6-week old female C57BL/6 mice (Australian BioResources, NSW, Australia) were housed under specific-pathogen-free conditions in the animal facility of Charles Perkins Centre, The University of Sydney. Mice were given ad libitum access to standardised rodent diet AIN-93G (Specialty Feeds, Glen Forrest, Australia) and drinking water. Mice were weighed twice weekly and monitored for humane endpoints. Humane endpoints for this study included greater than 15% loss of bodyweight, as well as observed changes in behaviour such as hunching, ruffled fur, and inactivity persisting over 24 hours. Loose stool consistency over a 24-hour period was also used as a parameter to assess animal wellbeing. All experiments were performed with approval from the University of Sydney Animal Ethics Committee (2023/2308).

To deplete gram-positive or gram-negative gut bacteria, mice were gavaged with either Vancomycin (5mg/mL; targeting gram-positives) or Polymyxin B (7.5mg/mL; targeting gram-negatives) dissolved in PBS, based on previous studies (24, 25) (n=10 mice per group). Mice were also gavaged with the anti-fungal Amphotericin B (1mg/mL in PBS) (Sigma-Aldrich) for 1 week prior to, and throughout the antibiotic regimen. Mice were gavaged twice daily for all antibiotic treatments, in a maximum volume of 200uL per gavage. Control mice (PBS group) were administered PBS for all gavages.

Caecal content was weighed and homogenised in deuterium oxide (Sigma-Aldrich) at a concentration of 100mg/mL. The sample was centrifuged 5 mins at 14 000 xg at 4°C, caecal pellet was retained for DNA extraction. The supernatant was filtered through 3kDa centrifugal filtration unit (Merck Millipore) and added to chloroform-D/methanol-D (Sigma Aldrich). The upper aqueous phase was diluted into sodium triphosphate buffer with 4,4-dimethyl-4-silapentane-1-sulfomnic acid spiked in as an internal standard. Samples were run on a Bruker 600 MHz spectrometer and data was analysed with Chenomx Profiler software (V10.0).

DNA from pelleted caecal content was extracted using the FastDNA SPIN Kit for Feces (MP Biomedicals) according to manufacturer’s instructions, and quantified with the Nanodrop for normalisation to caecal weight. The V3–V4 region of the 16S rRNA gene was amplified with Q5^® High-Fidelity 2X Master Mix (New England Biolabs) using the Pro341F (5′- CCTACGGGNBGCASCAG -3′) and Pro805R (5′- GACTACNVGGGTATCTAATCC -3′) primers as previously described (26). Following amplification, library cleanup was performed using AMPure XP reagent (Beckman Coulter). Amplicon sequencing was done on the Illumina MiSeq system (2x250 bp PE). Data was analysed using the dada2 package (1.30.0) using R software (4.3.1) to generate Amplicon sequence variants (ASV). ASV were classified according to taxonomy using the Ribosomal Database Project classifier (27). Analysis was performed with the phyloseq (1.46.0), vegan (2.6-4) and microbiome (1.24.0) R packages and differentially abundant taxa were analysed using ALDEx2 (1.34.0). Bacterial load was assessed by absolute 16S copy number per gram of caecal content, as described previously (28). Copy number was determined by qPCR with the 16S rDNA universal primers, UniF: GTGSTGCAYGGYYGTCGTCA and UniR: ACGTCRTCCMCNCCTTCCTC, and compared with a standard curve (10² to 10⁸ copies per uL). Fungal load was determined using primers for the Fungal ITS1 rRNA region (ITS1-30F: GTCCCTGCCCTTTGTACACA, ITS1-217R: TTTCGCTGCGTTCTTCATCG) (29) and the 18S region (F: GGRAAACTCACCAGGTCCAG, R: GSWCTATCCCCAKCACGA) (30), following the relevant cycling conditions.

Small intestinal content was homogenised at 100mg/mL in PBS with complete Protease Inhibitor cocktail (Roche). Samples were centrifuged 14 000x g for 10 min at 4°C and supernatant collected. Immunoglobulin A (IgA) was assessed with the Mouse IgA Quantitation Set (Bethyl Laboratories) following the manufacturer’s instructions.

Mice were euthanised via CO₂ and mesenteric lymph nodes (MLNs) and spleens were collected and processed into single cell suspension. Tissues were disrupted in FACS buffer (2% FCS, 0.5M EDTA in PBS). Red blood cells were lysed for spleen samples using RBC Lysis Buffer 10X (Biolegend). Cell suspensions were passed through a 100µm mesh filter and resuspended in FACS buffer for counting with trypan blue (0.4%) on a haemocytometer to exclude dead cells.

Cells were stained for viability using LIVE/DEAD™ Fixable Blue Dead Cell Stain (ThermoFisher Scientific), and non-specific binding was prevented using anti-mouse CD16/32 (BioLegend). Surface antigen and intracellular staining was performed using antibodies listed in Table 1 . For intracellular staining of cytokines and FoxP3, cells were fixed and permeabilised using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience) according to the manufacturer’s instructions. Cells were washed and then acquired on the Aurora (Cytek) with SpectroFlo software (Cytek) and data was analysed in FlowJo (Treestar Inc. Ashland, OR, USA), based on the gating strategies in Supplementary Figures 4–9 .

Splenocytes (10⁶ cells) were stimulated with either lipopolysaccharide (LPS) (100ng/mL) (Invivogen) for 48 hours, or with phorbol 12-myristate 13-acetate (PMA) (50ng/mL) (Sigma Aldrich), ionomycin (500ng/mL) (Sigma Aldrich), and brefeldin A (BFA) (5µg/mL) (Sigma Aldrich) for 4 hours. Cells were incubated at 37°C 5% CO2, then washed and stained for intracellular cytokines as described above.

Global antibiotic consumption data was sourced from the Global Research on Antimicrobial Resistance (GRAM) database (31). Antibiotic usage was documented as defined daily doses per 1000 population per day (DDD/1000 inhabitants/day). Age-standardised incidence rates of inflammatory bowel disease (IBD) was collated from Global Burden of Disease (GBD) Study 2021 (https://vizhub.healthdata.org/gbd-results/). Global gross domestic product (GDP) data was obtained from the Maddison project (32). Countries or timepoints without available data were excluded from analysis, and the final dataset included data from close to 200 countries, spanning the years 2000 to 2018.

Details of analysis were adapted from our previous works (33, 34). Generalised additive model (GAM) is a statistical tool facilitating the simultaneous analyses of multiple parameters, including their plausible interactions and non-linear effects (35–37). They are modelled using non-parametric smoothed functions, which are generally in forms of splines and represent flexible manners to examine non-linear associations. In this study, GAMs were used to model age-standardised IBD incidence rates, using antibiotic consumption, GDP and time as predictors with the “gam” function from the mgcv package (37, 38). When modelling, countries that the data were based on were corrected as random effects. An array of GAMs were run using aforementioned predictors as well as their different combinations, capturing their individual, additive and interactive effects. Effects from each parameters were modelled as smooth terms with the “s()” function, and their interactions were modelled with the “ti()” or “te()” functions, considering their different scales, such as antibiotic consumption and GDP. Modelling results were evaluated following the Akaike information criterion (AIC) (39), and the one with the lowest AIC was chosen.

Comparisons between groups were performed with GraphPad Prism (v10) with ordinary one ANOVA, repeated measures two-way ANOVA, or a mixed-effects model with Geisser-Greenhouse correction, followed by Tukey’s multiple comparisons test. Differences were considered statistically significant when p<0.05, with *p<0.05, **p<0.01, ***p<0.001, **** p<.0001

In order to understand the broad impact of G- or G+ bacteria on immune homeostasis we used antibiotics to selectively eliminate G+ or G- taxa within the gut microbiota towards a dominantly G+ or G- composition. We utilised Polymyxin B (PmB), a cationic peptide that binds to LPS to weaken the bacterial outer membrane to target G- species and skew the microbiota towards a G+ composition (40). To shift the microbiota towards a G- composition we used vancomycin, which is active against most enteric G+ species. Before beginning the antibiotic regimen, mice were administered the antifungal amphotericin B (1mg/mL) for 1 week, to avoid expansion of enteric fungi that often occurs with antibiotic treatment. Mice were then maintained on amphotericin B and gavaged either vancomycin (5mg/mL), PmB (7.5mg/mL) or PBS twice daily for 4 weeks ( Figure 1A ), to generate microbiota that was G- or G+ dominant, respectively. All three drugs are reported to have poor bioavailability after oral administration, with little systemic absorption (41–47). Therefore, we expect minimal absorption into peripheral circulation, and the drugs’ primary activity to be confined to the gastrointestinal tract. We selected to administer antibiotic treatments via oral gavage to allow for a standardised dose to be delivered and avoid dehydration and severe weight loss often observed if antibiotics are given in drinking water (25, 48). We observed, however, that vancomycin-treated mice had significant increase in percentage weight gain compared to PmB treated mice, and PmB mice tended to have reduced weight compared to PBS mice towards the end of the experiment ( Figure 1B ). Vancomycin has been linked to weight gain in humans (49) and was found to promote food intake and weight gain in mice (50), which may explain this discrepancy. Caecal weight was also significantly increased by vancomycin treatment ( Supplementary Figure 1A ), as previously described (50, 51), likely due to elimination of fibre-digesting bacteria. While histopathological analysis was beyond the scope of this study, previous studies suggest vancomycin does not induce gross changes to the colonic epithelium (52). However, short-term oral vancomycin was found to disrupt the colonic mucus barrier by inducing ER stress in goblet cells that impaired mucus secretion in goblet cells (53), and both vancomycin and PmB may impair colonic permeability (54).

To confirm our treatments were successful in manipulating gut microbiota composition, we extracted caecal DNA and performed 16S rRNA sequencing. Principal component analysis based on Aitchison distance revealed that all three groups had a distinct microbiota composition ( Figure 1C ). Measures of alpha diversity, which evaluate the species richness and evenness within the gut microbiota, were also examined ( Figure 1D ). Mice treated with vancomycin had a significantly reduced richness compared to PBS and PmB-treated mice, as shown by the decreased number of unique amplicon sequence variants (ASV). Observed ASV was unchanged in PmB-treated mice compared to PBS, indicating this treatment did not significantly reduce the number of unique taxa. Other measures of alpha diversity, including Evenness and Shannon and Inverse Simpson indices, were also significantly reduced in antibiotics-treated groups compared to PBS ( Figure 1D ), however, vancomycin-treated mice exhibited a more pronounced reduction across these measures and were significantly lower compared to PmB-treated mice ( Figure 1D ).

Next, to confirm the efficacy of antibiotic treatment, we examined the relative abundance of phyla in the gut microbiota and classified these as G- or G+ ( Figure 1E ). Basally, the composition of PBS microbiota was mostly G+. As expected, vancomycin treatment led to predominantly G- phyla and PmB resulted in predominantly G+ phyla. Surprisingly, PmB-treated mice had a slightly reduced abundance of G+ phyla compared to PBS. Examination of the G- phyla at the genus level for PmB mice revealed that G- species comprised Parabacteroides and Bacteroides spp ( Figure 1F ). PmB resistance in commensal Bacteroides and Parabacteroides has been reported previously, and has been attributed to enzyme modifications to the lipid A component of LPS, which disrupts PmB binding (55–57). Remaining G- genera, Akkermansia and Escherichia/Shigella, were present at relative abundance of <1% ( Supplementary Figures 1B, C ), indicating that susceptible G- species were successfully eliminated by PmB treatment. Altogether, this demonstrates that our treatments were successful in eliminating the majority of susceptible G+ or G- species within the gut microbiota.

To gain further insight into the gut microbiota composition, we examined the relative abundance of Phyla ( Figure 1G ) and Genus (top 10 genera) ( Figure 1H ). The mostly G+ composition of PBS microbiota was attributed to the presence of Firmicutes, primarily represented by the genus Acetatifactor ( Figures 1G, H ), which is strongly associated with the rodent gut (58, 59). As described earlier, PmB led to expansion of G- phyla Bacteroidetes (comprising genera Bacteroides and Parabacteroides) and G+ Actinobacteria (comprising the genus Olsenella). Vancomycin resulted in a microbiota primarily composed of G- Proteobacteria phyla, which was due to expansion of the genus Escherichia/Shigella, which likely represents the proliferation of a commensal or facultatively pathogenic species (pathobiont) within the existing gut microbiota. Altogether these data further confirm that our treatments effectively targeted susceptible bacteria to significantly perturb gut microbiota composition and induce a shift to a G+ or G- dominant composition.

We also measured 16S copy number, as a readout of bacterial load in the gut, which is an influential factor on host immunity ( Supplementary Figure 1D ). We found antibiotics decreased bacterial load compared to PBS mice, and PmB treated mice had significantly reduced load compared to vancomycin treated mice. Thus, the selective antibiotic treatments effectively targeted their respective species and reduced overall bacterial burden in the gut. We also assessed fungal load within the microbiota, which, despite comprising only 1% of gut microbiota, play a critical role in systemic immune development and responses (60, 61). We observed that vancomycin-treated mice had increased fungal load compared to PBS, as quantified by the relative expression of 18S rRNA, however, there were no significant differences between groups for ITS rRNA expression ( Supplementary Figure 1E ). Altogether, these findings reveal the differential impact of select antibiotic treatments on the bacterial and fungal compartments of the gut microbiota.

Gut microbiota-derived metabolites, particularly SCFAs, profoundly influence host physiology and immunity. To assess how antibiotic-mediated changes in microbiota composition impact the production of these metabolites, we quantified the levels of caecal SCFAs by nuclear magnetic resonance spectroscopy. We examined levels of the major SCFAs, acetate, propionate, and butyrate, as these play a key role in promoting tolerance (11, 62). Moreover, the majority of SCFA-generating bacteria belong to the predominantly G+ phyla Firmicutes (63), which was decreased in abundance in antibiotic-treated groups. Accordingly, these mice had decreased total SCFA levels, with vancomycin-treated mice having the most altered ratios of SCFAs ( Figures 2A, B ). Compared to PBS, acetate was significantly decreased in PmB-treated mice, and close to undetectable in vancomycin-treated mice ( Figure 2B ). Surprisingly, a proportion of vancomycin mice had significantly elevated propionate levels, whereas propionate was undetectable in the PBS and PmB mice ( Figure 2B ). Butyrate was reduced in both antibiotic-treated groups to a similar extent ( Figure 2B ). Finally, succinate, another metabolite that has significant associations with dysbiosis as well as intestinal inflammation and stress (64, 65), was significantly elevated in PmB mice ( Figure 2C ) compared to other groups.

Next, we measured intestinal secretory Immunoglobulin A (sIgA), a key component of the host’s luminal response to the gut microbiota, which has also been linked to high levels of acetate (66). One function of sIgA is binding to luminal bacteria to maintain exclusion of the microbiota from the host epithelium. sIgA was substantially elevated in vancomycin mice, approximately two-fold compared to other groups ( Figure 2D ) suggesting a potential expansion of colitogenic bacteria (67).

Finally, we performed a correlation analysis between bacterial-derived metabolites, host sIgA, and the composition of the gut microbiota ( Figure 2E ). We identified a positive correlation with acetate and butyrate with some G- genera (Desulfovibrio, Alistipes, Oscillibacter) and many G+ genera (including Bifidobacterium, Sporobacter, Acetatifactor, Anaerosporobacter, and Dorea). Similarly, succinate levels correlated with both G- taxa (Bacteroides and Parabacteroides) and G+ phyla Firmicutes (including Clostridia spp). Although propionate production is commonly correlated with G- Bacteroides in human microbiota (68), we observed propionate correlated highly with G- Escherichia/Shigella as well as G+ Lactococcus and Lactobacillus spp, all of which have the capacity to generate propionate (69, 70). Similarly, sIgA correlated strongly with Escherichia/Shigella, Lactococcus and Lactobacillus spp, in line with a previous study in humans treated with vancomycin (71). Together these findings suggest that perturbing microbiota composition can dramatically alter gut microbiota-derived metabolites, some of which may be attributed to the large reduction in Firmicutes phyla we observed.

As the mesenteric lymph nodes (MLNs) drain the gut and are the critical sites of antigen-specific response against gut bacteria, we first assessed impact of antibiotic-mediated changes in gut bacteria on this site. To identify how these changes may influence the local immune system in the gut, we quantified proportions and absolute numbers of innate and adaptive immune cells in the MLNs (gating strategy shown in Supplementary Figure 4 ).

Total MLN cellularity was similar among groups ( Figure 3A ), suggesting changes to gut microbiota composition did not induce a localised immune response in this site. We observed that compared to PBS control, a antibiotics-perturbed microbiota impacted the proportions of a range of innate and adaptive immune cell subsets ( Supplementary Figures 2A, B ). Both antibiotics treated groups had increased proportions of migratory conventional DCs (MHCII^hi cDCs), and PmB mice had increased neutrophils compared to vancomycin and PBS mice. Vancomycin treatment increased proportions of plasmacytoid DCs (pDCs), resident cDCs, Natural Killer (NK) cells, Ly6C⁺ inflammatory monocytes compared to the other groups, and led to decreases in TCRγδ⁺ T cells compared to PBS and decreased CD4⁺ T cells compared to PmB mice( Supplementary Figures 2A, B ). For absolute cell number, we found mainly innate cells were influenced by antibiotic treatment ( Figure 3B , Supplementary Figure 2C ). PmB treated mice had increased neutrophils compared to PBS and vancomycin mice, while vancomycin mice had increased NK cells and Ly6C⁺ monocytes compared to the other groups ( Figure 3B ). We found both antibiotic-treated groups had significantly increased numbers of migratory DCs compared to PBS, while vancomycin mice had increased numbers of resident cDCs and pDCs compared to PmB and PBS mice ( Figure 3B ). DCs are key antigen-presenting cells that are crucial for the development of antigen-specific responses in the gut and in the induction of tolerance, and cDCs and pDCs differ in their ontogeny as well as their functional role, with cDCs being highly efficient in antigen-presentation and activation of naïve T-cells whereas pDCs play a key role in production of type I interferons and the anti-viral response (72). Altogether our findings suggest that skewing the microbiota towards a dysbiotic composition mainly impacts innate cell numbers, primarily DCs.

Migratory and resident cDCs can be further classified into cDC1 (CD8a⁺CD11b^-) and cDC2 (CD8a^-CD11b⁺) subsets, which are also functionally specialised: cDC1 are key in the cytotoxic immune response and presentation to CD8⁺ T cells, and cDC2s are key in helper T cell responses and presentation to CD4⁺ T cells. Both migratory cDC subsets were increased in proportion and number for antibiotics-treated mice ( Figure 3C , Supplementary Figure 2D ). For vancomycin mice, both resident cDC1 and cDC2 subsets were also increased in number compared to other groups ( Figure 3C ). The fold-change for increased migratory cDC numbers was approximately 3 times greater for both antibiotics-treated groups compared to the PBS group, while the increase in resident cDC in vancomycin mice was twice that of PBS and PmB groups ( Figure 3D ). Altogether, these data indicate that a dramatically altered microbiota composition impacts innate cells in the gut-draining lymph nodes, and a PmB- or vancomycin-skewed microbiota appear to differentially impact neutrophils and pDCs, but both increasing migratory cDC numbers.

As we observed DC numbers were highly impacted by a skewed microbiota, we next examined the impact on T cells. Neither the proportion nor the number of total T cells, CD4⁺ T cell or CD8⁺ T cell subsets were altered in mice with perturbed microbiota compared to PBS ( Supplementary Figures 2B, C ), suggesting there was no initiation of an antigen-specific immune response, consistent with the unchanged MLN cellularity. This further confirms that antibiotic treatments perturbing the microbiota mainly impact innate, and not adaptive cell subsets, in the MLNs.

Altogether these results suggest that an antibiotics-perturbed microbiota increase recruitment of DC to the MLNs, but without initiation of an antigen-induced T cell response.

To determine if a dramatically skewed microbiota had a systemic impact on immunity, we next investigated proportions and numbers of innate and adaptive cells among splenocytes, as well as their cytokine secretion profile.

Similar to the MLN we found no changes in overall splenic cellularity ( Supplementary Figure 3A ). However, mice with perturbed microbiota had altered proportions of immune subsets including eosinophils, cDCs, putative red pulp macrophages (F480⁺Ly6C^- cells) and Ly6C^med monocytes, as well as T cell subsets ( Supplementary Figures 3B, C ; gating strategy shown in Supplementary Figure 5 ), suggesting the systemic impact of the microbiota shift. As in the MLNs, DC subset numbers were primarily impacted, with a significant reduction of total cDCs as well as cDC1 and cDC2 subsets in PmB-treated mice compared to other groups ( Figures 4A, B ). Contrary to the MLNs, we found no change to pDC numbers.

To further investigate the impact of antibiotic-mediated dysbiosis on immune function, we quantified the cytokine secretion before and after challenge in vitro. To do this, we stimulated splenocytes with LPS from E. coli for 48 hours and examined the intracellular production of pro-inflammatory cytokines IL-6 and TNF by flow cytometry ( Figures 4C–E ; gating strategy shown in Supplementary Figure 6 ). As expected, LPS treatment in pDC from PBS-treated mice led to increased production of IL-6 but this was to the same extent as pDC from vancomycin mice ( Figure 4C ). Interestingly, there was no effect of LPS on pDCs from PmB mice ( Figure 4C ). This differential effect of the gut microbiota on IL-6 was not observed for TNF, as pDCs across groups produced comparable levels of TNF upon LPS treatment ( Figure 4C ). LPS stimulation of cDCs induced robust IL-6 expression for all groups, however, cDCs from vancomycin mice showed reduced expression compared to PBS and PmB mice ( Figure 4D ). Although we did not measure circulating or faecal LPS levels in our mice, which may reflect differences in LPS tolerance, the fact that LPS is the primary antigen of the G- membrane suggests that reduced LPS levels in PmB mice may affect the DC response to this stimulus. This may be supported by the fact that Bacteroides is the only G- phyla present in PmB-treated microbiota, and Bacteroides-derived LPS is well-known to be less immunogenic than E. coli-derived LPS (73). Although PmB is known for its endotoxin-binding capability, the low likelihood of systemically circulating PmB suggests that the gut microbiota remains a critical source of antigens that shapes immune responses in distal organs. LPS stimulation did not induce TNF expression in cDCs, and there was in fact a slight reduction in TNF levels in PmB mice ( Figure 4D ). Overall, these results indicate that a PmB treated microbiota primarily impacts cDC activation by LPS, possibly reducing cDC capacity to respond to stimuli from immunogenic G- bacteria.

To further examine the impact of antibiotic-mediated dysbiosis on immune cells, we measured the expression of the cytokines IL-17, IFN-γ IL-10 and TNF, in splenocytes after PMA/ionomycin stimulation ( Figure 5 ; gating strategy and FMOs shown in Supplementary Figures 7-9 ). Many studies have identified G+ and G- bacteria induce differing cytokine immune profiles in vitro (74–77), as well as during sepsis (78), highlighting the capacity of the immune system to discriminate between the two classes of bacteria. We first examined levels of IL-17 and IL-10, as these cytokines are highly influenced by gut microbiota composition (1, p. 20). The development of IL17-producing cells TCRγδ⁺ T cells in the intestines, as well as in the liver and lungs, is regulated by the gut microbiota (79, p. 201; 3, 80). We observed antibiotics-treated mice showed no changes in IL-17⁺ cells for CD4⁺ T cells. TCRγδ⁺ and CD8⁺ T cells ( Figure 5A ). The induction of intestinal Th17 cells is highly dependent on a subset of G+ bacteria known as segmented filamentous bacteria (3), thus while we did not observe changes in IL-17 production in the spleen, measuring production of IL-17 in gut-local T cells may provide more relevant insights into the influence of antibiotics-perturbed microbiota. The anti-inflammatory cytokine IL-10 is primarily derived from Tregs that are induced by tolerogenic DCs. The generation of tolerogenic DCs is also strongly associated with G+ species, highly prevalent in PmB treated mice (81, 82). In line with this, we found PmB mice had significantly elevated IL-10 expression across T-cell subsets, including Tregs, non-Treg CD4⁺ cells, as well as CD8⁺ T cells ( Figure 5B ). Vancomycin mice also had elevated proportions of non-Treg IL10⁺ CD4⁺ T cells and tended to have elevated IL10⁺ CD8⁺ T cells (p=0.0603) compared to PBS mice ( Figure 5B ).

Finally, we examined the expression of IFN-γ and TNF, pleiotropic cytokines that are instrumental in regulating immune response and also associated with autoimmune diseases. IFN-γ, a type II interferon, is thought to contribute to the pathological intestinal inflammation seen in irritable bowel disease (IBD) (83, 84) and TNF is widely recognised for its involvement across autoimmune conditions such as IBD, arthritis, psoriasis, and systemic lupus erythematosus (85). Previous studies have also identified increases in TNF mRNA in colon tissue after PmB and vancomycin treatment (54, 86). While these studies differed from ours in antibiotic administration methods and resulting microbiota compositional changes, they are suggestive of a link between antibiotics exposure and a pro-inflammatory response. We observed that compared to PBS, PmB mice showed IFN-γ production across a range of subsets including NK cells, TCRγδ ⁺ T cells, CD4⁺ and CD8⁺ T cells, as well as B cells ( Figure 5C ). IFN-γ⁺ cells also tended to be increased compared to vancomycin mice for NK, TCRγδ⁺,and CD8⁺ T cell subsets. Vancomycin mice exhibited IFN-γ CD8⁺ T cells only, compared to PBS ( Figure 5C ). Strikingly, we observed that an antibiotics-skewed microbiota, resulted in a significantly increased TNF expression across a variety of cell types, including NK cells, TCRγδ⁺ T cells, Tregs, CD4⁺ T-cells and CD8⁺ T-cells ( Figure 5D ). This increase was most prominent for TCRγδ⁺ and CD4 subsets, where the average % of TNF positive cells was more than double the PBS group ( Figure 5D ). Altogether these data show that while PmB may promote IL-10 expression, contributing to a protective response, both PmB and vancomycin-treated microbiota significantly upregulate TNF expression, which may potentially exacerbate inflammation.

Given antibiotic-treated mice showed significant upregulation of TNF and IFN-γ, critical mediators of IBD, this prompted us to examine the link between antibiotic consumption and global burden of IBD. Previous studies have identified associations between antibiotic exposure and development of chronic inflammatory diseases, such as metabolic disease, allergies, and autoimmune conditions, including colitis (23). Therefore, we gathered antibiotic usage data from the Global Research on Antimicrobial Resistance database, as well as age-standardised IBD incidence rates from Global Burden of Disease 2021 (GBD2021) and analysed their relationship using a generalised additive model (GAM). GAMs allow for comprehensive analysis of multiple parameters, simultaneously covering their individual, interactive effects, including non-linear effects. Our GAM also adjusted for confounders such as socioeconomics, reflected by gross domestic product (GDP), and corrected for inter-country differences as random effects (see Supplementary Information in Supplementary Data Sheet 1 ).

Figure 6 presents the modelling results for 2018, the most recent timepoint with complete data coverage, illustrating the predicted response curve of IBD incidence rate as a function of antibiotic consumption (Defined Daily Dose per 1000 inhabitants) at the median-grouped GDP. Globally, increased antibiotic consumption correlated with higher IBD disease burden. However, this relationship was also influenced by GDP, as the association dissipates at a high GDP bracket ( Supplementary Figure 10 ). Here, our global analysis aligns with previous observational studies, which have demonstrated an association with antibiotics exposure and IBD incidence rates (87–89).

Overall, these data suggest a perturbed microbiota primes host cells for heightened pro-inflammatory expression under activation conditions. A PmB-treated microbiota in particular may skew the host towards a Th1-biased phenotype, in line with data that found G+ bacteria preferentially induce IFN-γ and TNF in human PBMCs (90).