics in the 48 h after birth, and 31 (97%) of these infants received intravenous (i.v.) benzylpenicillin and gentamicin for at least 48 h. Of these infants, 19 were also exposed or potentially exposed to maternal antibiotics intrapartum or postnatally (Supplementary Table 1). In the intrapartum antibiotic exposure (IP-ABX) group, 36 of 49 (73%) of mothers received i.v. benzylpenicillin within 48 h before delivery, seven (14%) received i.v. benzylpenicillin in combination with at least one other oral or i.v. antibiotic, four (8%) received i.v. clindamycin, one received i.v. clindamycin plus oral cephalexin and—consistent with the a priori definition of this group, which included mothers treated with antibiotics in the last 28 days of pregnancy—one received oral cephalexin from day 12 to day 10 before birth. Mothers of infants in the postnatal antibiotic exposure group (PN-ABX, n = 30) received a variety of different antibiotics orally in the postnatal period from birth to day 45, and seven also received i.v. antibiotics (Supplementary Table 1). Maternal and infant demographics, as well as breastfeeding rates, were comparable across exposure groups, except for probiotic usage, which was higher in the ABX group (Extended Data Table 1).

Fig. 1: Vaccine-specific antibody responses are frequently impaired in infants directly exposed to antibiotics in the neonatal period.

a, Overview of the AIR study design. Responses to the polio, measles, mumps, rubella and meningococcal ACWY vaccines were not measured in this study. b, Antibiotics administered to infants and their mothers in the direct neonatal antibiotic exposure group (Neo-ABX, n = 32 infants), intrapartum antibiotic exposure group (IP-ABX, n = 49 infants) and postnatal antibiotic exposure group (PN-ABX, n = 30 infants). c, Heatmap showing fold change in the geometric mean concentration (GMC) of vaccine antigen-specific antibodies in serum collected from Neo-ABX (n = 27 at 7 months, n = 26 at 15 months), IP-ABX (n = 43 at 7 months, n = 42 at 15 months) and PN-ABX (n = 26 at 7 months, n = 27 at 15 months) infants, relative to No-ABX infants (n = 64 at 7 months, n = 65 at 15 months). The blue–red gradient indicates decreased and increased fold change, respectively. Statistical significance in c was assessed using a generalized linear model adjusting for sex and baseline antibody titres. *P < 0.05 (unadjusted). All statistical tests were two-sided. Exact P values are provided in Supplementary Table 3a. Icons in a were created using BioRender (https://biorender.com). DTP, diphtheria, tetanus and pertussis; Hep., hepatitis.

Source data

Eighty-six per cent of infants in this study received hepatitis B vaccine at birth. At approximately week 6 of life, the first doses of the other primary childhood vaccines were administered as recommended by the Australian National Immunisation Program (NIP) schedule as of 2017. Further primary doses were administered at approximately 4 and 6 months of age (Fig. 1a). Serum IgG titres against 20 different vaccine antigens and anti-rotavirus vaccine IgA titres were measured at baseline and at 7 and 15 months (Extended Data Fig. 2a–u), using previously validated immunoassays^{13,14,15,16,17,18}. To assess the relationship between ABX group and antibody responses to vaccination, we used multivariate regression adjusting for sex and baseline antibody titres; this was because baseline titres, as expected¹⁹, were frequently negatively correlated with subsequent antibody titres (Extended Data Fig. 2v). Compared with No-ABX infants, infants exposed to any antibiotics before the first dose of the 13-valent pneumococcal conjugate vaccine (PCV13), 6-in-1 Infanrix Hexa and rotavirus vaccines had significantly (P < 0.05) decreased IgG titres against six of the capsular pneumococcal polysaccharides (PPS1, 6B, 9V, 18C, 19F and 23F) in the PCV13 vaccine and significantly lower anti-tetanus toxoid IgG titres at age 7 months (Fig. 1c and Supplementary Table 2a). After subdivision by exposure group, this decrease in anti-PPS IgG titres was evident in Neo-ABX infants (PPS1, 6B, 9V, 18C, 19F and 23F) but not IP-ABX infants (Fig. 1c, Extended Data Fig. 3a–f and Supplementary Table 3a). A similar trend towards reduced anti-PPS4 and PPS5 titres (but not those for other polysaccharides) was observed (Extended Data Fig. 3g–m). IgG titres against diphtheria toxoid and the Haemophilus influenzae type b (Hib) polyribosylribitol phosphate (PRP) antigen (antigens in the 6-in-1 Infanrix Hexa vaccine) were also significantly lower in Neo-ABX infants at 7 months (Fig. 1c and Extended Data Fig. 3n,o). A similar trend was observed for anti-tetanus toxoid IgG titres (P = 0.09) (Extended Data Fig. 3p). There was no detectable association of ABX group with IgG titres against pertussis toxoid (PT), hepatitis B surface antigen (HB-sAg), pertactin (PRN) or filamentous haemagglutinin (FHA), or with serum oral rotavirus vaccine IgA titres (Extended Data Fig. 3q–u). PN-ABX infants had significantly reduced anti-PPS19A and 23F IgG titres at 7 months (Fig. 1c), indicating that maternal postnatal antibiotic exposure may have an impact on subsequent infant responses to vaccination but to a lesser extent than direct administration of antibiotics to neonates.

Antibody titres were significantly lower at 15 months compared with 7 months for all of the antigens assessed, except anti-tetanus toxoid IgG titres, consistent with waning of immunity over time²⁰ (Extended Data Fig. 2a–u). Any antibiotics before the first dose of the PCV13, 6-in-1 and rotavirus vaccines was associated with significantly lower IgG titres at 15 months against six different serotypes in the PCV13 vaccine (PPS4, 6A, 9V, 19A, 19F and 23F) and Hib-PRP (Fig. 1c and Supplementary Table 2a). After subdivision by exposure group, significantly reduced IgG titres were observed only in Neo-ABX (PPS14, PPS19A, Hib-PRP) and PN-ABX (PPS6A, 9V, 19A, 19F, 23F) infants at this time point (Extended Data Fig. 4a–u and Supplementary Table 3a). There were no significant differences in IP-ABX infants. We also analysed the antibody data longitudinally (rather than separately at 7 and 15 months) and found that IgG titres against nine of 13 serotypes were significantly lower in Neo-ABX infants (Supplementary Table 3b). Taken together, our data indicate that antibiotic-exposed infants, particularly those exposed directly to antibiotics in the neonatal period, have impaired immunogenicity against several different vaccine antigens, particularly those in conjugate polysaccharide vaccines.

We next used opsonophagocytic assays to assess the functionality of antibodies against three PCV13 serotypes (6B, 9V and 18C) that had significantly reduced binding antibody titres in Neo-ABX infants. Compared with No-ABX infants, the opsonophagocytic activity of serum from Neo-ABX infants at 7 months was significantly lower (P < 0.05) against serotype 6B but not 9V or 18C (Fig. 2a–c). These data indicate that neonatal antibiotic exposure can not only impair vaccine-induced binding antibody titres months later but also, in some cases, reduce the functionality of those antibodies.

Fig. 2: Antibiotic exposure in early life is associated with impaired antibody functionality and lower seroprotective responses against some vaccine antigens.

a–c, Opsonophagocytic activity of serum collected from Neo-ABX (n = 26) and No-ABX (n = 59) infants at 7 months against PPS6B (a), PPS9V (b) and PPS18C (c). d,e, Proportions of No-ABX (n = 55) and Any-ABX (n = 83) infants achieving a seroprotective response against each vaccine antigen at 7 (d) and 15 (e) months. f, Proportions of No-ABX (n = 55), Neo-ABX (n = 22), IP-ABX (n = 38) and PN-ABX (n= 23) infants achieving a seroprotective response against each vaccine antigen at 15 months. g, Serum bactericidal activity against N. meningitidis NZ98/254 at week 7 (prevaccination, n= 140) and at 7 (n = 139) and 15 (n= 138) months. h, Serum bactericidal activity geometric mean titres (GMTs) subdivided by no or any antibiotic exposure at 7 months (No-ABX, n= 57; Any-ABX, n = 82). i, Serum bactericidal activity GMTs at 7 months subdivided by the indicated antibiotic exposure groups (No-ABX, n = 57; Neo-ABX, n = 22; IP-ABX, n = 37; PN-ABX, n = 23). j, Serum bactericidal activity GMTs subdivided by no or any antibiotic exposure at 15 months (No-ABX, n = 56; Any-ABX, n = 82). n represents the sample size of infants in each case. For the box plots in a–c and g–j, the box denotes the 25th and 75th percentiles, the whiskers the 5th and 95th percentiles, and the middle bar the median. Bar charts in d–f represent data as percentages. Statistical significance was assessed in a–c and g–j using a generalized linear model (Gaussian distribution) and in d–f using logistic regression, adjusting for sex. *P < 0.05, ***P < 0.001 (unadjusted). All statistical tests were two-sided. Exact P values are provided in the source data.

Source data

Reduced PCV13 seroprotective responses

Next, we assessed the association between ABX group and the proportion of infants with antibody responses above the seroprotective threshold defined for each vaccine antigen (Supplementary Table 2b). At 7 months, 1 month after the three primary doses of PCV13 and 6-in-1 vaccines had been administered, most of the infants (69–100%, dependent on the vaccine antigen) achieved a seroprotective response (Fig. 2d and Supplementary Table 2b), although there was a significantly reduced proportion of Neo-ABX infants above the seroprotective threshold for one PCV13 polysaccharide, PPS6B (Supplementary Table 3c). By 15 months, the percentage of infants in any of the exposure groups maintaining antibody titres above the seroprotective thresholds had dropped considerably, below 50% for several antigens (Fig. 2e). Despite this waning, the proportion above the seroprotective threshold for infants exposed to any antibiotics was lower for all 13 PCV13 serotypes at 15 months compared with No-ABX infants. These reductions were statistically significant for four serotypes (PPS6B, 9V, 14 and 19A), and there was a trend for anti-diphtheria toxoid titres (P = 0.06). The proportions of Neo-ABX and PN-ABX infants above the seroprotective threshold were significantly lower (P < 0.05) for PPS14 and 19A, and PPS9V and 19A, respectively (Fig. 2f and Supplementary Table 3c). No significantly impaired seroprotective responses were observed in IP-ABX infants. These analyses used the seroprotective threshold accepted by the World Health Organization for PCV13 (0.35 µg ml⁻¹). Reanalysis of the data using serotype-specific seroprotective thresholds²¹ showed the same trend (Supplementary Table 3d). These results indicate that there may be more rapid waning of vaccine-induced protection against certain pneumococcal serotypes in antibiotic-exposed infants, particularly those directly treated with antibiotics in the neonatal period.

Responses to meningococcal B vaccination

Infants in our study also received four doses of the four-component meningococcal B vaccine (4CMenB) (Fig. 1a). Sera from infants with matched samples at 7 weeks (pre-4CMenB vaccination), 7 months and 15 months were tested for bactericidal activity against Neisseria meningitidis NZ98/254, an internationally accepted surrogate of protection for meningococcal serogroup B vaccines, for which a titre greater than or equal to 1:4 is considered to be protective²². There was a significant increase in serum bactericidal activity titres postvaccination in the 7- and 15-month samples compared with baseline (Fig. 2g). Despite these very high seroprotective responses, we observed a trend (P = 0.06) towards reduced serum bactericidal activity titres at 7 months in infants exposed to any antibiotics (Fig. 2h). This was not observed when infants were divided further by antibiotic exposure group (Fig. 2i), possibly owing to reduced statistical power. No significant differences were observed at 15 months (Fig. 2j), around 3 months after the fourth dose of 4CMenB had been administered.

Antibiotics alter the infant transcriptome

In adults, transcriptional responses in blood at baseline or induced postvaccination have been shown to be correlated with subsequent vaccine-specific antibody responses²³. Here we profiled the peripheral blood transcriptome in 329 infant blood samples collected at baseline and approximately 1 week after first-dose vaccination (Supplementary Table 4). Gene set enrichment analysis, a powerful approach that can be used to detect subtle but coordinated changes in gene expression^23,24,25, showed upregulation of several inflammation-related blood transcriptional modules (BTMs) in Neo-ABX infants at baseline (Fig. 3a and Supplementary Table 5). Few of these BTMs were found to be enriched among genes upregulated in IP-ABX infants, which may be functionally relevant, given that impaired vaccine antibody responses were observed in Neo-ABX infants and not in IP-ABX infants. Upregulation of inflammatory modules has also been previously reported in adults treated with a cocktail of antibiotics¹¹. By contrast, B cell-related and plasma-cell-related modules were frequently downregulated in Neo-, IP- and PN-ABX infants. The activity of upregulated monocyte-related BTMs was negatively correlated with antibody responses at 7 months, whereas the activity of B cell-related BTMs was positively correlated (Extended Data Fig. 5a). Taken together, these data indicate that BTMs associated with antibody responses to vaccination are altered at baseline, particularly in Neo-ABX infants. Notably, several BTMs related to natural killer cells were enriched among genes downregulated in IP-ABX infants, indicating that although IP-ABX may not be associated with impaired humoral responses to vaccination, it may alter the immune system in early life in other ways.

Fig. 3: Antibiotic-exposed infants have altered blood transcriptional profiles pre- and postvaccination.

a, Heatmap showing normalized enrichment scores (NES) for selected BTMs before vaccination in Neo-ABX (n = 24), IP-ABX (n= 44) and PN-ABX (n = 25) infants compared with No-ABX infants (n= 66). n represents the number of infants in each group prevaccination when RNA sequencing was performed. The blue–red gradient indicates decreased and increased BTM activity, respectively, relative to that of No-ABX infants. b, UMAP projection of whole-blood gene expression data pre- and postvaccination (n = 329 infant blood samples), adjusted for sex and batch using SVAseq³⁸. c, Volcano plot of differentially expressed genes postvaccination. d, Selected BTMs that were statistically enriched among differentially expressed genes. e, Volcano plot showing fold change in circulating immune cell populations in Neo-ABX (n = 27), IP-ABX (n = 40) and PN-ABX (n = 26) infants relative to No-ABX (n = 63) infants. Statistical significance was assessed using the fgsea³⁹ R package in a and d and using the edgeR⁴⁰ package in c. The Benjamini–Hochberg method was used to adjust for multiple comparisons in a, c and d, with an FDR < 0.05 considered to indicate statistical significance. Statistical significance in e was assessed using two-tailed Wilcoxon signed-rank tests without adjustment for multiple comparisons. Asterisk indicates FDR < 0.05. Exact P values can be found in the supplementary tables or source data. NK, natural killer; T_CM, central memory T cell; T_reg, regulatory T cell.

Source data

Compared with the prevaccination baseline, there was a clear difference in peripheral blood transcriptomes approximately 1 week after the first dose of vaccines had been administered (Fig. 3b), with 3,089 differentially expressed genes (false discovery rate (FDR) < 0.05) identified postvaccination (Fig. 3c and Supplementary Table 6). These differences were driven by vaccination and not the 1-week difference in the age of the infants between the two time points (Extended Data Fig. 5b,c). BTMs upregulated postvaccination (Fig. 3d) included several that have been associated with response to vaccination in adults²³ (Extended Data Fig. 5d and Supplementary Table 7). BTMs related to B cells, monocytes and neutrophils were enriched among genes downregulated postvaccination. Broadly, the magnitude of the transcriptional response to vaccination was similar across the different ABX groups (Extended Data Fig. 5e,f). However, because Neo-ABX infants had altered transcriptional activity at baseline, these infants had notably different BTM activity postvaccination compared with that of No-ABX infants (Extended Data Fig. 5g and Supplementary Table 5).

Circulating immune cell populations

We performed flow cytometry analysis of more than 40 different immune cell populations in fresh peripheral blood samples collected concurrently with the week-7 blood samples used for RNA sequencing (Supplementary Table 8 and Supplementary Figs. 1–3). Although there was a modest reduction in the number of CD45RA⁻ regulatory T cells in IP-ABX infants (Fig. 3e and Extended Data Fig. 5h,i), the numbers and frequencies of all other immune cell populations were not significantly different in any of the ABX groups compared with unexposed infants (Extended Data Fig. 5j–o and Supplementary Table 9). In addition, we explored whether any circulating immune cell populations differed between high and low vaccine responders, but we did not detect any significant differences (data not shown). These results indicate that the transcriptional differences observed in ABX infants are not primarily driven by differences in the frequency of major circulating immune cell populations.

Impact of antibiotics on microbiota

We collected stool specimens (n = 409) at weeks 1 and 6, with 89% of the week-6 samples being collected within 1 day of the first dose of the PCV13, 6-in-1 and oral rotavirus vaccines being administered. Shotgun metagenomic sequencing of these samples therefore enabled unique insight into the composition of the faecal microbiota in these infants at the time of vaccination (Supplementary Table 10). More than 12 million scaffolds were assigned into 3,774 metagenome-assembled genomes (MAGs), encompassing more than 300 species from 43 bacterial families. Actinomycetota and Pseudomonadota were the most abundant phyla at both time points, consistent with findings of previous studies²⁶. Bacterial load did not significantly differ between ABX groups at either time point (Extended Data Fig. 6a). Alpha diversity (both evenness and richness) was significantly lower (P ≤ 0.05) in Neo-ABX and IP-ABX infants at week 1 (Extended Data Fig. 6b,c). From week 1 to week 6, both bacterial load and alpha diversity significantly increased (P < 0.001) (Extended Data Fig. 6a–c). We also assessed antimicrobial-resistant-gene carriage and observed increased carriage in Neo-ABX and IP-ABX infants (Extended Data Fig. 6d,e).

Using Dirichlet multinomial mixtures modelling to cluster metagenomic samples, we identified six community state types (CSTs) (Fig. 4a,b). CST-2 had a high centred log ratio (CLR) abundance of Bifidobacterium species, particularly Bifidobacterium breve but also included Bifidobacterium longum, Bifidobacterium bifidum and Bifidobacterium infantis, and was significantly overrepresented among No-ABX infants at both week 1 and week 6 (Fig. 4c and Extended Data Fig. 6f). Notably, this was not observed for CST-3, which had a higher CLR abundance of B. longum compared with CST-2. We next assessed whether specific bacterial taxa were differentially abundant among the ABX groups. Compared with No-ABX infants, numerous taxa, including several from the genera Enterobacter, Klebsiella, Haemophilus, Citrobacter and Streptococcus, had significantly higher (FDR < 0.05) CLR abundance in the IP-ABX infants at week 1 but not at week 6 (Fig. 4d,e, Extended Data Fig. 6g,h and Supplementary Table 11). Correspondingly, the CLR abundances of several different Bifidobacterium species were significantly lower (FDR < 0.05) in IP-ABX infants at week 1 but not at week 6 (Fig. 4d,e, Extended Data Fig. 6i and Supplementary Table 11). Consistent with the changes in the composition of the microbiota in IP-ABX infants, we detected 298 metagenomically encoded pathways that were differentially abundant (FDR < 0.05) at week 1 but none at week 6 (Supplementary Table 12). These data indicate that the effects of intrapartum antibiotics on the composition and functional capacity of the faecal microbiota were transient and had resolved by the time the first doses of vaccines were administered.

Fig. 4: Depletion of Bifidobacterium species at the time of first-dose immunization is associated with impaired antibody responses to vaccination.

a, Taxonomic composition of stool samples (n = 409) collected from study infants, as determined by shotgun metagenomics sequencing. Samples are grouped by CST. b, Multidimensional scaling (MDS) analysis of the six CSTs. c, Odds ratios of a sample (n = 396) in each CST at week 1 and week 6 of life being from an infant not exposed to antibiotics. d,e, Volcano plots showing fold change in the CLR abundance of taxa in the faecal microbiota of Neo-ABX, IP-ABX and PN-ABX infants, relative to No-ABX infants, at week 1 (No-ABX, n = 79; Neo-ABX, n = 36; IP-ABX, n = 55; PN-ABX n = 33) (d) and week 6 (No-ABX, n = 80; Neo-ABX, n = 33; IP-ABX, n = 49; PN-ABX, n = 31) (e). n represents the number of infants. f, Heatmap representing correlations between the CLR abundance of the top ten most positively correlated and top ten most negatively correlated taxa in the infant faecal microbiota at week 6 and antibody responses to vaccination (adjusted for sex and baseline titres) at month 7. The blue–red gradient represents negative and positive Spearman correlations, respectively (n = 139 with matched data across the three time points). See Supplementary Table 13 for exact correlation coefficients. Error bars in c indicate 95% confidence intervals. Statistical significance was assessed using permutational multivariate analysis of variance in b, Fisher’s exact test in c, the LinDA method in the MicrobiomeState R package in d and e, and the cor.test R function in f. All statistical tests were two-sided. The Benjamini–Hochberg method was used to adjust for multiple comparisons in d and e, with FDR < 0.05 considered to indicate statistical significance. *P < 0.05 (unadjusted). Exact FDR values for d and e can be found in Supplementary Table 11.

Source data

By contrast, the CLR abundances of various Bifidobacterium species including B. breve, Bifidobacterium pseudocatenulatum and B. bifidum (Fig. 4d,e and Extended Data Fig. 6j–l) and several Collinsella species were significantly decreased (FDR < 0.05) in Neo-ABX infants at week 6 (Supplementary Table 11), the time that the first doses of PCV13, 6-in-1 and oral rotavirus vaccines were administered. Reduced Bifidobacterium relative abundance has been reported in some but not all previous studies assessing the impact of antibiotics on the gut microbiota in early life²⁷. Consistent with the persistent changes in the composition of the faecal microbiota of Neo-ABX infants at week 6, we identified 406 metagenomically encoded pathways that were differentially abundant (FDR < 0.05) in Neo-ABX infants at this time point compared with No-ABX infants (Supplementary Table 12). These data indicate that the persistent changes induced in the composition and encoded function of the gut microbiota at the time of vaccination by neonatal antibiotic exposure—in particular, reduced abundance of Bifidobacterium species—could play a part in the suboptimal antibody responses to vaccination observed in these infants. Consistent with this, the CLR abundances of several Bifidobacterium species at week 6 were positively correlated with IgG titres against several different vaccine antigens at 7 months (Fig. 4f and Supplementary Table 13a). Correlations between abundances of taxa at week 1 and antibody responses at month 7 were weaker, although there was positive correlation (P < 0.05) between the CLR abundance of Bifidobacterium species and anti-PPS6B IgG titres (Extended Data Fig. 7a and Supplementary Table 13b). Given these associations, we next assessed whether Bifidobacterium at 6 weeks was increased in high versus low vaccine responders. The CLR abundances of B. breve, B. bifidum and B. longum—but not B. pseudocatenulatum—were significantly increased in high responders to several different vaccine antigens, particularly PPS6B (Extended Data Fig. 7b and Supplementary Table 13). Causal mediation analysis^28,29 also indicated significant positive direct relationships of both B. breve and B. bifidum with PPS6B IgG titres (P < 0.05) that were mediated by inflammation-related (BTM M11.0, P < 0.05) but not B cell-related transcriptional activity. These data indicate that antibiotic exposure may trigger a pro-inflammatory state that impedes Bifidobacterium colonization and/or growth and subsequently affects optimal vaccine responses.

Impaired PCV13 responses in GF mice

Our clinical data indicated that antibody responses to the PCV13 vaccine might be particularly dependent on signals from the gut microbiota. To assess whether this was also reflected in preclinical models, we compared antibody responses in sera collected from PCV13-vaccinated conventional specific-pathogen-free (SPF) and GF mice. PCV13-specific IgG, IgG1, IgG2c, IgG3 and IgM responses in sera were markedly impaired in GF mice (Fig. 5a and Extended Data Fig. 8a–d), as were IgG titres against PPS1, PPS3, PPS6B, PPS9V and PCV13 conjugate protein CRM₁₉₇ (Extended Data Fig. 8e–i). GF mice immunized with the 6-in-1 vaccine also had significantly impaired anti-diphtheria toxoid and anti-tetanus toxoid IgG responses; however, analogous to our observations in Neo-ABX infants, their anti-pertussis toxoid responses were not significantly different to those of SPF mice (Extended Data Fig. 8j–l). GF mice also had antibody responses comparable with those of SPF mice following immunization with the 4CMenB vaccine (Extended Data Fig. 8m). We recently reported that GF mice have B and T cell responses to the Pfizer/BioNTech COVID-19 mRNA vaccine that are comparable with those of SPF mice¹⁰. Taken together, these data indicate that as in antibiotic-treated infants, responses to PCV13 in mice are particularly dependent on signals from the gut microbiota.

Fig. 5: Immune responses to PCV13 in mice strongly depend on the gut microbiota and signalling through a MyD88-dependent pathway.

a, PCV13-specific IgG_total in sera from GF (n = 9) and SPF (n = 10) mice vaccinated with two doses of PCV13. Mock, PBS-vaccinated mice (n = 3). b–d, Frequencies and total numbers of total GC B cells (b) or CRM₁₉₇⁺ GC B cells (c) in the spleens of mock-vaccinated mice (n = 9) and PCV13-vaccinated GF (n = 10) and SPF (n = 10) mice at 2 weeks postboost; and frequencies and total numbers of T_FH cells (d) in the spleens of these mice. e, CD86 mean fluorescence intensity (MFI) on indicated myeloid cell populations in the dLN of GF and SPF mice 24 h following PCV13 or mock vaccination (n = 5 mice per group). f, PCV13-specific IgG_total antibodies in the serum of mock-vaccinated mice (n = 5) and PCV13-vaccinated Myd88^−/− (n = 10) and littermate wild-type (n = 10) mice. g, PCV13-specific IgG_total in the serum of mock-vaccinated mice (n = 5) and PCV13-vaccinated (two doses intraperitoneally, 2 weeks apart) SPF (n = 10) and GF (n = 9) mice, and GF mice born to dams colonized with a consortium of Bifidobacterium species in pregnancy and administered these strains again at days 7 and 14 postbirth (GF + Bif, n = 6). Data shown are at 2 weeks postboost. h, Frequencies and total numbers of GC B cells in the spleens of mice from g at 2 weeks postboost. i,j, PCV13-specific IgG_total (i) and IgG1 (j) in the sera of mock-vaccinated mice (n = 3), SPF mice (n = 8), GF mice (n = 8 mice), and GF mice colonized with probiotic strains L. acidophilus and B. bifidum at day 21 of life (GF + Probiotic, n = 7). Mice were vaccinated at day 28 of life. Data in a–j represent the mean ± s.e.m. Data in a–d are representative of three or more independent experiments; data in e–j are representative of two independent experiments. Statistical significance was assessed in a, f, i and j using two-way analysis of variance with Tukey’s posttest analysis for multiple comparisons, in b–e using two-tailed Student’s t-tests, and in g and h using one-way analysis of variance with Dunnett’s posttest analysis for multiple comparisons. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.001; NS, not significant. Exact P values are provided in the source data.

Source data

To investigate why, we next assessed whether germinal centre (GC) responses, which have a critical role in optimal T-dependent humoral immune responses, were suboptimal in PCV13-immunized GF mice. Compared with SPF mice, GF mice immunized with two doses of PCV13 had significantly lower total and CRM₁₉₇-specific GC B cells in their spleens (Fig. 5b,c) and draining lymph nodes (dLN) (Extended Data Fig. 8n–q). The proportion of T follicular helper (T_FH) cells in the spleen (Fig. 5d) but not in the dLN (Extended Data Fig. 8r,s) was also significantly lower in GF mice. Confocal microscopy confirmed impaired GC formation in PCV13-immunized GF mice (Extended Data Fig. 9). These data indicate that PCV13-immunized GF mice have impaired T-dependent antibody responses to vaccination because of a failure of optimal GC formation in these mice.

We next proposed that GF mice might fail to mount optimal GC responses to PCV13 because of impaired innate immune activation postvaccination. Consistent with this hypothesis, 24 h after the first dose of PCV13, we observed significant increases in surface expression of activation marker CD86 on cells of several different myeloid populations, including dendritic cells (DC), conventional dendritic cells (cDC), plasmacytoid dendritic cells (pDC), migratory dendritic cells (mDC) and resident dendritic cells (rDC) in the dLN of SPF but not GF mice (Fig. 5e). Expression of MHCII or CD80 was not induced in either GF or SPF mice at 24 h postvaccination (Extended Data Fig. 10a,b). Notably, given the associations we observed between increased CLR abundance of Bifidobacterium species and higher antibody titres in infants, mice born to GF dams colonized with B. breve during pregnancy (to ensure colonization from birth) had significantly higher CD86 expression at 24 h after PCV13 vaccination compared with mice that remained GF (Extended Data Fig. 10c).

Given that GF mice seemed to have impaired innate immune responses to PCV13 immunization, we proposed a further hypothesis that signals from the microbiota might act through an MyD88-dependent pathway. Consistent with this, serum IgG, IgG1, IgG2c and IgM responses to PCV13 were severely impaired in Myd88^−/− mice (Fig. 5f and Extended Data Fig. 10d–f). A previous study found that TLR5-mediated sensing of the microbiota was required for antibody responses to a non-adjuvanted influenza vaccine⁹; however, we have previously shown that antibody responses to the alum-adjuvanted PCV13 vaccine are not impaired in Tlr5^−/− mice¹. Given this, we instead investigated responses to PCV13 in Tlr4^−/− and Tlr2^−/− mice and found a trend suggestive of impaired PCV13-specific antibody responses in Tlr4^−/− but not Tlr2^−/− mice (Extended Data Fig. 10g,h). These data indicate that sensing of the microbiota through an MyD88-dependent pathways may be required for optimal antibody responses to the PCV13 vaccine; however, further work is needed to determine the upstream pattern recognition receptor(s) involved, noting that redundancy between different receptors is possible.

Bifidobacteria enhance PCV13 responses

Our clinical data suggested that loss of Bifidobacterium species in infants at the time of immunization leads to impaired antibody responses to PCV13 and other vaccines. To assess whether administering a consortium of Bifidobacterium species to GF mice could restore PCV13-specific antibody responses, we administered a mixture of B. breve, B. longum subsp. longum and B. longum subsp. infantis to pregnant dams to ensure colonization of their offspring from birth. These pups (GF + Bif) and pups born to GF or SPF mice were vaccinated with PCV13 at around day 21 of life and again 2 weeks later. Consistent with our hypothesis, PCV13-specific IgG, IgG1 and IgM antibody responses in GF + Bif pups were significantly higher 2 weeks following the second dose of PCV13 compared with those in GF mice (Fig. 5g and Extended Data Fig. 10i,j). Compared with GF mice, GF + Bif mice also had significantly higher proportions or total numbers of GC B cells in their spleens (Fig. 5h) and dLN (Extended Data Fig. 10k). Proportions or numbers of T_FH cells were also higher in GF + Bif mice (Extended Data Fig. 10l,m). Notably, we observed no increase in antibody responses to PCV13 in GF + Bif mice after only a single dose of vaccine (Extended Data Fig. 10n–p).

Next, we wanted to establish whether responses in GF mice to PCV13 could be restored by directly administering a probiotic intervention to the GF mice closer to the time of vaccination (rather than through their mothers). To assess this, we compared responses to PCV13 in SPF mice, GF mice and GF mice colonized with strains from commercial probiotic Infloran, 1 week before vaccination with PCV13 (GF + Probiotic). We selected Infloran as it contains a mixture of B. bifidum and Lactobacillus acidophilus and also because it is commonly used in neonatal intensive care units to prevent necrotizing enterocolitis^30,31 and thus has an established clinical safety profile with clear translational potential. In addition, B. bifidum was among the Bifidobacterium species that we observed to show decreased abundance in Neo-ABX infants and increased abundance in infants that were high vaccine responders (Extended Data Fig. 7b). PCV13-specific IgG and IgG1 responses in GF mice were significantly enhanced by colonization of the mice with the Infloran probiotic strains (Fig. 5i,j). Notably, colonization of GF mice with either strain alone was insufficient to enhance antibody responses to PCV13 (Extended Data Fig. 10q–s). Bifidobacterium spp. are known to use a variety of cross-feeding synergistic strategies involving other members of the gut microbiota to access carbohydrates that they cannot metabolise;³² this potentially explains why neither culture alone enhanced antibody responses. Consistent with this, the bacterial load in mice colonized with both the B. bifidum and L. acidophilus strains simultaneously was two log-fold higher than that of mice colonized with either strain alone (Extended Data Fig. 10t). Colonizing GF mice with a range of other commensal isolates before PCV13 vaccination did not enhance responses to PCV13 vaccination (Extended Data Fig. 10u–w). In conclusion, our data indicate that restoring a healthy Bifidobacterium-rich microbiota in antibiotic-exposed infants, before vaccination, may lead to enhanced antibody responses to vaccination and increased protection against infectious diseases.

Discussion

We have previously shown that early-life antibiotic exposure in mice leads to impaired responses to a range of different infant vaccines¹; this motivated us to conduct the prospective clinical study we report here. Consistent with these previous preclinical data, we found that infants exposed to antibiotics, particularly direct antibiotic treatment in the neonatal period, had significantly reduced serum IgG titres against several vaccine antigens at 7 and 15 months of age. Our results, which add to the findings of several important previous studies (Supplementary Discussion), indicate that immune responses to certain types of vaccine, particularly protein-polysaccharide conjugate vaccines such as PCV13, may be more dependent on signals from the gut microbiota than others. Further investigation is needed to determine why this is the case; however, it may be related to the lower immunogenicity of polysaccharide conjugate vaccines compared with adjuvanted protein vaccines and other platforms such as mRNA vaccines¹⁰.

Our clinical and preclinical data suggest that Bifidobacterium species in the infant gut microbiota have an important role in mediating optimal innate, GC and serum antibody responses to PCV13 and other vaccines and that administering probiotics to infants exposed to neonatal antibiotics may be a feasible, cheap and safe intervention to enhance responses to vaccination. This needs to be rigorously tested in randomized controlled trials in appropriate populations (Supplementary Discussion). Consistent with our data, some previous correlative studies have reported an association between the relative abundance of Bifidobacterium species and responses to the Bacillus Calmette–Guérin (BCG), oral polio and tetanus toxoid vaccines^33,34. In addition, GF mice colonized with B. longum subp. infantis have recently been found to have increased T cell responses to the BCG vaccine³⁵. Notably, intrapartum antibiotic exposure was not associated with reduced antibody responses to vaccination in our study (Supplementary Discussion). A possible explanation is that at the time of first-dose vaccination, abundances of Bifidobacterium species were not significantly reduced in infants exposed to intrapartum antibiotics, whereas they were significantly reduced in infants who were directly treated with antibiotics.

Despite our important findings, our study had several limitations. The sample size was relatively modest, especially when the antibiotic exposure group was broken down further according to the type of antibiotic exposure. Further studies in larger cohorts are now warranted. Encouragingly, our results in infants were strongly concordant with our observations in animal models. Our study excluded infants born by caesarean section and those born to mothers with a body mass index greater than 30. Future research should investigate whether these and other factors that alter the infant gut microbiota³⁶ can also influence immune responses to vaccination. A recent study reported that infants born by caesarean section had significantly reduced IgG titres against pneumococcal capsular polysaccharides in saliva (no serum was collected in that study)³⁷. Furthermore, assessment of gene expression responses at more time points postvaccination and using single-cell technologies could further refine our understanding of the link between altered transcriptional responses in infants exposed to antibiotics and suboptimal responses to vaccination (Supplementary Discussion). Finally, our study focused predominantly on binding antibody titres, given that these are the established correlates of protection for most vaccines; however, we found that the opsonophagocytic activity of serum from infants treated with antibiotics in the neonatal period against one of the three PCV13 polysaccharides, PPS6B, was significantly reduced. Furthermore, there was a modest reduction in the bactericidal activity (against N. meningitidis) of serum collected from infants vaccinated with the 4CMenB meningococcal B vaccine that had been exposed to any antibiotics, compared with those that had not been exposed to antibiotics. Future studies should also assess the impact of antibiotic exposure on functional antibody responses in more detail, as well as T cell responses to vaccination. In conclusion, we found that early-life antibiotic exposure was associated with significantly impaired vaccine immunogenicity, with a particular impact on immune responses to the PCV13 vaccine. These effects seemed to be mediated by a reduction in abundance of Bifidobacterium species, which in turn led to altered baseline immune status. Our findings emphasize the need to broaden the rationale for antibiotic stewardship in early life beyond antimicrobial resistance, as well as the potential of precision vaccination approaches to optimize vaccine-mediated protection from infectious diseases.

Methods

Study design

Infants were enrolled in the study at the Women’s and Children’s Hospital Adelaide, Australia, between April 2017 and March 2021 as part of the Antibiotics and Immune Responses (AIR) study, a prospective, observational clinical study (ACTRN12617000856314). Potential participants were approached by study research nurses in postnatal wards and invited to participate if the mother and their infant met the study eligibility criteria. Study information had also been previously provided to women delivering at the Women’s and Children’s Hospital through pamphlets in antenatal care packs. Following written informed consent, stool collection packs were provided to families, and medical case notes were reviewed to confirm eligibility criteria and antibiotic exposure status. Ethics approval for the study was obtained from the Human Research Ethics Committee of the Women’s and Children’s Health Network (approval number HREC/17/WCHN/19), and the authors affirm that they have complied with all relevant ethical regulations.

Exclusion criteria were gestation of less than 37 weeks; maternal body mass index greater than 30 (at first-trimester antenatal visit); maternal sepsis (defined by laboratory-confirmed bacterial infection in blood cultures or cerebrospinal fluid) during pregnancy; infant delivered by caesarean section; confirmed sepsis in infant (defined by laboratory-confirmed bacterial infection in blood cultures or cerebrospinal fluid); known or suspected disorder of the immune system that would prevent an immune response to the vaccines, such as congenital or acquired immunodeficiency, or receiving systemic immunosuppressive therapy; infant with suspected or confirmed HIV, major congenital abnormality or serious illness; maternal or infant participation in a clinical study that could interfere with participation in this study; and anything that would place the individual at increased risk or preclude their full compliance with or completion of the study. For immunogenicity analyses, only participants who received at least one vaccine at study visit 1 were included.

Participants were allocated to an antibiotic exposure group for immunogenicity analyses on the basis of medical record review and discussion with the mother to ascertain any antibiotic exposure before vaccination at study visit 1 (approximately 6 weeks of age). Antibiotic exposure groups were defined a priori as follows: infants with no direct or maternal antibiotic exposure (n = 80, No-ABX); (1) infants who received at least 48 h of antibiotic treatment in the neonatal period (the first 28 days after birth), with or without maternal antibiotic exposure (n = 32, Neo-ABX); (2) infants whose mothers received intrapartum antibiotics (within 28 days before or during delivery) and no direct infant antibiotic exposure up to 6 weeks of age (n = 49, IP-ABX); and (3) infants who did not meet the criteria for inclusion in the No-ABX, Neo-ABX or IP-ABX group. Upon study completion, all infants in group 3 were those who could have been exposed to maternal antibiotics during breastfeeding in the postnatal prevaccination period, so we called this group the postnatal exposure group (PN-ABX, n = 30). Researchers conducting the serological assays, omics analyses and other data analyses were blinded to antibiotic exposure group allocations. Participant data were recorded in a secure REDCap (https://project-redcap.org/) database.

Sample collection

Faecal samples for metagenomics were collected from the nappies (diapers) of each infant by parents at week 1 (days 2–10 of life) and week 6 (days 38–56 of life) and placed into Zymo DNA/RNA Shield faecal collection tubes (Zymo Research) before being stored at −80 °C. Blood (3–5 ml) was obtained from infants at approximately 6 weeks, 7 weeks, 7 months and 15 months. At 6 weeks, 1–3 ml of whole blood was placed into BD Vacutainer serum tubes, and 2 ml was placed in a PAXgene blood RNA tube (BD) for RNA preservation and stored at −80 °C. At 7 weeks, up to 1 ml of blood was collected in a BD Vacutainer K3 EDTA tube, 2 ml was collected for flow cytometry analysis, and the remainder (up to 2 ml) was placed into a PAXgene (BD) tube. At 7 months and 15 months, 3–5 ml of blood was collected into BD Vacutainer serum tubes.

Vaccine schedule

Infants were vaccinated with PCV13 (Prevenar 13, Pfizer) and the combination 6-in-1 vaccine (Infanrix Hexa, GSK) at 6 weeks and 4 and 6 months of age and with the oral rotavirus vaccine (Rotarix, GSK) at 6 weeks and 4 months of age as per the NIP Schedule for South Australia as of 2017. Most of the infants (86%) were also vaccinated with the hepatitis B vaccine at birth (either H-B-Vax II Paediatric or Engerix B Paediatric). Infants in this study were also vaccinated with the meningococcal B vaccine (4CMenB/Bexsero, GSK) at 7 weeks, 4 months and 6 months of age, with a booster dose at 12 months of age. The meningococcal ACWY (Nimenrix, Pfizer) and the measles, mumps, rubella (M-M-R II, Merck Sharp & Dohme; or Priorix, GSK) vaccines were administered at 12 months of age as part of the standard NIP schedule (although antibody responses were not evaluated for these vaccines in this study). In this study, the booster dose of PCV13 was provided at 15 months rather than the current NIP scheduled time point of 12 months of age. All vaccines were administered intramuscularly except for the oral rotavirus vaccine. All vaccines were administered in line with Australian Immunisation Handbook recommendations.

Pneumococcal polysaccharide-specific IgG

Antibodies against PCV13 serotypes were measured as previously described^13,15,16, with modifications. In brief, Bio-Plex COOH-microspheres (6.25 × 10⁶, 500 µl, Bio-Rad) were conjugated overnight with optimized doses of DMTMM-modified pneumococcal polysaccharides (ATCC) in 1× PBS pH 7.2 (Life Technologies) as per Schlottmann et al.⁴¹. The optimal coating doses were a 1:1 dose, 500 µl microspheres with 500 µl DMTMM-modified polysaccharides for serotypes 1, 3, 5, 7F and 9V; and a 1:0.5 dose, 500 µl microspheres with 250 µl DMTMM-modified polysaccharides for serotypes 4, 6A, 6B, 14, 18C, 19A, 19F, 23F and 11A.

A ten-step, three-fold serial dilution of the international reference standard for human antipneumococcal capsule, 007sp (NIBSC), two in-house quality controls and test samples were diluted in adsorbent buffer containing 0.05% Tween-20, 2% N-butyl benzenesulfonamide (Sigma-Aldrich) and 2.5 µg ml⁻¹ CWPS Multi (SSI Diagnostica) for 30 min to remove non-specific antibodies before incubation with polysaccharide-conjugated microspheres. Equal volumes (25 µl) of diluted samples and bead-mix containing 4,000 microspheres per region were combined on a MultiScreen filter plate (Merck) prewetted with 50 µl of 0.05% PBS with Tween-20 (PBS-T). Plates were incubated at room temperature in the dark on an orbital shaker at 500 rpm for 20 min. They were then washed twice with 100 µl 0.05% PBS-T, followed by addition of 100 µl (1:150) RPE-conjugated goat anti-human IgG Fc secondary antibody (Jackson ImmunoResearch Laboratories Inc.) and incubation for a further 20 min. Following washing, the beads were resuspended in 125 µl 0.05% PBS-T and read on a Bio-Plex-200 machine (Bio-Rad).

Diphtheria, tetanus and pertussis-specific IgG

Antibody concentrations specific to pertussis toxoid, pertactin, filamentous haemagglutinin, tetanus toxoid and diphtheria toxoid were measured as previously described^14,42.

Hib-PRP- and HB-sAg-specific IgG

Antibodies against Hib-PRP and HB-sAg were measured as previously described for diphtheria, tetanus and pertussis (DTP)^14,42, changing only the antigen-coated microspheres, quality controls and reference sera used. In brief, DMTMM-modified⁴¹ Hib-PRP 12/306 (NIBSC) was conjugated to Bio-Plex COOH-microspheres (Bio-Rad) as outlined above for pneumococcal polysaccharides, with an optimal dose of 1:1, 500 µl microspheres with 500 µl DMTMM-modified polysaccharide. Recombinant HB-sAg subtype adr (Prospec Bio) was conjugated as previously described for DTP antigens, with an optimal coating concentration of 10 µg ml⁻¹. A ten-step, three-fold serial dilution of the international standard for anti-H. influenzae type b, 09/222 (NIBSC), was used as the reference serum, after anti-HB-sAg titres had been determined in that serum. Anti-HB-sAg titres in 09/222 were quantified against the second international standard for anti-HB-sAg, 07/164 (NIBSC), using the final Bio-Plex protocol; they were further validated using the current gold-standard method, the Alinity i Anti-HBs kit (PathWest), which showed excellent concordance (coefficient of variation less than 5%).

IgA Serology

Rotavirus IgA antibodies were measured by validated enzyme-linked immunoassay, as previously described^17,18.

High and low vaccine responders

High and low vaccine responders were defined as infants with IgG titres in the 75th percentile and 25th percentile, respectively, for each of the antigens assessed.

Multiplex opsonophagocytosis method

Multiplex opsonophagocytosis assays were performed on three serotypes (6B, 9V and 18C) using a previously published method^43,44. Serum samples were heat-inactivated at 56 °C for 30 min; then, serial dilutions were made, and 20 μl per well of each dilution was tested in duplicate in 96-well u-bottomed microtitre plates (Interpath Services). Serial three-fold dilutions were used. Frozen aliquots of target pneumococci were thawed, washed with opsonization buffer (HBSS with magnesium and calcium, and 0.1% gelatin), and diluted to approximately 5 × 10⁴ CFU ml⁻¹ of each serotype. All serotypes were then pooled in one bacterial suspension, and 10 μl of the bacterial suspension was added to each well. Plates were incubated at room temperature with shaking (mini Orbital Shaker; Bellco Biotechnology) at 700 rpm for 30 min. Then, 10 μl of baby rabbit complement and 40 μl of HL60 cells (approximately 1 × 10⁷ cells ml⁻¹) were then added to each well (the HL60 cells had been washed twice with HBSS before use). Plates were incubated in a tissue culture incubator (37 °C in a 5% CO₂ atmosphere) with shaking at 700 rpm. After 45 min, plates were placed on ice to stop the reaction, and 10 μl of the final reaction mixture was spotted onto Todd-Hewitt agar plates supplemented with 0.5% yeast extract. An overlay agar containing one of the selective antibiotics (selective markers for target bacteria) was added to one of each of the Todd-Hewitt agar plates, which was then incubated overnight at 37 °C with 5% CO₂. The number of bacterial colonies was enumerated using a ProtoCol3 automatic colony counter (Synbiosis). Results were expressed as opsonization indices, defined as the interpolated dilutions of serum that killed 50% of bacteria. The lower limit of detection in the assay was 4. The opsonization indices of samples that did not kill 50% of bacteria were reported as 2 for purposes of analysis. A cut-off of 8 was used to define a positive response.

Human serum bactericidal activity assays

Human serum bactericidal activity assays were performed on 140 matched study participant serum samples (from blood draws at 6 weeks, 7 months and 15 months of the study) as described previously⁴⁵. Briefly, heat-inactivated study participant sera were serially diluted two-fold in assay buffer and inoculated with approximately 1 × 10³ CFU of N. meningitidis strain NZ98/254. Exogenous human complement (25% v/v, Pel-Freez Biologicals) was added, and assay microtitre plates were incubated at 37 °C with 5% CO₂ for 60 min; then, samples were plated on GC agar base plates prepared using the tilt method and incubated at 37 °C with 5% CO₂ overnight to enable CFU enumeration. Bactericidal titres are reported as the reciprocal of the lowest dilution of participant serum that induced at least 50% killing relative to the no-serum control. The serologic correlate of protection for N. meningitidis is considered to be a human serum bactericidal activity titre greater than or equal to 4 (ref. ²²). Statistical analyses were performed using one-way analysis of variance and two-tailed paired Student’s t-test.

Infant blood preparation for flow cytometry

For flow cytometry analysis, 450 µl of infant blood was transferred from a BD Vacutainer K3 EDTA tube into a sterile 10-ml tube within 3 h of collection, and red blood cells were lysed by addition of 4.5 ml of 1× BD Pharm Lyse for 15 min at room temperature. Cells were then pelleted by centrifugation at 200g for 5 min, the supernatant was aspirated, and the cell pellet was resuspended in 10 ml flow cytometry buffer and centrifuged again. Cells were subsequently resuspended in 150 µl flow cytometry buffer and divided equally into three tubes for staining. Fc receptors were first blocked using Human TruStain FcX (BioLegend) in Brilliant Stain Buffer (BD) for 10 min on ice. Three panels were designed for same-day flow cytometry analysis (see Supplementary Table 8 for a complete list of cell subsets identified in each panel). Panel 1 was a pan-leukocyte panel enabling identification of neutrophils, monocytes, dendritic cells, eosinophils, basophils, T cells, B cells and natural killer cells. Cells were stained with a cocktail of anti-CD3 FITC (HIT3a, 1:20, BD Biosciences), anti-CD11c PE (B-ly6, 1:20, BD Biosciences), anti-HLA-DR APC-H7 (G46-6, 1:40, BD Biosciences), anti-CD15 V500 (HI98, 1:80, BD Biosciences), anti-CD45 BUV395 (HI30, 1:80, BD Biosciences), anti-CD16 BV421 (3G8, 1:143, BD Biosciences), anti-CD14 AF647 (Mop-9, 1:143, BD Biosciences), anti-CD123 BV711 (9F5, 1:143, BD Biosciences), anti-CD56 PE-Cy7 (5.1H11, 1:40, BioLegend), anti-CD19 BV605 (HIB19, 1:80, BioLegend), anti-CD20 BV786 (2H7, 1:80, BioLegend), anti-FcεRIα PerCP Cy5.5 (AER-37, 1:80, BioLegend) and anti-Siglec8 PE Dazzle (7C9, 1:143, BioLegend). Panel 2 enabled characterization of around 20 functional subsets of B and T cells including naive, central memory and effector memory CD4⁺ and CD8⁺ T cells; regulatory T cells; and naive, memory and plasma B cells. The antibody cocktail consisted of anti-CD19 PE (HIB19, 1:10, BD Biosciences), anti-CCR7/CD197 AF647 (150503, 1:10, BD Biosciences), anti-CD127 BV421 (A019D5, 1:40, BioLegend), anti-CD38 BV510 (HIT2, 1:40, BD Biosciences), anti-HLA-DR APC-H7 (G46-6, 1:40, BD Biosciences), anti-CD3 BUV395 (UCHT1, 1:40, BD Biosciences), anti-CD4 BV605 (RPA-T4, 1:80, BD Biosciences), anti-CD8 Cy7 (RPA-T8, 1:80, BD Biosciences), anti-IgD PerCP Cy5.5 (IA6-2, 1:80, BD Biosciences), anti-CD27 BV711 (L128, 1:143, BD Biosciences), anti-CD25 PE Dazzle594 (M-A51, 1:80, BioLegend), anti-CD20 BV786 (2H7, 1:80, BioLegend), and anti-CD45RA FITC (HI100, 1:143, BioLegend). Panel 3 was designed to assess specific T helper (T_H) subsets, including T_H1, T_H2, T_H9, T_H17, T_H22 and natural killer T cells, using anti-CD3 BUV395 (UCHT1, 1:50, BD Biosciences), anti-CD4 BV605 (RPA-T4, 1:100, BD Biosciences), anti-CD8 PE Cy7 (RPA-T8, 1:100, BD Biosciences), anti-CXCR3 BV421 (1C6/CXCR3, 1:100, BD Biosciences), anti-CCR10 APC (1B5, 1:100, BD Biosciences), anti-CCCR6 BV711 (11A9, 1:50, BD Biosciences), anti-CCR4 PECF594 (1G1, 1:50, BD Biosciences), anti-CD45RA FITC (HI100, 1:100, BioLegend) and CD1d-Tet PBS57 PE (1:5,000, NIH). See Supplementary Table 8 for a comprehensive list of antibody sources, catalogue numbers and fluorophores.

After each antibody cocktail had been added, samples were incubated in the dark on ice for 30 min before being washed and resuspended in 300 µl of flow cytometry buffer. Panel 1 was added to a BD Trucount FACS tube (BD), whereas panels 2 and 3 were added to round-bottomed 5-ml FACS tubes alongside 10 µl liquid counting beads (BD). Counts were generated as previously described²⁵. Dead cells were stained by addition of DNA binding dye DAPI and were excluded from analysis. Samples were run on a FACSymphony (BD) flow cytometer using FACSDiva software, and results were analysed using FlowJo v.10.6.1 (Tree Star).

Extraction of RNA from blood

RNA extraction and elimination of genomic DNA were carried out using a PAXgene Blood RNA kit (Qiagen) per the manufacturer’s instructions, with final elution into 80 μl RNase-free water. A further RNA precipitation reaction was carried out. Briefly, RNA was resuspended in 2.5 × 100% ethanol and 10% sodium acetate and spun at 12,000g for 30 min at 4 °C. Samples were washed in 75% ethanol. Pellets were air-dried and resuspended in 40 µl of RNase-free water. The total RNA yield was determined using a Bioanalyzer 2100/TapeStation (Agilent) and Qubit (Thermo Fisher Scientific). Before library preparation, 0.2 ng of synthetic reference RNA standards (Sequins, Garvan Institute) were added to 250 ng of RNA.

RNA library preparation

Total RNA was converted to strand-specific Illumina-compatible sequencing libraries using a Nugen Universal Plus mRNA-Seq library kit from Tecan, with Any Deplete probes and custom gamma globin depletion per the manufacturer’s instructions (MO1442 v.2). Briefly, 250 ng of total RNA was polyA-selected, and the mRNA was fragmented before reverse transcription. Any deplete and custom gamma globin depletion was performed before strand selection II complementary DNA (cDNA) synthesis with pooled gamma depletion oligos, per the manufacturer’s instructions. The resulting cDNA was end-repaired, followed by ligation of Illumina-compatible barcoded sequencing adaptors. The cDNA libraries were strand-selected and PCR amplified for 12 cycles before quality control assessment using a TapeStation (Agilent Technologies). Library concentrations were assessed using a Qubit fluorescence assay (Thermo Fisher Scientific). Sequencing pools were generated by mixing equimolar amounts of compatible sample libraries on the basis of the Qubit measurements. The library pool was sequenced using an Illumina NovaSeq 6000 S1 flow cell with 2 × 100-bp paired-end reads.

RNA sequencing analysis

Sequence read quality was assessed using FastQC v.0.11.4 (ref. ⁴⁶) and summarized with MultiQC v.1.8 (ref. ⁴⁷), followed by quality control with Trimmomatic v.0.38 (ref. ⁴⁸) with a window size of 4 and an average quality score of 25. Reads of fewer than 50 nucleotides after trimming were discarded. Reads that passed all quality control steps were then aligned to the human genome (GRCh38) using HISAT2 v.2.1.0 (ref. ⁴⁹). The gene count matrix was generated with FeatureCounts version 1.5.0-p2 (ref. ⁵⁰) using the union model with Ensembl v.93 annotation. Data were then imported into R v.4.2.0 for further analysis. Counts were normalized using the trimmed mean of M values method in EdgeR v.3.38 (ref. ⁴⁰), followed by multidimensional scaling analysis and differential gene expression analysis performed with the glmLRT function. For differential expression analysis comparing week 6 and week 7, a paired design was used, fitting each participant into the model as well as each time point. As each participant’s samples were processed together, this intrinsically controlled for batch effects. For non-paired comparisons, processing batch and infant sex were fitted into the model to adjust for those effects. Unknown sources of variation were identified using svaseq v.3.46.0 (ref. ³⁸) and adjusted for before multidimensional scaling and gene set variation analysis. The gene count matrix was filtered to remove genes with a count per million of less than 5 in 15 samples before differential expression analysis. The Benjamini–Hochberg method was used to correct for multiple comparisons. Gene set enrichment analysis was carried out using the fgsea R package v.1.22.0 (ref. ³⁹) with predefined BTMs²⁴, with input ranks as fold changes calculated by edgeR. Gene set variation analysis was used to calculate a per-sample activity score for each of the BTMs (excluding unannotated modules labelled ‘TBA’) using R Bioconductor package GSVA v.1.44.1 (ref. ⁵¹). Spearman correlation analysis was performed using the Hmisc v.4.7-0 package in R v.4.2.

DNA extraction from faecal samples

Two millilitres of the faecal suspension from the Zymo DNA/RNA Shield faecal collection tube (Zymo Research) was transferred into a sterile microcentrifuge tube and pelleted by centrifugation at 16,200g for 20 min at 4 °C. The supernatant was then removed, and DNA extraction was performed on the faecal pellet using the PowerLyzer PowerSoil Kit protocol (Qiagen) as described previously⁵².

Quantitative real-time PCR analysis to determine bacterial load in infant stool samples

Total bacterial load in the samples was determined by quantitative PCR (qPCR) of the 16S ribosomal RNA (rRNA) gene using a SYBR-based assay, as described previously⁵³. Each qPCR reaction comprised 1× PowerUP SYBR Green Master Mix (Applied Biosystems by Thermo Fisher Scientific), 0.2 µM of each forward and reverse primer, and 1 µl of DNA in a 35-µl total reaction volume. Each reaction was then aliquoted into three technical replicates of 10 μl for qPCR analysis. Numbers of 16S rRNA gene copies were calculated on the basis of a standard curve generated by serial dilutions of Escherichia coli genomic DNA and normalized against the faecal weight (in mg) of each sample.

Metagenomic library preparation

Shotgun metagenomics libraries were generated as previously described with modifications⁵⁴. Briefly, libraries were generated for samples with a DNA concentration of at least 0.2 ng µl⁻¹ using a Nextera DNA XT Library Preparation Kit and indexed using a Nextera XT Index Kit v.2 (Illumina). Quality control checks for DNA fragment size following PCR enrichment were performed using a DNA 1000 chip on a Bioanalyzer (Agilent Technologies). All libraries with a concentration of 5 nM and above were pair-end sequenced (2 × 151 bp) on the NovaSeq 6000 platform (Illumina).

Metagenomic data analysis

Read quality was assessed using FastQC v.0.11.4 (ref. ⁴⁶) and summarized with MultiQC v.1.8 (ref. ⁴⁷). Adaptors were removed with cutadapt v.1.18, and quality-filtering and trimming were performed using Trimmomatic v.0.38 (ref. ⁴⁸) with a window size of 4, average quality score of 25 and minimum length of 50 nucleotides. The percentage of human DNA per sample was estimated using Kraken2 v.2.1.1 (ref. ⁵⁵). MAG bins were generated by assembling each of the 409 samples individually using metaSPAdes v.3.15.2 (ref. ⁵⁶). Reads were then aligned back to this assembly catalogue, and the mapped output was used for binning with MetaBAT2 v.2.15 (ref. ⁵⁷). Counts per bin were then calculated on the basis of SAM files output from bowtie2 v.2.5.0. Each MAG bin was taxonomically classified using GTDB-Tk v.2.1.0 and reference data release 207 (ref. ⁵⁸). This MAG bin count matrix was then imported into R for further analyses. Count data were transformed to relative abundances before being visualized using bar plots and were normalized by CLR elsewhere as implemented in R library PhyloSeq v.1.28. Differential abundance analysis was performed with LinDA as implemented in R library MicrobiomeStat v.1.18 with raw count data as input. Counts were filtered for taxa detected in at least 5% of samples before testing with the prev.filter flag in LinDA⁵⁹. Spearman correlation analyses were performed using functions in base R with the CLR-normalized data as input. Alpha and beta diversity analyses were performed in R using PhyloSeq v.1.28 (ref. ⁶⁰). Metagenomic samples were clustered into CSTs with Dirichlet multinomial mixture modelling (R package DirichletMultinomial v.1.38.0), with the optimal number of clusters calculated using the lowest LaPlace approximation. Quality-filtered reads were also used as input to HUMAnN v.3.0 (ref. ⁶¹) for generation of microbial pathway data and summed to represent counts for each MetaCyc pathway and gene ontology term (biological processes only). Pathway count data were also CLR-normalized. To account for variation in 16S rRNA copy numbers between samples, which could have biased qPCR estimates of bacterial load, a mean copy number per sample was calculated on the basis of the species-level relative abundance profiles from the shotgun metagenomic data and the copy number from rrnDB⁶². Antimicrobial resistance genes were analysed by prediction of the encoded protein sequences from MAGs using prodigal v.2.6.3, clustering at 95% identity with cd-hit v.4.8.1, and conversion to amino acid sequences using transseq to generate a non-redundant predicted protein catalogue. The non-redundant catalogue was aligned to the Comprehensive Antibiotics Resistance Database (v.3.2.7) using the Resistance Gene Identifier (v.5.2.1) with default parameters except the addition of ‘–exclude_nudge’. Per-gene counts were generated by aligning reads using bowtie2 against the gene catalogue and generating a per-sample count with SAMtools v.1.17. Reads per kilobase per million mapped reads values were generated in R, summed by antibiotic class and compared between groups using a Wilcoxon rank sum test, controlling for multiple comparisons with the Benjamini–Hochberg method. Causal mediation analysis was performed in R with the mediation R package (v.4.5.0)²⁸.

Preclinical mouse experiments

C57BL/6J mice (The Jackson Laboratory; RRID: IMSR_JAX:000664) were bred and maintained under specific and opportunistic pathogen free (SPF) conditions or gnotobiotic conditions at the South Australian Health and Medical Research Institute (SAHMRI). SPF mice were kept in cages with free access to commercial pelleted food and water and were housed under standardized conditions with regulated daylight, humidity and temperature. Gnotobiotic (GF) C57BL/6 mice (Translational Research Institute) were housed in positively pressurized HEPA-filtered isolators at the SAHMRI Preclinical, Imaging and Research Laboratories (PIRL) with access to autoclaved commercially pelleted food and sterilized water ad libitum, with regulated 12 h day/night and regulated humidity (45–75%) and temperature (18–24 °C). All experiments were carried out in accordance with protocols approved by the SAHMRI Animal Ethics Committee. Each vaccine experiment was conducted as an independent experiment. Equal numbers of male and female mice were used. Young mice (aged approximately 3 weeks) or dams (aged 8–16 weeks) and pups were randomized to the treatment groups. Investigators were not blinded to the mouse treatment groups. The appropriate sample size for the murine experiments was determined on the basis of data from our previous mouse vaccine experiments.

GF mouse dams and their pups remained GF (as determined by culturing of faecal samples and 16S rRNA gene qPCR) or were colonized as previously described⁶³ by means of oral gavage with a consortium of Bifidobacterium species consisting of B. breve (DSM 20213, Leibniz Institute), B. longum subspecies longum (DSM 20219) and B. longum subspecies infantis (DSM 20088) at approximately 14 days postcoitus (E14) and at day 7 and day 14 postbirth. Successful colonization was confirmed by culturing and Sanger sequencing. For culturing, Bifidobacterium isolates were plated on MRS agar (Thermo Fisher Scientific) enriched with 1 mg ml⁻¹ cysteine, and single colonies were isolated, confirmed using a MALDI Biotyper (Bruker) and cultured for 48 h at 37 °C using BD GasPak pouch systems (BD). To assess whether administration of a commercially available probiotic could restore antibody responses to PCV13, GF mice were colonized by means of oral gavage at day 21 of life with 2 × 10⁷ CFU L. acidophilus and B. bifidum isolated from commercial probiotic Infloran (Laboratorio Farmaceutico). Further GF mice were similarly colonized with 2 × 10⁷ CFU with each of the strains individually. L. acidophilus and B. bifidum cultures were prepared by serially diluting Infloran on to MRS agar plates (Thermo Fisher Scientific) enriched with cysteine for L. acidophilus or reinforced clostridial agar (Thermo Fisher Scientific) for B. bifidum, under anaerobic conditions. Single colonies were identified using a MALDI Biotyper (Bruker). Further GF mice, aged 21 days, were colonized with Lactobacillus murinus (DSM 20452), Enterococcus gallinarum (DSM 24841), Akkermansia muciniphila (DSM 22959), Bacteriodes acidifaciens (DSM 15896), Blautia producta (DSM 2950), Lactobacillus plantarum (ATCC 202195, The Global Bioresource Centre) or Enterobacter cloacae¹.

To produce littermate wild-type and knockout mice for the experiments in Myd88^−/−, Tlr2^−/− and Tlr4^−/− mice, pairs of heterozygous Myd88^+/−, Tlr2^+/− and Tlr4^+/− dams (8–12 weeks of age) were bred with heterozygous studs. Equal numbers of wild-type and knockout offspring were cohoused for studies.

Vaccination of mice

Mice were vaccinated with 3 μg PCV13 (Pfizer), 8 μg Infanrix Hexa combination vaccine (GSK) or 17.5 μg Bexsero meningococcal serogroup B vaccine (GSK) intraperitoneally at day 28 postbirth and boosted with the same dose 2 weeks later. Control mice were mock-vaccinated with 100 μl of PBS. Four to eight sex-matched mice per group were vaccinated or mock-vaccinated in each experiment. Sera were collected 2 and 4 weeks postimmunization in each experiment and also at 6 and 8 weeks in selected experiments.

Murine faecal sampling

Faecal samples were collected from mice at the start of each experiment and at each experimental time point to confirm the GF or colonization status of the mice. Faecal pellets were aseptically collected, immediately snap-frozen and stored at −80 °C.

DNA extraction from murine faecal samples

Faecal samples were individually weighed, resuspended in 1 ml of phosphate-buffered saline (PBS) (pH 7.2) by vortexing and pelleted by centrifugation at 13,000g for 5 min. Microbial DNA extraction was performed on the faecal pellets using a DNeasy PowerSoil Kit (Qiagen) with the following minor modifications: faecal pellets were resuspended in 750 μl of PowerSoil bead solution and 60 μl solution C1 in a PowerSoil bead plate and incubated at 65 °C for 10 min before bead beating as per previously described protocols⁶³.

16S rRNA gene sequence analysis

Murine faecal DNA extracts were used to generate amplicons of the V4 hypervariable region of the 16S rRNA gene as described previously⁶³. Sequencing of the amplicon library was performed using an Illumina MiSeq system (2 × 300-bp run). Paired-end 16S rRNA gene sequences were demultiplexed and imported into QIIME2 (release 2023.2) for processing⁶⁴. Sequences were error-corrected, and counts of error-corrected reads per sample, which we refer to herein as exact sequence variants, were generated with DADA2 (ref. ⁶⁵). Taxonomy was assigned to sequences with the sklearn plugin for QIIME2 with an 80% confidence threshold, using the SILVA v.138 database⁶⁶. Exact sequence variant counts were converted to relative abundance, which was plotted in R v.4.2.0 with ggplot2 v.3.3.6.

Quantitative real-time PCR analysis to determine bacterial load

Real-time PCR was performed on a Quant Studio 7 Flex Real-time PCR system (Thermo Fisher Scientific) using primers that targeted conserved regions of the 16S rRNA gene as described in ref. ⁶⁷. Each reaction comprised 5 μl of SYBR Green qPCR (Thermo Fisher Scientific), 0.2 μM of each forward and reverse primer, 1 μl of DNA template and sterile water to a total reaction volume of 22 μl. A no-template DNA control, with the DNA substituted with 1 μl sterile water, was included in each run. Real-time PCR reactions for each sample were performed in duplicate, consisting of 10 μl per reaction. The amplification program was 50 °C for 2 min, 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. Melting curve analysis was performed at the end of the program. The total bacterial load in each sample was quantified against a standard curve of serial dilutions of E. coli genomic DNA, which were performed in the same run, and was detectable within a range of 3 to 3 × 10⁶-equivalent E. coli cells. Numbers of cells were then calculated against the faecal weight used for each sample. A Mann–Whitney test was used to assess statistical significance.

Assessment of antibody responses in mice

Levels of antigen-specific IgG_total IgG1, IgG2c, IgG3 and IgM were measured by enzyme-linked immunosorbent assay (ELISA) in serum collected from each vaccine experiment. Briefly, Nunc MaxiSorp ELISA plates (Sigma) were coated either with CRM₁₉₇ (0.5 μg ml⁻¹, Finabio), PCV13 vaccine (0.05 μg ml⁻¹, Pfizer), pneumococcal polysaccharide (4 μg ml⁻¹ PPS1, 161-X, ATCC), PPS3 (4 μg ml⁻¹ 169-X, ATCC), PPS6B (4 μg ml⁻¹ 105-X, ATCC), PPS9V (4 μg ml⁻¹ 305-X, ATCC) or diphtheria toxoid (1 μg ml⁻¹, List Biological Laboratories), tetanus toxoid (1 μg ml⁻¹, List Biological Laboratories), pertussis toxoid (1 μg ml⁻¹, List Biological Laboratories) or Bexsero vaccine (2 μg ml⁻¹, GSK). Plates were washed with 0.5% Tween-20/PBS and blocked with 1× ELISA buffer (eBioscience). Mouse serum samples were diluted in 1× ELISA buffer and incubated on blocked plates. Antigen-specific serum antibodies were detected using horseradish peroxidase (HRP)-conjugated antibodies: anti-mouse IgG (Novex) or anti-mouse IgG1 (Abcam) at a dilution of 1:1,000 or IgG2c, IgG3 or IgM (Abcam) at 1:500. HRP activity was detected using tetramethylbenzidine substrate (Thermo Fisher Scientific) and stopped with 2 N H₂SO₄. Developed plates were recorded using a Synergy HTX microplate absorbance reader (BioTek) at 450 nm with correction at 595 nm by subtraction. Data were analysed using Prism 10 (GraphPad).

Tissue dissociation and flow cytometry analysis

Single-cell suspensions of murine spleen or lymph nodes were prepared by mechanical dissociation and filtration through a 70-µm nylon filter (Merck Millipore). Cells were stained with Fixable Viability Stain 780 (BD Biosciences) and murine Fc-blocking antibody clone 2.4G2 (1:100, BD Biosciences) in PBS for 15 min. Cells were washed and then stained with fluorochrome-conjugated antibodies in FACS buffer (PBS, 0.1% bovine serum albumin, 2 mM EDTA) against CD3 (142-C11, 1:300, BD Biosciences), CD4 (RM4, 1:300, BD Biosciences), CD8 (53-6.7, 1:600, BD Biosciences), CD19 (1D3, 1:300, BD Biosciences), CD38 (90, 1:600, BioLegend), CD44 (IM7, 1:600, BD Biosciences), CD62L (MEL-14, 1:600, BD Biosciences), CD95 (J02, 1:300,BD Biosciences), CD138 (281-2, 1:600, BD Biosciences), B220 (RA3-6B2, 1:600, BD Biosciences), CXCR5 (L138D7, 1:200, BioLegend), GL7 (GL7, 1:600, BD Biosciences), IgD (11-26 c.2a, 1:300, BioLegend), IgM (R6-60.2, 1:300, BD Biosciences), PD-1 (RMP1-30, 1:300, BD Biosciences) and streptavidin-PECF549 (1:400, BioLegend).

For assessment of CRM⁺ GC B cells, cells were stained with Fixable Viability Stain 780 (1:1000, BD) and murine Fc block (1:100, BD) in PBS for 15 min; then, they were washed and incubated with biotinylated CRM (1:100, Fina BioSolutions) in FACS buffer. Surface antibody cocktails were added directly after staining for 20 min at 4 °C. Data were acquired on a BD FACS Symphony and analysed using FlowJo v.10. For assessment of CD86, CD80 and MCHII expression, single-cell suspensions of spleen or lymph nodes were stained with Zombie Aqua Fixable Viability Stain 780 (1:1000, BioLegend) and murine Fc-blocking antibody clone 2.4G2 (1:100, BD Biosciences) in PBS for 15 min. Cells were washed and then stained with fluorochrome-conjugated antibodies in FACS buffer (PBS, 0.1% bovine serum albumin, 2 mM EDTA) against CD19 (ID3, 1:200, BD), CD3 (145-2C11, 1:150, BD Biosciences), CD8 (53-6.7, 1:200, BD), NK1.1 (PK136, 1:200, BD), Ly6G (1A8, 1:600, BD), B220 (RA3-6B2, 1:3000, BD), CD11b (M1/70, 1:500, BD), CD11c (HL3, 1:400, BD Biosciences), MHCII (M5/114.15.2, 1:150, Miltenyi), CD86 (PO3.3, 1:150, Miltenyi) and CD80 (16-10A1, 1:150, Tonbo Biosciences).

Immunofluorescence staining and imaging

Mediastinal lymph nodes were fixed in 1 ml 4% paraformaldehyde (Sigma) for 4 h and then incubated in 1 ml 30% sucrose solution overnight at 4 °C before being embedded in OCT (Tissue Tek) as described previously⁶⁸. Slides were blocked in blocking buffer (normal goat serum 4% v/v (Southern Cross Science), normal donkey serum 4% v/v (Stratech Scientific APAC) and bovine serum albumin 15% v/v (AusGeneX)) for 2 h at room temperature. Slides were permeabilized with Triton X-100 2% v/v (Sigma) and stained with CD16/32 (1:100, BD Biosciences) and CD21/35 clone 7E9 (1:200), GL7 (1:200), CD3 clone 17A2 (1:200) and IgD clone 11-26c.2a (1:200) (all from BioLegend). Images were acquired using a Leica TCS SP8X confocal microscope and analysed using Fiji⁶⁹ in ImageJ.

Statistical analysis

The primary objective of the AIR study was to determine whether antibody responses to the PCV13 vaccine at 7 months of age were different between antibiotic-exposed and unexposed groups, specifically comparing the following:

• 1. The percentage of participants achieving a seroprotective antibody response (greater than or equal to 0.35 μg ml⁻¹) against pneumococcal polysaccharide serotypes in the PCV13 vaccine at approximately 7 months of age.

• 2. Geometric mean concentrations in μg ml⁻¹ of anti-PPS IgG as measured by serum assay at approximately 7 months of age.

As secondary outcomes, we compared PCV13 vaccine responses at 15 months of age and antibody responses to the Infanrix Hexa 6-in-1 vaccine (6-in-1, DTPa-HepB-IPV-Hib) and the Rotarix oral rotavirus vaccine in antibody-exposed and unexposed infants at both time points. Multivariable linear regression adjusting for baseline prevaccination titres and sex was used to test for any association between antibiotic exposure group and IgG geometric mean concentrations for each of the vaccine antigens assessed. Further longitudinal analyses were conducted by fitting a mixed effects multivariable linear regression, including a random effect for participant ID, to assess the interaction between time point and antibiotic exposure group on IgG geometric mean concentrations for each of the vaccine antigens. Multivariable logistic regression was used to test for any association between antibiotic exposure group and the proportion of infants achieving a seroprotective response. Statistical significance was defined as P < 0.05. Further sensitivity analyses were conducted to assess the robustness of associations to adjustment for probiotic consumption, formula consumption, baseline IgG/IgA concentration, infant sex and gestation at delivery. The AIR study also included a range of exploratory systems immunology assessments for which analyses were undertaken in line with best practices for each data type and as described in the sections above. Measurements were taken from distinct samples. All statistical tests throughout were two-sided; the specific statistical analyses used are specified in the figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.