Parenteral BCG vaccine induces lung-resident memory macrophages and trained immunity via the gut–lung axis

Aside from centrally induced trained immunity in the bone marrow (BM) and peripheral blood by parenteral vaccination or infection, evidence indicates that mucosal-resident innate immune memory can develop via a local inflammatory pathway following mucosal exposure. However, whether mucosal-resident innate memory results from integrating distally generated immunological signals following parenteral vaccination/infection is unclear. Here we show that subcutaneous Bacillus Calmette–Guérin (BCG) vaccination can induce memory alveolar macrophages (AMs) and trained immunity in the lung. Although parenteral BCG vaccination trains BM progenitors and circulating monocytes, induction of memory AMs is independent of circulating monocytes. Rather, parenteral BCG vaccination, via mycobacterial dissemination, causes a time-dependent alteration in the intestinal microbiome, barrier function and microbial metabolites, and subsequent changes in circulating and lung metabolites, leading to the induction of memory macrophages and trained immunity in the lung. These data identify an intestinal microbiota-mediated pathway for innate immune memory development at distal mucosal tissues and have implications for the development of next-generation vaccine strategies against respiratory pathogens.

Aside from centrally induced trained immunity in the bone marrow (BM) and peripheral blood by parenteral vaccination or infection, evidence indicates that mucosal-resident innate immune memory can develop via a local inflammatory pathway following mucosal exposure. However, whether mucosal-resident innate memory results from integrating distally generated immunological signals following parenteral vaccination/infection is unclear. Here we show that subcutaneous Bacillus Calmette-Guérin (BCG) vaccination can induce memory alveolar macrophages (AMs) and trained immunity in the lung. Although parenteral BCG vaccination trains BM progenitors and circulating monocytes, induction of memory AMs is independent of circulating monocytes. Rather, parenteral BCG vaccination, via mycobacterial dissemination, causes a time-dependent alteration in the intestinal microbiome, barrier function and microbial metabolites, and subsequent changes in circulating and lung metabolites, leading to the induction of memory macrophages and trained immunity in the lung. These data identify an intestinal microbiota-mediated pathway for innate immune memory development at distal mucosal tissues and have implications for the development of next-generation vaccine strategies against respiratory pathogens.
There is growing recognition of the importance of innate immune memory and trained innate immunity (TII) in host defense and vaccinology [1][2][3] . Epidemiological studies have shown that parenteral immunization with live attenuated vaccines (including Bacillus Calmette-Guérin (BCG)) offers protection against both the target and unrelated pathogens 4 . Thus, antituberculosis (TB) BCG vaccination reduces all-cause mortality and/or respiratory infections in young children and the elderly [5][6][7] . Such systemic/parenteral microbial-, inflammation-or vaccine-induced TII is mediated primarily through centrally trained circulating leukocytes including monocytes resulting from metabolic/epigenetic rewiring of myeloid progenitors in the bone marrow (BM) 8-13 . Article https://doi.org/10.1038/s41590-022-01354-4 Eleven-analyte Luminex analysis of secreted cytokines/ chemokines shows that at baseline (US), levels of cytokines/chemokines were low in control and BCG AM cultures (Fig. 1f-l), whereas upon stimulation, compared to control AMs, BCG AMs produced significantly higher levels of IL-6, IL-12p40, MCP-1, MIP-1α and RANTES (Fig. 1g,h,j-l) with no difference observed in TNF production (Fig. 1i).
We next analyzed the metabolic state of airway AM. BCG AM demonstrated significantly increased glycolysis compared to their mildly changed rates of oxidative phosphorylation (Fig. 1m,n and Extended Data Fig. 1d). Because local viral infection-trained AM underwent low-rate in situ proliferation for maintenance 14 , the proliferating capability of BCG AM was examined by in vivo BrdU incorporation (Extended Data Fig. 1e). BCG AM showed significantly increased BrdU incorporation over control AM (Extended Data Fig. 1f), and their increased MHC II was independent of their proliferating status because both BrdU + and BrdU − AMs expressed increased MHC II (Extended Data Fig. 1g).
We next examined the immunophenotype and metabolic state of airway AM at an earlier 2-week time point post-BCG (Extended Data Fig. 1h). In contrast to their trained phenotype acquired at 5 weeks, the phenotype of airway and lung tissue AM of BCG hosts was comparable to the controls (Extended Data Fig. 1i-k). The 2-week BCG AM also demonstrated significantly reduced glycolysis compared to their controls (Extended Data Fig. 1l). To assess the mechanisms for such time-dependent training of airway AM, we first examined its relationship to mycobacterial dissemination post-BCG. We observed a small extent of BCG dissemination to the mediastinal lymph nodes (MLN) (40 ± 20 colony forming unit (CFU)/MLN) but not to the lung at 2 weeks. Because BCG possesses a slow-doubling time, this finding thus implicated the time-dependent mycobacterial dissemination in memory AM induction. We next addressed the role of BCG viability/replication and dissemination in AM training by comparing subcutaneous injection of viable BCG with heat-inactivated BCG (BCG-ia). Contrary to trained AM in viable BCG-vaccinated hosts, BCG-ia AM exhibited a profile in MHC II and IL-6 expression similar to untrained AM (Extended Data Fig. 1m).
We also determined whether, besides the lung, BCG vaccination had a global effect on macrophages in the peritoneal cavity. The peritoneal macrophages (PM) were identified as CD11b + F4/80 hi SiglecF − population 37 . BCG PM demonstrated a significantly altered immune phenotype with reduced F4/80 expression (CD11b + F4/80 Low SiglecF − ) compared to PBS control (Fig. 1o) and a trained phenotype with constitutively increased MHC II expression without restimulation (US) and increased IL-6/TNF production upon ex vivo restimulation (S) (Fig. 1p). On the contrary, consistent with its failure to train AM (Extended Data Fig. 1m), BCG-ia also failed to train PM (Fig. 1p).
The above data indicate that parenteral BCG leads to a time-dependent induction of memory AM. Such memory AM is characterized by increased MHC II and TLR2 expression, glycolysis and cytokine production upon stimulation. BCG vaccination also globally trains macrophages in the peritoneal cavity.

Distinct gene profile and microbial control by memory lung macrophages
We next examined the transcriptional profile of memory AM in BCG hosts. Airway BAL BCG or control cells were cultured with (S) or without (US) stimulation, and isolated RNA was sequenced (Fig. 2a). The principal component analysis (PCA) shows that each of the groups clustered into its own pattern, indicating a unique gene expression profile in each ( Fig. 2b and Extended Data Fig. 2a). A total of 248 genes were differentially expressed (DE) between BCG and control AM populations at baseline and after WCL restimulation (Extended Data Fig. 2b,c). BCG AMs with and without stimulation were enriched in gene sets involved in cell cycle and division ( Fig. 2c and Extended Data Fig. 2d). For instance, Mki67 (Ki67), a cellular proliferation marker, Kif11, Kif15, Kif23 encoding kinesin-like proteins involved in chromosome segregation and spindle formation and Rad51, Rad54b involved in DNA Until recently, little was known about whether mucosal-resident macrophages can directly be trained to store lasting innate memory 3 . We and others have discovered that respiratory mucosal exposure to microbes/vaccines can induce airway memory macrophages with TII 14,15 or immune-regulatory or tolerized property [16][17][18] . Respiratory adenoviral-vectored vaccination/infection induced a persisting memory phenotype in resident alveolar macrophages (AMs), independent of circulating monocytes. TII associated with such memory macrophages enhances innate protection against both the intended target and heterologous bacterial pathogens in the lung 14,19 .
Hence, growing evidence supports a paradigm of compartmentalization in the genesis of resting-state innate immune memory resulting from the recent history of immunological imprinting/training. This paradigm ascribes trained hematopoietic progenitors and circulating monocytes to systemic/parenteral microbial exposure/vaccination, while it attributes the memory phenotype of barrier mucosal-resident macrophages to the local microbial exposure/vaccination 3,20,21 . The latter is in line with the current concept of macrophage niche of tissue residence and its adaptation to local inflammation 22 . It has remained unclear whether, in the absence of local inflammation, innate memory at barrier tissues may develop as a way of integrating and adapting to distally generated immunological signals following systemic/parenteral microbial exposure/vaccination 23 . Recent studies show that subcutaneous viral infection/vaccination triggers a widespread immunological alert across multiple tissue sites 24,25 and that local tissue injury can activate resident macrophages in remote tissue sites 26 . One prototypic modality of the immunological cross-talk between tissue sites is the gut-lung axis whereby intestinal microbiota dysbiosis alters immune responses in the lung [27][28][29][30][31][32][33][34] . However, it is not well understood whether parenteral vaccination affects the intestinal microbiome and whether/how intestinal dysbiosis induces lung-resident innate memory 23 . As BCG is administered via the skin, like most current human vaccines (including coronavirus disease 2019 (COVID-19) vaccines), addressing these questions has far-reaching implications.
Here using an experimental model, we investigated whether and how subcutaneous BCG vaccination induces tissue-resident memory macrophages and TII in the lungs. Aside from its effect on myeloid progenitors, parenteral BCG independently induces memory AM and TII against Mycobacterium tuberculosis infection. This process occurs via the initial mycobacterial spread and a gut-lung axis involving a time-dependent alteration in intestinal microbiota, barrier function and metabolites. Our study thus identifies an intestinal microbiota-mediated pathway to innate memory development at distal mucosal sites and has implications for the development of next-generation vaccines against respiratory pathogens 35,36 .

Subcutaneous BCG induces a memory phenotype in lung macrophages
Subcutaneous BCG vaccination induces trained monocytes in the BM/ blood 9,13 . It is unclear whether it trains lung-resident AM. Thus, mice were vaccinated subcutaneously with BCG. At 5 weeks postvaccination, the bronchoalveolar lavage (BAL) and lung mononuclear cells were analyzed with (S) or without (US) stimulation by M. tuberculosis whole cell lysates (WCL) (Fig. 1a). AMs were identified as Ly6G − CD11b − CD11c hi CD64 hi Siglec-F hi myeloid cells 14 . Approximately, 95% of airway cells of BCG-vaccinated hosts were AMs. Compared to control AMs (PBS-US), BCG AMs (BCG-US) expressed higher levels of baseline and augmented MHC II upon stimulation (PBS-S versus BCG-S) (Fig. 1b). The stimulated AMs also exhibited significantly increased toll-like receptor 2 (TLR2) (Fig. 1c). Significantly higher frequencies of stimulated BCG AM produced IL-6, while tumor necrosis factor (TNF)-producing cells were comparable (Fig. 1d,e). Similar immune profiles were observed among lung tissue AM populations from control and BCG-vaccinated hosts (Extended Data Fig. 1a-c).
Article https://doi.org/10.1038/s41590-022-01354-4 repair, and the genes involved in sister chromatin segregations were upregulated in BCG AM. BCG AM also upregulated defense response genes, particularly those involved in chemotaxis of T cells (Cxcl10, Ccl5), monocytes (Ccl7) and downregulated cell activation genes Lck, Bcl2, Il7r and Slamf7 (Fig. 2d). Furthermore, the predefined gene sets related to TII 1 and antigen (Ag) processing and presentation 14 Cy5-5-A :: TNF-α  PE-A :: IL-6  BV421-A :: TLR2   MHC II  TLR2  IL-6  TNF   0  . f, Heatmap of cytokine/chemokine protein levels (geometric means) in supernatants of airway AM cultured with (S) or without (US) stimulation. Red asterisks denote significantly increased cytokine/chemokine production upon stimulation by airway AM of BCG hosts. g-l, Concentrations of IL-6 (*P = 0.0230) (g), IL-12p40 (****P < 0.0001) (h), TNF (i) and MCP-1 (**P = 0.0031) (j), MIP-1α (*P = 0.0408) (k) and RANTES (*P = 0.0137) (l) in supernatants of airway AM cultured with and without stimulation. m, Real-time ECAR in airway AM at 8 weeks post-BCG immunization. 2-DG: 2-deoxy-glucose. n, Glycolysis (***P = 0.0006), glycolytic capacity (***P = 0.0007) and glycolytic reserve (**P = 0.0013) in airway AM at 8 weeks post-BCG immunization. o, Dotplots of frequencies of cells gated out of total live peritoneal cells expressing F4/80 surface marker and the median fluorescent intensity (MFI) of F4/80 expression by PM at 5 weeks post-BCG vaccination (*P = 0.0231, ****P = <0001). p, Histograms of and the MFI of MHC II and frequencies of IL-6-or TNF-producing PM at 5 weeks postvaccination with viable BCG or inactivated BCG (BCG-ia) or PBS with (S) or without (US) stimulation (****P = <0.0001). Data in b, d and e are representative of three independent experiments. Data in f, g-l and o-p are representative of two independent experiments. Data in bar graphs are presented as mean ± s.e.m. and represent individual data points of biologically independent samples, n = 3 mice per group. One-way ANOVA was used for multiple comparison testing with Bonferroni's test for data in b-l, two-tailed t-test for data in n and one-way ANOVA followed by Dunnett's multiple comparison test for data in o and p.  (Fig. 1b), expression levels of the genes involved in antigen presentation significantly increased in BCG AM at baseline (US) and upon stimulation (S) (Fig. 2e), while those in glycolysis and mTOR pathways did not significantly differ (Extended Data Fig. 2e,f).   Il7r  Inhba  Clec1a  Cx3cr1  Aoah  Cfp  Ccl5  Oas2  Tlr9  Chst1  F3  Aqp9  Tcf7  Lck  Thy1  Bcl2  Stat4  Igfbp4 Ltb Ptx3 Ccl7     On the contrary, expression of genes related to fatty acid oxidation significantly decreased in stimulated BCG AM ( Fig. 2f and Supplementary  Table 1), consistent with a shifted metabolism from oxidation toward glycolysis (Fig. 1m,n) 12 . These data suggest that BCG vaccination leads to a unique transcriptional profile in memory AM. We then used the data in the current study on differentially expressed genes (DEG) by BCG AM versus PBS AM and analyzed them against the DEG in AM of intranasally adenoviral (Ad)-vaccinated mice 14 . About five times more genes (1,309 versus 248 genes) were differentially expressed by trained Ad AM (Extended Data Fig. 2g) compared to BCG AM (Extended Data Fig. 2b). However, both shared similar features in predefined gene sets related to antigen presentation, glycolysis, mTOR pathway and fatty acid oxidation (Extended Data Fig. 2h). The transcriptomic features distinguishing Ad AM from BCG AM were the upregulated cell activation genes, Lck, Bcl2, Il7r and Slamf7 and no enrichment for cell cycle and division-associated genes. These data suggest that although some features of trained AM are shared between certain vaccine strategies, the trained AM are unique depending on the vaccine type and delivery route.
Given that BCG-trained AM displayed enhanced MHC II and antigen presentation/processing genes, we assessed their capability of antigen presentation to T cells ex vivo by using transgenic M. tuberculosis Ag85B-specific CD4 T cells cocultured with Ag85B-laden BCG AM (Extended Data Fig. 3a). T cell proliferation rates were calibrated as the extent of CFSE dilution by FACS. While ~30% of T cells cultured with BCG AM underwent at least three generations of proliferation (G), only 15% of those cultured with control AM underwent mostly one-generation proliferation (Extended Data Fig. 3b,c), indicating enhanced antigen presentation by BCG AM.
To examine their antimicrobial activity, BCG and control airway AM were infected ex vivo with M. tuberculosis and mycobacterial inhibition/killing rates were determined by CFU assay (Fig. 2g). Compared to control AM, BCG AM exhibited a significantly greater ability to control M. tuberculosis at both 24-h and 48-h postinfection ( Fig. 2h) or to control Mycobacterium bovis expressing a fluorescent protein (BCG-dsRed) ( Fig. 2i.j). Augmented M. tuberculosis control was also seen in CD11C + /CD11b + antigen-presenting cells (APC) from BCG lung tissue (Extended Data Fig. 3d). Both trained and control AM exhibited similar phagocytosis (Fig. 2i,j) and cell death/apoptosis (Extended Data Fig. 3e) rates. In keeping with their increased ex vivo MHC II-mediated M. tuberculosis antigen presentation (Extended Data Fig. 3b,c), trained AM rapidly further upregulated MHC II expression upon in vivo M. tuberculosis infection (Fig. 2k,l). These data indicate that besides their memory phenotype in BCG hosts, trained AMs show a distinct transcriptional profile, increased antigen presentation and antimycobacterial activities.

Trained immunity by memory lung macrophages against pulmonary TB
Anti-TB host defense has long been attributed solely to adaptive Th1 immunity induced by parenteral BCG vaccination. Airway AMs are known to harbor most of M. tuberculosis bacilli and contribute to its dissemination in the early stages of TB 38 . Enhanced mycobacterial control/responses by BCG-trained AM in ex vivo settings (Fig. 2h-l) suggest that such AM could offer TII against TB independent of T cell immunity in vivo. To begin examining whether memory AM offers anti-TB TII, 4-week BCG-vaccinated animals were infected with M. tuberculosis and lung CFU was assessed at days 7 and 14 (Fig. 3a). Compared to unvaccinated control, BCG lungs had ~0.3 log and ~1 log-reduced M. tuberculosis CFU at days 7 and 14, respectively (Fig. 3b). Consistent with airway AM being the primary M. tuberculosis reservoir within the first 7-9 d 38 , BCG airway cells (BAL cells) at days 3 and 7 (Fig. 3c) contained significantly reduced M. tuberculosis CFU compared to nonvaccinated counterparts (Fig. 3d), coincided with significantly reduced M. tuberculosis CFU in cell-free fluid (BALF) of BCG hosts at day 3 (Fig. 3e). These data suggest that BCG-trained AM can better control M. tuberculosis in vivo.
We next investigated the potential role of BCG-activated Th1-cells in enhanced AM control of M. tuberculosis in BCG hosts. We first characterized the kinetics of antigen-specific CD4 T-cells in the airway and lung parenchymal tissue (LPT) at early time points post-M. tuberculosis (Fig. 3f) by using Ag85B-CD4 T-cell tetramers (tet) 39 . The T-cells within LPT were differentiated from intravascular counterparts via intravascular CD45.2 immunolabeling 40 (Extended Data Fig. 4a). Substantially activated CD4 + CD44 + T cells appeared in BCG airways as early as day 3 postinfection, whereas, they did not populate the control airways until day 7 (Fig. 3g,h). Likewise, tet + CD4 T-cells (CD4 + Ag85B + ) were seen only in BCG airways at day 7 while being absent in the control airways (Fig. 3g,i). However, both CD4 + CD44 + and CD4 + Ag85B + T-cells became comparable at day 14 between BCG and control hosts ( Fig. 3g-i). Similarly, the BCG LPT had significantly greater numbers of activated CD4 + CD44 + T-cells at days 3, 7 and 14 compared to the controls ( Fig. 3j and Extended Data Fig. 4a) and also contained more CD4 + Ag85B + T-cells, particularly at day 7 ( Fig. 3k,l). These data suggest the enhanced M. tuberculosis control by trained AM to be accompanied by increased Ag-specific T cells within the airway early during M. tuberculosis infection.
To address the direct relationship of trained AM to enhanced M. tuberculosis control, we depleted the T-cells in BCG hosts before

Fig. 4 | Induction of memory AMs is independent of trained circulating monocytes and T cell-derived signals. a, Experimental schema.
b, Representative dotplots of myeloid (MMP3) and lymphoid (MMP4) progenitors and frequencies of MMP3 out of total multipotent progenitors in the BM (*P = 0.0120) (n = 3 mice in PBS group, n = 4 mice in BCG group). c,d, MFI of MHC II on circulating Ly6C high (*P = 0.0228) (c) and Ly6C low (P = 0.0510) (d) monocytes with (S) and without (US) stimulation (n = 3 mice per group per cell type). e, Frequencies of Ly6C low TNF + monocytes with (S) and without (US) stimulation (*P = 0.0157) (n = 3 mice per group). f, Heatmap of cytokine/chemokine protein levels (geometric means) in the plasma from whole blood culture samples with (S) and without (US) stimulation. Red asterisks denote significant differences upon stimulation of airway AM of BCG hosts. g,h, Representative dotplots of SiglecF + Ly6C − airway AM (g) and lung tissue (h). MDM and IM in lung tissue were identified as SiglecF − Ly6C + and SiglecF − Ly6C − , respectively. The total numbers of macrophage subsets in airway and lung tissue are presented in the bar graph. Representative of two independent experiments (n = 3 mice per group per tissue). i, MFI of MHC II on airway AM from BCG-vaccinated or PBS-treated CCR2KO mice with (S) and without (US) stimulation and cytokine/chemokine levels in culture supernatant of airway AM with stimulation (S) (*P = 0.0280 for TNF; *P = 0.0335 for IL-6; *P = 0.0239 for IL-10). n = 3 mice in PBS group, n = 4 mice in BCG group. j, MFI of MHC II on lung tissue AM from BCG-vaccinated or PBS-treated CCR2KO mice with (S) and without (US) stimulation (US: *P = 0.0238; S: *P = 0.0246) and frequencies of lung tissue AM producing IL-6 (US: *P = 0.0284; S: *P = 0.0297) and TNF with and without stimulation. k, Experimental schema. l, Representative histograms of PKH-labeled AM in the airway of BCG-vaccinated or PBS-treated WT and CCR2KO animals, compared to naïve mouse AM without PKH-labeling (no PKH). n = 3 mice per group. m,n, Signature scores for embryonic origin (AM) (m) and circulating monocyte genes (n) in airway AM in PBS and BCG-vaccinated hosts. Horizontal lines in violin plots denote medians and dotted lines denote lower and upper quartiles. o, Experimental schema. p, Heatmap of cytokine/chemokine protein levels (geometric means) in culture supernatants of AM with stimulation, comparing PBS, BCG-vaccinated and BCG/T cell depletion (dep) groups. Red asterisks denote significant differences upon stimulation of airway AM of BCG hosts. q,r, MFI of MHC II on airway AM (**P = 0.0066; ****P ≤ 0.0001) (q) and frequencies of IL-6-producing airway AM (*P = 0.0159; ***P = 0.0003; ****P ≤ 0.0001) (r) with (S) and without (US) stimulation, comparing PBS, BCG-vaccinated and BCG/IFN-γ-depleted (anti-IFNγ) groups. n = 3 mice per group. Data in bar graphs are presented as mean ± s.e.m. Statistical analysis was determined by two-tailed t-test for b-e, i, and j, comparing BCG with PBS. Data in q and r were analyzed by one-way ANOVA, followed by multiple comparisons with Bonferroni's test.
Article https://doi.org/10.1038/s41590-022-01354-4 M. tuberculosis infection by using mAbs (Fig. 3m) and compared day 7 M. tuberculosis CFU in BCG lungs (BCG ΔT cells) with those in unvaccinated (PBS) and control Ab-treated BCG hosts (BCG). Consistent with the earlier data ( Fig. 3b), BCG lungs (BCG) contained significantly reduced M. tuberculosis CFU (Fig. 3n). Depletion of T-cells in BCG hosts (BCG ΔT cells) did not compromise enhanced protection, but it rather further reduced M. tuberculosis CFU (Fig. 3n). Depletion of T-cells in unvaccinated (PBS) hosts did not affect M. tuberculosis CFU (4.27 ± 0.06 in PBS versus 4.16 ± 0.08 in PBS ΔT cells). On the contrary, the naive animals receiving the adoptively transferred BCG AM (BCG-AM) had moderately reduced lung M. tuberculosis CFU compared to those receiving the control AM (PBS-AM; Extended Data M. tuberculosis both in airway macrophages (Fig. 3p) and lung tissue (Fig. 3q) than the unvaccinated control. The above data together indicate the enhanced early TB protection in BCG hosts to be independent of T cells or circulating monocytes, further supporting the role of trained AM.

Independence of monocytes and T cells for memory macrophage induction
Because parenteral BCG vaccination was shown to train circulating monocytes via imprinting the BM myeloid progenitors 9,13 , we examined whether the BM myeloid cells and circulating monocytes were trained in our model (Fig. 4a). Indeed, significantly increased frequencies of MPP3 myeloid progenitors were observed in BCG hosts compared to the control (Fig. 4b). Using a gating strategy (Extended Data Fig. 5a), both Ly6C hi and Ly6C low monocytes in the peripheral blood of BCG hosts were found to express higher levels of MHC II upon stimulation compared to their controls (Fig. 4c,d and Extended Data Fig. 5b); however, they did not differ in their ability to produce IL-6 (Extended Data Fig. 5c). On the contrary, significantly increased Ly6C low monocytes from BCG hosts produced TNF upon stimulation (Fig. 4e) while TNF + Ly6C hi monocytes were comparable (Extended Data Fig. 5d). Consistent with enhanced activation of circulating monocytes, IL-1β, IL-6, IL-12p40, IP-10, MIP-1α and RANTES production in stimulated whole blood cultures from BCG hosts significantly increased compared to the controls ( Fig. 4f and Extended Data Fig. 5e). These data show that subcutaneous BCG vaccination leads to increased myelopoiesis in the BM and trained circulating monocytes.
The circulating monocytes may contribute to the pool of AM, particularly under inflammatory conditions in the lungs 21,22 . To address the relationship of trained circulating monocytes to the genesis of memory AM in BCG hosts, we first assessed the levels of pro-inflammatory cytokines/chemokines in the airway and found all of them to be undetectable. We next examined the airway macrophage and monocyte surface markers. At 2 weeks (Extended Data Figs. 6a) and 5 weeks (Fig. 4g) post-BCG, most airway (BAL) macrophages in both BCG and control hosts were Siglec-F + resident AM. Furthermore, frequencies and total numbers of major macrophage populations, monocyte-derived-macrophages (MDM) (Siglec-F -Ly6C + ), interstitial macrophages (IM) (Siglec-F − Ly6C − ) and AM (Siglec-F + Ly6C − ) were similar in lung tissues of both groups (Extended Data Figs. 6b and 4h). These data thus do not support a substantial contribution of circulating monocytes to the induction of airway-resident memory AM in BCG hosts.
To investigate this further, CCR2KO mice lacking classical Ly6C hi monocytes were BCG-vaccinated for 5 weeks. Compared to the controls, MHC II expression remained elevated in both airway and lung tissue AM of BCG-vaccinated CCR2KO animals with and without stimulation (Fig. 4i,j). Furthermore, upon stimulation, compared to the controls, BCG AM from CCR2KO hosts produced significantly higher levels of IL-6 and TNF (Fig. 4i). Like the cytokine profile in wild-type (WT) BCG AM (Fig. 1d,e), there were significantly higher frequencies of stimulated CCR2KO BCG AM producing IL-6 while TNF-producing cells were comparable (Fig. 4j). The trained phenotype/immunity of AM from BCG CCR2KO animals was further supported by the functional data that these cells, upon ex vivo infection with M. tuberculosis, significantly better controlled mycobacterial infection (Extended Data Fig. 6c), consistent with ex vivo-infected WT BCG AM (Fig. 2g-j) and anti-TB TII in the lung of BCG CCR2KO animals ( Fig. 3o-q). These findings together indicate the circulating monocyte-independent induction of functional memory AM by BCG vaccination.
Using a different approach, we delivered a stable fluorescent dye PKH26 to label airway-resident AM via phagocytosis 42 in both WT and CCR2KO mice, and any contribution of circulating monocytes toward BCG-trained AM would have diluted PKH26 within airway AM (Fig. 4k). In nonvaccinated WT (WT-PBS) and CCR2KO (CCR2KO-PBS) hosts, the majority of airway AM remained stably labeled by PKH over 5 weeks (Fig. 4l). There was no loss/dilution of PKH from AM of both WT and CCR2KO BCG hosts (Fig. 4l), suggesting a minimum contribution of circulating monocytes to BCG-trained AM. Because the autonomously induced AM would remain similar to their steady-state counterparts in their embryonic (AM) and monocytic gene signatures 14 , we compared these genes in control and BCG and found no substantial differences (Fig. 4m,n), thus further supporting the independence of circulating monocytes for memory AM induction.
Because BCG activates Th1-cells (Fig. 3g,I,k,l) which produce IFN-γ involved in training monocytes/macrophages 13,14 , we determined the role of T-cells and IFN-γ in BCG-trained AM. T-cells or IFN-γ were depleted by using mAbs from 2 weeks post-BCG when there was a lack of trained AM (Extended Data Fig. 1h-l) and their depletion was maintained over the next 3 weeks (Fig. 4o). While in keeping with earlier observations (Fig. 1f), BCG AM produced greater amounts of cytokines upon stimulation (S), and T-cell depletion did not compromise such enhanced responses by BCG AM to restimulation but it rather led to further increased IL-1β, IL-6, MIP-1α and MIG production (Fig. 4p). Similarly, IFN-γ neutralization did not impair the trained AM phenotype with elevated MHC II (Fig. 4q)   6d) production. These results are consistent with the fact that T-cell depletion in BCG-vaccinated hosts did not impair increased protection in the early stages of TB (Fig. 3n). The above data indicate that BCG induction/maintenance of airway-resident memory AM is independent of trained circulating monocytes, T cell help or IFN-γ.

Alterations in intestinal microbiota, metabolites and barrier function
Recent evidence suggests that gut microbiome may undergo changes in response to distal infections and such changes can alter immune responses in the lung via the gut-lung axis [27][28][29][30][31][32]34 . To address whether the gut-lung axis was involved in mucosal-resident memory AM induction in our model, we first characterized the cecal microbiome at 5 weeks. BCG vaccination significantly reduced the cecum size at 5 weeks (Extended Data Fig. 7a). Alpha diversity comparison based on the operational taxonomic unit (OTU) revealed a significantly lower microbial richness or alpha-diversity index in BCG cecum (Fig. 5a). Actual abundance at the rank of bacterial phylum was also reduced in BCG animals with the mean abundance of top four phyla being 20,329 versus 33,125 (P = 0.06) in control animals (Extended Data Fig. 7b). The intestinal microbiota clusters from BCG hosts were significantly separated from those in control animals despite some overlap by principal coordinate analysis (PCoA) (P < 0.004, PERMANOVA; Fig. 5b). Moreover, at rank of bacterial family, while Muribaculaceae, also known as S24-7, predominated both in control and BCG hosts, a significantly increased frequency of Lactobacillaceae was present in the gut microbiome of BCG hosts (Fig. 5c,d). Nineteen specific OTUs were differentially changed in the BCG host (false discovery rate (FDR) < 0.05;    Table 2). Because significant induction of TII in AM was observed at 5 weeks (Fig. 1a-p) but not at 2 weeks (Extended Data Fig. 1h-l), we also characterized the intestinal microbiome at 2 weeks post-BCG. Indeed, 2-week intestinal microbiome did not change much, resembling the control gut microbiome (Extended Data Fig. 7c-f), despite their partial overlap (Extended Data Fig. 7e) and 14 specific OTUs being differentially changed (Supplementary Table 3). These results indicate a time-dependent development of intestinal dysbiosis postparenteral BCG vaccination. We next evaluated whether intestinal dysbiosis in BCG hosts was accompanied by microscopic histologic changes at 5 weeks postvaccination. Compared to the control, marked changes were observed in the distal colon of BCG hosts characterized by irregular villi, shortened crypts, enlarged lumen, epithelial disruption/sloughing and inflammatory infiltrates in the mucosa (Fig. 5e,f and Extended Data Fig. 7g). Although these changes are distinct, they were mild in severity according to the guidelines on murine intestinal inflammation 43 . To address whether BCG vaccination-triggered colitis was time-dependent and self-limited, we examined colon histology at 2 weeks and 8 weeks post-BCG and compared it to 5 weeks when the colitis was overt. Consistent with limited changes in intestinal microbiome at 2 weeks (Extended Data Fig. 7c-f), colon histology remained unchanged except for the low-degree lymphocytic infiltration ( Fig. 5g and Extended Data Fig. 7h). Contrast to the colitis at 5 weeks (Fig. 5e,f), by 8 weeks postvaccination, colitis largely resolved and the colon architecture restored ( Fig. 5g and Extended Data Fig. 7h). These findings indicate that parenteral BCG triggers time-dependent but self-limited intestinal inflammation. Given the colitis at 5 weeks, we immunohistochemically examined the expression of epithelium tight junction proteins, zonula occludens (ZO-1) and occludin. Irregular and disrupted the distribution of ZO-1 (Fig. 5h) and occludin (Extended Data Fig. 7i) was seen in the colonic epithelium of BCG hosts, compared to their even/ intact distribution in control hosts. As a result, there was significantly increased intestinal permeability in BCG hosts shown by a fluorescein isothiocyanate (FITC)-labeled dextran method (Fig. 5i). These findings suggest an association of intestinal dysbiosis with intestinal structural changes and increased translocation of intestinal luminal molecules across the epithelium.
As intestinal dysbiosis is often linked to changes in its metabolites 28 , we profiled the metabolome in the cecum, colon and serum from 5-week BCG animals. Partial least square-discriminant analysis (PLS-DA) indicated an intergroup clustering of cecal metabolites with some overlap (Fig. 5j). Compared to control hosts, deoxycarnitine/γ-butyrobetaine, a precursor of l-carnitine (Fig. 5k), and creatinine (Extended Data Fig. 8a) levels increased significantly in the cecum of BCG hosts. Metabolites in the colon of each group also clustered out into their own patterns (Extended Data Fig. 8b) with substantial increased lactic acid levels in BCG hosts (Extended Data Fig. 8c). Besides changes in intestinal microbial metabolites, the PLS-DA model showed a clear between-group clustering of serum metabolites (Fig. 5l). Based on variable importance in projection (VIP) scores (>1.5), ten discriminating serum metabolites were rank-ordered, showing decreased concentrations of all metabolites, except creatine, in BCG hosts compared to the controls (Extended Data Fig. 8d, colored boxes on the right). Pathway analysis of the murine serum metabolome revealed the arginine metabolic pathway to be predominately impacted in BCG hosts compared to controls (Extended Data Fig. 8e). Because short-chain fatty acids (SCFAs) are among the major metabolites of gut microbiota generated upon dietary fiber breakdown and have inflammatory/metabolic impacts within and beyond the gut 28 , we quantified the major SCFAs. While acetate in the cecum was not impacted by BCG-induced intestinal dysbiosis, the relative proportions of propionate and butyrate were altered (Extended Data Fig. 8f) with butyrate levels significantly increased (Fig. 5m). Using PICRUSt 44 as a predictive tool, we also explored the functional potential of the intestinal microbiome based on its differences between BCG and control animals. Nineteen predictive functional molecules significantly differed (P < 0.05), categorized into energy metabolism, transporters, signaling and cellular processes and genetic information processing (Supplementary Table 4). In keeping with elevated butyric acid levels in the cecum (Fig. 5m), acetolactate synthase I/II/III large subunit involved in butonate (salts and esters of butyric acid) metabolism was significantly increased in BCG hosts. Because besides metabolites, other luminal molecules including microbial-associated molecular patterns (MAMPs) may also translocate into the circulation due to increased permeability, we measured serum LPS levels but found them not to differ (<0.01 EU per ml; Extended Data Fig. 8g).
Given the metabolomic changes in the gut/serum of BCG hosts, we examined the metabolome in the lung at 2 and 5 weeks. Consistent with little and marked changes in gut microbiome/metabolome at 2 weeks (Extended Data Figs. 7c-f) and 5 weeks (Fig. 5a-d,j-m), respectively, the profile of lung metabolites significantly differed only at 5 weeks, but not at 2 weeks (Fig. 5n). Based on VIP scores (>1.5), seven discriminating lung metabolites were rank-ordered (Extended Data Fig. 8h). In keeping with increased deoxycarnitine levels in the cecum at 5 weeks (Fig. 5k), the carnitine products, butyryl carnitine and hexanoyl carnitine, significantly elevated in the lung at 5 weeks post-BCG (Fig. 5o,p). These data demonstrate a time-dependent association in metabolomic changes between the gut, serum and lung.
Considering that besides lung AM, BCG vaccination had a global training effect on PM (Fig. 1o,p) and there exists a biological connection between the peritoneal cavity/macrophages and the organs including the gut within the peritoneal cavity 45,46 , to address how BCG vaccination mediated changes in the gut we examined the possibility of BCG translocation to the peritoneal cavity at 2 weeks post-BCG. Indeed, not only was the length of the colon significantly different between control and BCG hosts with the latter being shortened by ~2 cm, suggesting mild colitis (Fig. 5q), but substantial BCG CFU was detected in the MLN, cell-free peritoneal washes (PW) and total macrophage fraction (peritoneal cell lysate (PCL)) only from BCG hosts (Fig. 5r), not from unvaccinated controls. These data establish a mechanistic linkage between distal BCG vaccination and the marked alterations in the gut.
The above data collectively suggest that via mycobacterial dissemination to the gut-associated sites, BCG vaccination leads to time-dependent alterations in intestinal microbiota, metabolome and barrier function which, in turn, result in metabolomic changes in serum and lung.

BCG vaccine-conditioned intestinal microbiota induces memory lung macrophages
We next used a microbiota transplant approach to address the relationship of BCG-induced intestinal dysbiosis to induction of memory AM. Naive mice were treated with broad-spectrum antibiotics for 10 d to perturb the original microbiota before transplantation with cecal microbiota (CM) from control (PBS-CM) or BCG (BCG-CM) hosts and allowed to colonize over the next 5 weeks (Fig. 6a). Consistent with the trained phenotype of BCG AM (Fig. 1a-n), airway AM of those colonized with BCG-CM demonstrated elevated MHC II and IL-6 and TNF production at baseline (US) and upon stimulation (S) compared to those colonized with PBS-CM (Fig. 6b-d). A similar trained phenotype was also seen with AM in BCG-CM, but not PBS-CM, lung tissue ( Fig. 6e-g). These data, thus, establish a causal relationship between BCG-conditioned microbiota and memory AM induction.
As expected, unlike in BCG-vaccinated hosts (Fig. 5e,f and Extended Data Fig. 7g), BCG-CM colonization of naive animals did not cause major colonic architectural changes (Fig. 6h). However, it led to epithelial hyperplasia with reduced goblet cells and mild epithelium disruption (Fig. 6h) and reduced epithelial mucin-2 (muc-2) production compared to PBS-CM controls (Fig. 6i). Because gut dysbiosis is linked to changes in BM myeloid hematopoiesis 47 , we examined the  circulating monocytes and found BCG-CM colonization to significantly activate circulating Ly6C high and Ly6C low monocytes over PBS-CM controls (Fig. 6j).
To investigate whether BCG-CM colonization-trained AM might translate to lung TII as seen in BCG-vaccinated hosts (Fig. 3b,d,e,n,p,q), the CM-colonized naive animals were infected with M. tuberculosis and CFU was assessed at 3 d (Fig. 6k). Indeed, airway BCG-CM AM (BAL) contained significantly reduced mycobacterial bacilli compared to PBS-CM controls (Fig. 6l). Correspondingly, CFUs in cell-free BALF and lung tissue were also trending smaller in BCG-CM animals (Fig. 6m,n). The above data together indicate that transplantation of BCG-conditioned intestinal microbiota alone can induce intestinal barrier changes, memory AM and anti-TB TII in the lung.

Role of circulating microbial metabolites in training lung macrophages
To examine the relationship of circulating metabolites to BCG-trained AM, we adapted an in vitro model well-established for monocyte training 48 . Naive airway AMs were incubated in a culture medium supplemented with serum from BCG-vaccinated or control animals (training). After 24-h training and 24-h or 3-d resting, cells were stimulated, microscopically analyzed and immunophenotyped (Fig. 7a). Different from control serum, BCG-conditioned serum (BCG-S) caused remarkable morphologic changes of AM after training or resting and upon restimulation ( Fig. 7b and Extended Data Fig. 9a). These AM congregated in clusters and were larger with cytoplasmic inclusions (after training; Fig. 7b). Upon resting, there appeared increased cell divisions and spreading (after resting), consistent with their enriched genes involved in cell division in AM from BCG hosts (Fig. 2c and Extended Data Fig. 2b-d). The most marked morphologic changes were seen upon restimulation (after stimulation), accompanied by significantly elevated MHC II and IL-6 production (Fig. 7c-e). This trained immunophenotype was similar to memory AM in BCG-vaccinated hosts (Fig. 1b,d,e) and was observed only with the AM rested for 3 d but not with those rested for 24-h after training, consistent with previous observations 48 . Furthermore, production of training biomarkers IL-1β, IL-6 and TNF along with MCP-1 and KC significantly increased from the AM exposed to BCG-S over the controls (Fig. 7f,g). These data suggest a role for circulating soluble factors in BCG-trained AM. Because innate training involves epigenetic reprogramming via histone methylation/acetylation 1 , we assessed whether histone modification was involved in the observed training effect of BCG-S on AM. Thus, during the training with BCG-S, the culture medium was supplemented with either histone methyltransferase inhibitor, 5′-deoxy-5′-methylthioadenosine (MTA) or histone acetyltransferase inhibitor epigallocatechin-3-gallate (EGCG) 48 . Inhibition of histone modification enzymes, particularly histone acetyltransferase (BCG + EGCG-S), significantly reduced IL-1β, IL-6 and TNF production by BCG-S-trained AM upon restimulation (Fig. 7h-j). A relatively minor inhibitory effect was observed with methyltransferase inhibitor (BCG + MTA-S). These data suggest the involvement of epigenetic modification in AM training by circulating metabolites in BCG hosts.
Given the immunomodulating role of intestinal microbiomederived circulating SCFAs 28 and our observed increases in deoxycarnitine and butyrate in BCG-vaccinated hosts (Fig. 5j-p), we determined the relationship of these SCFAs to memory AM induction. A mix of l-carnitine and butyrate hydrochloride was introduced to the drinking water (DW + M) of naive animals for 3 weeks 49 and the control animals received the regular drinking water (DW) (Fig. 8a). DW + M AM demonstrated enhanced IL-6 and MIP-1α production among the cytokines examined over the control (DW) upon stimulation (S) (Fig. 8b,c and Extended Data Fig. 9b), similar to the profile of BCG-trained AM (Fig. 1g,k). However, unlike the circulating monocytes in BCG hosts (Fig. 4f and Extended Data Fig. 5), the DW + M monocytes did not assume a trained phenotype, displaying a suppressed immune profile with reduced Ly6C and MHC II expression (Extended Data Fig. 9c,d). Their ability to secrete cytokines/chemokines also remained comparable (Extended Data Fig. 9e). Furthermore, as expected, unlike in BCG-vaccinated hosts (Fig. 5e,f and Extended Data Fig. 7g), the metabolite supplementation (DW + M) did not cause colonic architectural changes except a mild lymphocytic infiltration (Extended Data Fig. 9f). We next compared the transcriptomic profiles in trained AM by metabolite supplements and BCG vaccination. DW + M airway AM (Fig. 8a) were subjected to transcriptional analysis with (S) and without (US) stimulation. Each of the groups was found to cluster into its own pattern, suggesting the transcriptional alteration in AM following metabolite treatment (Fig. 8d). A total of 265 genes were differentially expressed in DW + M AM compared to DW controls (Extended Data Fig. 9g,h). Like BCG AM (Fig. 2c and Extended Data Fig. 2d), the genes associated with cell differentiation/proliferation (Nov, Hbegf, Kitl and Six5) were also upregulated in DW + M AM compared to controls (Extended Data Fig. 9g). The gene, Snca, a microphage/microglial activation gene required for inflammatory responses was also upregulated in both AM (Fig. 2d and Extended Data Fig. 9g). Also similar to BCG AM (Figs. 1b and 2c,e), the immune genes including HLA genes (Extended Data Fig. 9h) and antigen presentation genes (Fig. 8e) were significantly increased in stimulated DW + M AM. Furthermore, the levels of the genes associated with fatty acid oxidation, glycolysis and mTOR pathway in DW + M AM (Extended Data Fig. 9i-k) were in general similar to those of BCG AM (Fig. 2f and Extended Data Fig. 2e,f). Of importance, there was significantly enhanced protection in the early stages of M. tuberculosis infection in DW + M lungs, even to a greater extent than in BCG hosts (Fig. 8f). These findings indicate that supplementation of BCG-altered metabolites alone could induce trained immunophenotype, transcriptomic changes and TII in AM similarly as BCG vaccination.
Given that serum LPS was undetectable (Extended Data Fig. 8g), we further investigated the role of intestinal MAMPs potentially translocated into the circulation in BCG-vaccinated hosts. Considering microbial peptidoglycan, LPS and muramyl dipeptide are ligands for TLR2, TLR4 and NOD2, respectively, TLR2-, TLR4-and NOD2-deficient (KO) AM were trained ex vivo with BCG-S or PBS-conditioned serum using our ex vivo AM training model (Fig. 7a). While BCG-S induced significantly increased MHC II, an innate training marker, on WT AM without (US) and with (S) stimulation (Fig. 8g), it also induced increased MHC II on un-stimulated TLR2-or NOD2-KO AM (Fig. 8h,j). Stimulation with M. tuberculosis WCL enhanced MHC II further on NOD2-KO AM, particularly on those trained by BCG-S (Fig. 8j). Interestingly, TLR4-KO AM expressed reduced MHC II which did not differ between control and BCG serum before stimulation (Fig. 8i). Although the stimulation increased MHC II in these cells, it did not differ between control and BCG-S, suggesting the inherent requirement of TLR4 signaling for MHC II expression by AM. The above data suggest that even if present at heightened circulating levels, these MAMPs do not contribute significantly; however, circulating microbial metabolites have a critical role in AM training by BCG vaccination.

Discussion
It remains unclear whether, in the absence of local inflammation, lung-resident innate memory may occur following integrating and adapting to distally generated immunological signals postparenteral vaccination. Here, we show that subcutaneous BCG vaccination induces memory AM and TII in a time-dependent manner besides its training effects on circulating monocytes. Such memory AM develops and self-sustains independently of circulating monocytes. BCG vaccination does so via the gut-lung axis involving mycobacterial translocation, intestinal dysbiosis and increased permeability and changes in local/ systemic metabolites (Extended Data Fig. 10).
Our study, thus, reveals a new intestinal microbial metabolic pathway for innate memory/TII development at a distal mucosal site postparenteral vaccination (Extended Data Fig. 10), and it changes the current view that genesis of innate immune memory is compartmentalized according to the route of immunologic exposure 2,3,20 . Thus, parenteral vaccination could trigger a long-range immunological alert across multiple tissue sites, resulting in macrophage memory formation. Such knowledge shall enhance our understanding of host defense mechanisms by parenteral vaccines. It indicates that following immunization with a properly designed parenteral vaccine and upon respiratory entry of pathogens, trained mucosal-resident macrophages act as the first line of host defense which can be reinforced via the recruitment of trained circulating monocytes, a mechanism referred to as 'canonical tissue trained immunity' 20 . It also offers an additional mechanism for enhanced nonspecific innate protection in the lung of BCG-vaccinated humans [5][6][7]9 . It is noteworthy that different from parenteral BCG vaccination, parenteral adenoviral-vectored vaccination is unable to train AM 19 , suggesting the importance of choices of vaccine platform and route of delivery to mucosal-resident TII induction. The inability of BCG-ia to train AM suggests that the replicability of the parenteral vaccine is required for its widespread immunological alert and global macrophage-training effects, which is supported further by our finding that BCG spread appears required to initiate the gut-lung axis. That BCG replicates slowly may explain a slow build-up of its spread and a time-dependent manifestation of intestinal dysbiosis, colitis and metabolomic shifts. Of importance, mild colitis is self-limited as by 8 weeks it is largely resolved. Although two recent reports show the changes in intestinal microbiome following parenteral BCG vaccination 50,51 , there have not been any clinical reports on parenteral BCG-related colitis, let alone its linkage to metabolomic shifts in the gut and TII induction in the lung. The mild/transient nature of BCG vaccination-related colitis could explain its clinical insignificance. Furthermore, because it is only a proportion of parenteral BCG-vaccinated humans that develop innate protection against M. tuberculosis in the lung 52 , induction of lung-resident anti-TB TII via the gut-lung axis is likely genetically determined and ensues only in some human BCG vaccinees.
Our study also offers evidence that parenteral vaccination can induce intestinal dysbiosis-associated local/systemic metabolomic changes. We further demonstrate that induction of lung-resident innate memory via the gut-lung axis is independent of T cells or IFN-γ, different from their central role in the genesis of lung mucosal-resident macrophages via a local inflammatory pathway 14 . Besides its effects on the lung, intestinal dysbiosis/metabolites could also train circulating monocytes as shown in our current study, likely through influencing BM myelopoiesis 47 . As gut microbiota-derived SCFAs possess immune-modulatory properties 23,33 , a decline in acetate, a predominant SCFA, in prior flu-experienced mice affected AM bactericidality 30 and deficient SCFA production hampered microglia maturation 49 . Our finding that elevated deoxycarnitine and butyrate levels in gut-lung are linked to AM training significantly adds to our understanding of innate regulatory properties of SCFAs. Our approach of supplementing via drinking water of SCFAs to induce lung TII presents a potential immunotherapeutic strategy. Future studies shall address whether parenteral BCG-triggered intestinal dysbiosis may train macrophages in other mucosal tissues than the lung and peritoneal cavity.
Our findings also highlight the plasticity of tissue-resident AM in lung homeostasis and host defense 3,21 . The trainability, durability and autonomy of AM are in keeping with their ability to patrol the alveoli via crawling and to kill bacteria at a greater-than-neutrophil rate 53 . Thus, we show memory AM to enhance TB protection in BCG-vaccinated hosts independent of T cells or circulating monocytes. This is a plausible mechanism underlying the innate clearance of M. tuberculosis observed in a substantial proportion of BCG-vaccinated humans 52 . Having trained AM at the site of M. tuberculosis entry in BCG vaccinees is of importance to early control of infection given the ability of M. tuberculosis to hijack Article https://doi.org/10.1038/s41590-022-01354-4 airway macrophages for its dissemination 19,38,39 . Besides M. tuberculosis, we are currently investigating if parenteral BCG-induced lung TII can provide protection against heterologous bacterial infection.
In conclusion, our study reveals a new parenteral vaccine-triggered intestinal microbiota-mediated pathway to innate memory development in distal mucosal tissues. The work shows that parenteral immunization with a live vaccine can both centrally and peripherally induce TII. Such knowledge shall help design the next-generation vaccines against respiratory pathogens such as M. tuberculosis and SARS-CoV-2 (refs. 35,36 ).

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41590-022-01354-4.

Mice
WT female C57BL/6 mice were purchased from Charles River Laboratories (Saint Constant) or the Jackson Laboratory (Bar Harbor).
Female chemokine (C-C motif) receptor 2 (CCR2) (B6.129S4-Ccr2 tm1Ifc /J), TLR2 (B6.129-Tlr2 tm1Kir /J), TLR4 (B6(Cg)-Tlr4 tm1.2Karp /J), NOD2 (B6. 129S1-Nod2 tm1Flv /J) knock-out and P25 TCR-Tg transgenic mice containing CD4 T cells expressing Ag85B receptor (H2-K b -Tcra,-Tcrb)P25Ktk/J) on a C57BL/6 background were purchased from the Jackson Laboratory. All mice were 6-8 weeks of age upon arrival. Mice were housed in a specific pathogen-free level B facility or at the biosafety level 3 facility with ad libitum access to food and water, 12 h light cycle, 50-60% humidity and at 20-25 °C room temperature at McMaster University. Age-matched mice housed in the same room were used in each experiment. Control mice were administered subcutaneously with PBS used for the preparation of BCG. Animals were assigned experimental groups at random. All experiments were carried out in accordance with the institutional guidelines from the Animal Research and Ethics Boards of McMaster University (AUP 210822).

Bronchoalveolar and peritoneal lavages and mononuclear cell isolation
Mice were killed by exsanguination. In some instances, intravascular staining was carried out 3 min before exsanguination by injecting i.v. anti-CD45.2 antibody (clone 104) (BD Pharmingen) 40 . Cells in BAL and lung tissue were isolated as previously described 14,40 . The peritoneal cavity was lavaged as previously described 37 . Briefly, 3 ml of total wash solution (PBS containing 2 mM EDTA, 1 mM HEPES) was injected into the peritoneal cavity and the peritoneum was massaged gently for 30 s. Lavage fluid was collected with a pipette tip after making a small cut in the body wall. Spleen mononuclear cells were obtained as previously described 40 . BM cells were obtained by crushing the spine, femur and tibia bones in a mortar in PBS. BM cells were then filtered through a 40-μm basket filter (BD Biosciences). After lysing red blood cells, isolated cells were resuspended in either complete RPMI 1640 medium (RPMI 1640 supplemented with 10% FBS and 1% l-glutamine, with or without 1% penicillin/streptomycin) for ex vivo culture or in PBS for flow cytometry staining. When the BAL and lung cells were stimulated for intracellular cytokine staining (ICS) or cultured to measure cytokine/ chemokine levels in culture supernatants, cells were resuspended in complete RPMI 1640 medium containing 2% FBS.

Immunostaining, in situ cell proliferation and flow cytometry
Cell immunostaining and flow cytometry were performed as previously described 14 . Specifically, to determine alveolar macrophage activation levels and intracellular cytokine production, 250,000 mononuclear cells from BAL and 2 × 10 6 mononuclear cells from lung tissue were plated in a flat bottom 48-well plate and incubated for 3 h at 37 °C for macrophage adherence and tempering the irrelevant pro-inflammatory activities of freshly isolated AM. At the end of incubation, nonadherent cells were washed off and fresh media was added with and without M. tuberculosis WCL at a concentration of 1.6 μg ml −1 .To determine levels of trained circulating monocytes and intracellular cytokine production, whole blood was collected into EDTA blood tubes (Sarstedt) via cardiac puncture and diluted with an equal volume of RPMI 1640. Diluted whole blood was aliquoted to 300 μl and incubated with or without M. tuberculosis WCL at a concentration of 1.6 μg ml −1 . GolgiPlug (5 mg ml −1 ) (BD Biosciences) was added to BAL and lung cells and to diluted whole blood cultures 1 h after adding the stimulant. Cells were incubated for further 12-14 h. To determine activation levels of PM, 1 × 10 6 mononuclear cells were plated in a U-bottom 96-well plate with and without WCL at a concentration of 1.6 μg ml −1 (ref. 55 Table 5. A panel of mAbs was used to identify multipotent progenitors polarized toward myeloid (MMP3) and lymphoid (MMP4) progenitors 13 (Supplementary Table 5). For ICS of T cells, BAL, lung and spleen cells were cultured in the presence of GolgiPlug (5 mg ml −1 brefeldin A; BD Pharmingen) with or without a mixture of crude BCG and M. tuberculosis culture filtrate (2 μg per well) 54 . Stimulated cells were stained with cell surface antibodies, followed by fixation/permeabilization by using fixation/permeabilization solution kit (BD Biosciences) according to the manufacturer's instructions. Cells were then stained with anti-IFN-γ-APC mAb in Perm/Wash buffer (BD Biosciences) for 30 min on ice. Fluorochrome-labeled mAbs used for T cell surface and ICS were listed in Supplementary Table 5. For tetramer immunostaining, a tetramer for the immunodominant CD4 T cell peptide (FQDAYNAA-GGHNAVF) of Ag85B bound to the C57/Bl6 MHC class II allele (I-A(b) conjugated to PE fluorochrome (Ag85B:H-2I-A b ) (NIH Tetramer Core, Atlanta, GA) was used 39 .
For the determination of in situ AM proliferation, APC BrdU flow kit (552598; BD Biosciences) was used. Intranasal administration of BrdU was performed repeatedly at 5-week post-BCG immunization for a total of 9 d at a concentration of 0.5 mg per mouse in a total volume of 50 μl 14 . BrdU incorporated into DNA was then detected with an anti-BrdU-APC mAb (clone B44) 56 .
Unless otherwise indicated, all mAbs and reagents were purchased from BD Biosciences. All antibodies were validated and titrated for optimal conditions before their applications in the experiments. Immunostained cells were processed according to the BD Biosciences instructions for flow cytometry and run on a BD LSR II or BD LSRFortessa flow cytometer using FACSDiva software. Data were analyzed using FlowJo software (version 10.8.1; Tree Star).

Chemokine and cytokine quantification
To measure cytokines and chemokines produced by BAL AM, cells suspended in complete RPMI containing 2% FBS were plated in a 96-well flat bottom plate at 100,000 cells per well and rested for 3 h. After washing cells, fresh complete RPMI containing 2% FBS with and without 1.6 μg ml −1 of M. tuberculosis lysates was added to wells and incubated for 12-14 h at 37 °C and 5% CO 2 . Collected culture supernatants were frozen at −80 °C until measurement of protein levels. Whole blood Nature Immunology Article https://doi.org/10.1038/s41590-022-01354-4 was diluted with an equal volume of RPMI and 400 μl was plated in a 48-well flat bottom plate and incubated for 12-14 h at 37 °C and 5% CO 2 with or without stimulation with 1.6 μg ml −1 of M. tuberculosis lysates. Plasma was then collected and stored at −80 °C until the measurement of protein levels. Cytokines/chemokine levels were quantified by using MCYTOMAG-70K mouse chemokine and cytokine detection kit (Millipore Sigma, Oakville, ON) according to the manufacturer's instructions. Plates were read on a MagPix reader and concentrations of cytokines/chemokines were determined by using xPONENT software (Thermo Fisher Scientific).

Ex vivo mycobacterium phagocytosis and killing assays
Mycobacterial phagocytic and killing assay of AM from BAL and pooled CD11b + and CD11c + cells purified from lung using microbeads (Miltenyi Biotec) was performed as described previously 19 . Bacterial killing capacity (% killing) was calculated and CFUs determined in the culture plates by using the formula: % killing = (number of bacterial CFU at 4 h -number of bacterial CFU at 24/48 h)/(number of bacterial CFU at 4 h) ×100. Moreover, mycobacterial phagocytic and killing capacity of AM was also evaluated using a recombinant BCG expressing red fluorescence (BCG-dsRed) and flow cytometry. Apoptosis and necrosis among mycobacterium-infected cells (dsRed + ) were analyzed using a flow cytometer after staining the cells with Aqua dead cell staining kit and Annexin V-APC staining kit (BD Biosciences) according to the manufacturer's instruction.

Ex vivo alveolar macrophage antigen presentation to CD4 T cells
Ag85B-specific transgenic CD4 T cells purified from the spleen and lymph nodes of P25-Tg mice were labeled with Carboxyfluorescein succinimidyl ester (CFSE) (Invitrogen) 39 . CFSE-labeled CD4 T cells were cocultured with AM obtained from BAL at a ratio of 2:1 (2×10 5 T cells to 1×10 5 AM) at 37 °C and 5% CO 2 for 96 h. Cells were then washed and immunostained for CD3 and CD4 cell surface markers and CFSE dilution in CD3 + CD4 + cells was analyzed by flow cytometry.

In vitro macrophage training and epigenetic re-modeling inhibition
AM obtained from BAL of naive mice resuspended in RPMI 1640 with % l-glutamine, with or without 1% penicillin/streptomycin, were plated (100,000 cells per well in 96-well or 250,000 cells per well in 48-well plates) and rested for 3 h at 37 °C and 5% CO 2 . For innate immune training, the serum obtained from 5-week BCG-immunized or placebo mice was added to each well at 2% of the final volume and incubated for 24 h. After training, cells were washed once with culture medium (RPMI 1640 with % L-glutamine, with or without 1% penicillin/streptomycin and 2% FBS) and rested in the culture medium for 24 h or 3 d. The medium was changed once on day 2 for those rested for 3 d. After resting, cells were stimulated with or without M. tuberculosis WCL (1.6 μg ml −1 ) for 12-14 h. Cytokine/chemokine proteins in culture supernatants were measured by Luminex. Separately, cells lifted from the well were immunostained and analyzed by flow cytometry. To investigate whether inhibition of epigenetic modifying enzymes affects innate immune training, inhibitors of histone methyltransferase, 5′-deoxy-5′-MTA (Sigma-Aldrich) at the concentration of 1 mM or histone acetyltransferase, EGCG (Sigma-Aldrich) at the concentration of 15 μM were added to the cells incubated in BCG serum for 24 h 48 .

In vivo depletion of T cells and IFN-γ neutralization
To deplete CD4 and CD8 T cells in vivo, mice were injected i.p. with 200 mg of anti-CD4 mAb (clone GK1.5) and 200 mg of anti-CD8 mAb (clone 2.43) 14 . To achieve continuous T cell depletion, 2 d following the initial injection, repeated doses of 100 mg of anti-CD4 mAb and 100 mg of anti-CD8 mAb were administered i.p. at a 7-d interval as needed. For IFN-γ neutralization, mice were injected i.p. with 200 μl of rabbit antimurine IFN-γ serum or normal control rabbit serum or with 500 μg monoclonal anti-IFN-γ antibody (Clone XMG1.2) (Bio X Cell) or isotype control mAb at 2-week post-BCG immunization. This treatment was repeated once every 5 d for a total of 3 weeks.

In vivo alveolar macrophage labeling with PKH26-PCL
For selective labeling of AM, PKH26-Phagocytic Cell Linker (PCL; Sigma-Aldrich) was diluted in Diluent B (20 mM) as per the manufacturer's instruction and instilled intranasally into the lungs of mice (50 μl per mouse) 1 d before BCG immunization 42 . Control mice received PBS as control. The next day, one of the mice that received PKH26 or PBS was sacrificed to ensure selective labeling of AM.

In vivo intestinal permeability measurement
Tracer FITC-labeled dextran (4 kDa; Sigma-Aldrich) was used to assess in vivo intestinal permeability 57 .

In vivo metabolite supplementation
For treating mice with distinct metabolites identified in BCG-immunized mice, 40-mM sodium butyrate and 1.25-mg ml −1 l-carnitine (both from Sigma-Aldrich) were added to DW for 3 weeks 49 . Mice were then supplied with DW without metabolites for a week before the examination of AM and monocytes. The water intake was monitored every 2 d to ensure comparable water intake by mice provided with metabolite-supplemented water.

Adoptive transfer of AM
AM obtained by BAL from 5 weeks of subcutaneous BCG-immunized and unimmunized mice were transferred via the intratracheal route 14 . Four hours after transfer, mice were challenged with M. tuberculosis.

Cecal microbial transplantation
Cecum harvested from BCG-immunized and unimmunized mice were snap-frozen in liquid N 2 and stored at −80 °C until use 58 . Before cecal microbial transfer, mice were administered a broad-spectrum antibiotic cocktail containing vancomycin, neomycin, ampicillin and metronidazole at 0.5 g l −1 concentration each (Sigma-Aldrich) in sterile DW (ad libitum) for 10 d. During antibiotic administration, body weight was monitored daily. The cecal matter for transplantation was prepared after dilution in 10 ml PBS and incubating at 37 °C for 30 min. After cessation of antibiotics administration, freshly prepared 200 μl of the cecal matter was administered by oral gavage once daily for 3 d. The microbiota was allowed to colonize for 5 weeks postgavage.

Histological analysis and microscopy
To assess histological changes in the colon post-BCG, tissue sections were stained with H&E. To assess colonic barrier function, the tight junction components ZO-1 and occludin were visualized by immunohistochemistry. Briefly, after deparaffinization, rehydration and antigen-retrieval, tissue sections were treated with 3% hydrogen peroxide for 10 min to block endogenous peroxidase activity. Tissue sections were then blocked with 5% goat serum (Sigma) and were incubated overnight with either ZO-1 Polyclonal Antibody (1:50) (Invitrogen Life Technologies) or Occludin Rabbit Polyclonal Antibody (1:50), (Proteintech Group) at 4 °C. Signals were visualized using the HRP-DAB enzyme-chromatic reporter system (Dako EnVision+system HRP labeled polymer anti-rabbit from Dako and DAB from Sigma-Aldrich).
To assess the mucin produced in the colon epithelial surfaces, tissue sections were stained for MUC2 on the Leica Bond Rx automated Stainer (Leica Biosystems) at MIRC Core histology facility, Department of Medicine, McMaster University. Briefly, tissue sections were pretreated with Leica Epitope retrieval 2 and then stained with rabbit-monoclonal (EPR23479-47) anti-MUC2 (Abcam) (1:3000) and Leica Bond Refine detection kit. Histological examination and scoring were independently verified by two researchers blinded to the treatment groups.

Nature Immunology
Article https://doi.org/10.1038/s41590-022-01354-4 Cell morphology after training, resting and stimulating, the cells was studied by bright-field microscopy-EVOS cell imaging system (Thermo Fisher Scientific) and pictures were taken at ×5 and ×20 magnification. Images of representative micrographs were taken under a Zeiss Axio Imager 2 Research Microscope using AxioVision digital imaging software (Carl Zeiss Microscopy GmbH).

Metabolic assay of AM
Real-time cell metabolism of AM was determined by using the Seahorse XF Glycolysis Stress Test Kit (Agilent Technologies) according to the manufacturer's instructions and as previously described 14 . Extracellular acidification rate (ECAR) corresponding to glycolysis and oxygen consumption rates corresponding to mitochondrial respirations (Oxidative phosphorylation) were assessed by using a Seahorse XFe24 Analyser (Agilent Technologies). Glycolysis was represented by ECAR after the addition of 10-mM glucose. Glycolytic capacity was represented by maximum ECAR following the addition of 1-μM oligomycin. The glycolytic reserve was represented by the difference between glycolytic capacity and glycolysis. Data were analyzed using Wave Desktop software version 2.6 (Agilent Technologies) and normalized to protein.

RNA isolation and RNA-Seq
AMs were obtained via BAL. To ensure sufficient RNA for sequencing, two mice were pooled per sample. Triplicate samples were set up per group/condition. Unstimulated and stimulated samples were paired. Isolated AMs were cultured with or without stimulation with 1.6 μg ml −1 of M. tuberculosis WCL. Following 12-h incubation, total cellular RNA was isolated using an RNeasy mini kit (QIAGEN) containing RNase-free DNase kit according to the manufacturer's instructions. RNA samples were stored at −80 °C until use. The quality of RNA was verified, and subsequent RNA sequencing was carried out by Farncombe Metagenomic Facility at McMaster University. RNA integrity was checked using the Agilent bioanalyzer. It was ensured that all RNA samples had a RIN (RNA integrity number) of 7.0 or greater for the best-quality libraries. The RNA was then subject to a polyA bead enrichment (NEBNext_ PolyA_mRNA) process to enrich for mRNA with polyA tail, ensuring that high-quality RNA was obtained as any degraded transcripts would not be sequestered at this point. This step also ensured the removal of ribosomal RNA. The isolated mRNA was then converted to cDNA and made into a library containing adaptors and unique indexes using a ligation-based library prep kit (NEBNext_Ultra_II_Directional_RNA). The libraries were run on the bioanalyzer for a check of size and distribution and the concentration was checked using qPCR. The libraries were then pooled and run on two lanes of an Illumina HiSeq 1500 using onboard clustering in Rapid Mode. Furthermore, 25 M clusters were obtained per sample.

Bacterial profiling by deep sequencing analysis of 16sRNA
Cecum was collected sterilely, and the V34 region of 16S rRNA gene was amplified by PCR 59
Pall Nanosep Omega 3-kDa ultrafiltration tubes (VWR International) used for sample preparation were first prerinsed with deionized water to remove residual additives and background contamination. Tubes and filters were rinsed with 500 μl of deionized water, which were subsequently centrifuged at 10,000g for 5 min using Eppendorf 5430 (VWR International) to remove residual water. A mixture of 70:30 (MeOH:H 2 O) containing 40 μM of F-Phe and Cl-Try, as well as 2 mM 13 C-glucose, was prechilled on ice and used as internal or recovery standards for metabolite quantification.
All frozen cecum, colon and serum samples were slowly thawed on ice before sample pretreatment steps. In total, 100 μl of the above mixture containing 40 μM standards was added to the tubes containing preweighted cecum samples. The samples were then vortexed 10 min at room temperature at 3,000 rpm. Subsequently, the cecum samples were centrifuged at room temperature at 10,000g for 15 min and the supernatants were saved. The above-mentioned step was repeated, and the supernatants from both steps were combined. Subsequently, the supernatants were filtered through prerinsed ultrafiltration tubes at 10,000g for 15 min. Colon samples followed the similar sample pretreatment steps as cecum, but the ultrafiltration step was excluded. Serum samples were diluted fourfold. Briefly, 20 μl of the internal standard mixture containing 200 μM of F-Phe, Cl-Try and 2 mM 13 C-glucose in water and 55 μl of water was added to the thawed 25-μl serum samples. The resulting mixtures were vortexed for 5 s at room temperature at 3,000 rpm. Subsequently, the mixtures were transferred to prerinsed ultrafiltration tubes and centrifuged at 10,000 g for 15 min. All extracted samples were frozen at −80 °C until analysis, and a 20 μl aliquot was transferred into a polypropylene vial for CE-TOF-MS analysis 61 . Lung samples were processed by adding 120 μl of chloroform and 200 μl of the mixture (50:50, methanol:water) containing 5 μM recovery standards (F-phenylalanine, choline-d9) to the vials containing ~5 mg of freeze-dried lung samples. After shaking for 15 min at room temperature at 3,000 rpm, samples were centrifuged at 3000g at 4 °C for 15 min to sediment protein at the bottom of the vial followed by a biphasic chloroform and water/methanol (top) layer. A fixed volume (150 μl) was collected from the upper aqueous layer into a new vial. The above-mentioned step was repeated, and the combined upper aqueous layer was collected into a new vial. Combined upper aqueous layers were then dried under a gentle stream of nitrogen gas at room temperature. Lung extracts were then stored at −80 °C and before analysis reconstituted in 50 μl of water/methanol (70:30) with 40-μM chloro-tyrosine and naphthalene monosulfonic acid (internal standards).

Measurement of LPS in serum samples
Pierce Chromogenic Endotoxin Quant Kit from Thermo Fisher Scientific was used to measure the LPS levels in serum samples.

Prediction of functional profiles of the bacterial communities
PICRUSt2 was used to infer the functional metagenomic contents of each sample (in the unrarefied OTU table). For quality assurance purposes, Nearest Sequenced Taxon Index scores were examined, and they were <0.15 for all samples. Differential abundance of the predicted metagenomes between the experimental groups was analyzed using the KO (K identifiers in KEGG pathway maps) metagenome unstratified predictions obtained from PICRUSt2. Metagenomes Nature Immunology Article https://doi.org/10.1038/s41590-022-01354-4 with the predicted abundance of 0 in at least 10 of 11 samples were excluded from further analysis. Additionally, only metagenomes with the predicted abundance of at least ten in at least eight samples were included in further analysis. The differential abundance analysis was performed using limma package based on the limma-voom approach 62 . Specifically, the abundance values were rounded and then underwent VOOM transformation and TMM normalization before the differential analysis. Metagenomes exhibiting adjusted P value of <0.05 were considered to be significantly regulated between the groups, and KEGG Orthology database (https://www.genome.jp/kegg/ko.html) was used to annotate and categorize these terms, using the second and the third levels of the database, including terms from BRITE hierarchy, and pathways.

Quantification and statistical analysis
Statistical parameters including the exact value of n, the definition of center, dispersion and precision measures and statistical significance are reported in Figs. 1-8. The same samples were not repeatedly measured, and no data points were excluded from the analysis. All analyses were performed by using GraphPad Prism software (version 9.3.1, GraphPad Software). The confidence interval was set at 95% for statistical analysis. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in the previous publications 10,15,29 . Data distribution was assumed to be normal, but this was not formally tested. Data collection and analysis, except histological analysis, were not performed blind to the conditions of the experiments.
For RNAseq analysis, the reads were filtered by quality (at least 90% of the bases must have a quality score of 20 and higher) and then the remaining reads were aligned with mm10 (UCSC) reference using HISAT2. Next, the reads were counted by using HTSeq count. Genes, showing less than 10 counts in more than 30% of the samples per group, were removed using filterByExpr function in EdgeR package in R, resulting in 11,697 genes. Counts for these remaining genes were normalized with the TMM normalization method and then transformed using voom transformation. Differential expression between the groups of interest was examined using limma package in R. P values were corrected with BH correction for multiple testings 63 ; corrected values of <0.05 were considered to be significant. PCA was performed, and the results were visualized using rgl package in R (https://cran.r-project.org/web/package/rgl/index.html). Heatmaps were obtained using gplots package in R (https://cran.r-project.org/web/packages/gplots/index.html) and Volcano plots were obtained using limma package.
For microbiome analysis, reads were processed using DADA2. First, Cutadapt was used to filter and trim adaptor sequences and PCR primers from the raw reads with a minimum quality score of 30 and a minimum read length of 100 bp (ref. 64 ). Sequence variants were then resolved from the trimmed raw reads using DADA2, an accurate sample inference pipeline from 16S amplicon data. DNA sequence reads were filtered and trimmed based on the quality of the reads for each Illumina run separately, error rates were learned and sequence variants were determined by DADA2. Sequence variant tables were merged to combine all information from separate Illumina runs. Bimeras were removed, and taxonomy was assigned using the SILVA database version 1.3.2. The most abundant bacterial taxa were recognized at the genus level. Comprehensive statistical analysis of the microbiome was performed using a web-based platform MicrobiomeAnalyst (www. microbiomeanalyst.ca) 65 . Data filtering was performed using the criteria that feature containing all zeros or appearing in only one sample was excluded from the analysis. Data were total sum scaled (count data are divided by the total number of reads in each sample). Community diversity was profiled as "alpha diversity" and "beta-diversity". Differentially abundance of operational taxonomical units at the genus level was identified using DESeq2. The statistical significance level was set at 5%. P values were calculated and adjusted by the FDR.
For metabolite analysis, raw CE-TOF-MS data (d format) were processed using MassHunter Workstation Qualitative Analysis software (version B.06.00, Agilent Technologies, 2012). A comprehensive study of all detectable molecular features from the raw data was performed using MassHunter Molecular Feature Extractor, Molecular Formula Generator tools and an in-house compound database. Molecular features were extracted using a symmetric 10 ppm mass window, and all ions were annotated using their accurate mass (m/z), relative migration time (RMT) normalized to an internal standard (Cl-Tyr), and ionization mode of detection (p, positive; n, negative). RMTs are reported because they are an important parameter used to exclude redundant adducts and/or fragment ion peaks, which exhibit identical RMTs as the parent compound. Peak smoothing was performed using a quadratic/cubic Savitzky-Golay function (7 points) before peak integration. Peak areas and migration times for all molecular features were transferred to an Excel worksheet (Microsoft Office), and relative peak areas (RPA) for each unique molecular feature were saved as.csv file. RPAs were used for all statistical analyses. Pathway analysis (targeted) and multivariate data analysis, including PLS-DA, were performed using Metaboanalyst 5.0 (www.metaboanalyst.ca). In all cases, missing values were replaced with the default setting (one-fifth of the lowest detected value) and metabolomic data sets were (generalized) log-transformed and autoscaled unless otherwise stated.

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