Oxygen-responsive bacterial glycosphingolipid links symbiont fitness and immune development in neonatal host

Summary

Symbiotic gut bacteria must re-establish themselves in every host generation, yet the molecular strategies enabling this inheritance remain poorly understood. Here we show that Bacteroides fragilis uses a membrane glycolipid, alpha-galactosylceramide (BfaGC), to colonize the neonatal gut. Genome-wide fitness profiling revealed that BfaGC biosynthesis is selectively required during early life, when transient oxygenation creates a physiological bottleneck for strict anaerobes. Mechanistically, BfaGC reduces membrane proton permeability, sustaining the proton-motive force that supports aerobic respiration. This oxygen-responsive adaptation simultaneously generates a host-facing immunomodulatory signal that calibrates neonatal NKT cell development, linking bacterial fitness to immune maturation through a single metabolite. The same mechanism also enables niche expansion by enterotoxigenic strains, revealing context-dependent consequences. Notably, this strategy is distinct among gut Bacteroidales: other prominent members synthesize a different sphingolipid subclass supporting broader fitness, implying divergent evolutionary strategies. Our findings provide time-resolved insight into how bacterial metabolites shape host-microbiota symbiosis across development.

Graphical Abstract

In Brief

How do maternal gut bacteria seed the rapidly developing neonatal gut? Heo et al. show that Bacteroides fragilis produces a distinct membrane lipid that enables this strict anaerobe to tolerate a microaerobic environment. This finding highlights divergent evolutionary strategies among related species, most of which rely on lipids conferring broader fitness.

Introduction

Symbiotic microbiota have co-evolved with mammals, shaping host metabolism, immune development, and physiological homeostasis.^1–4 Yet this partnership must be re-established in every generation, as maternal microbes seed an initially uncolonized niche and assemble into a nascent community that matures in parallel with host development.^5–7 Despite its stability, the underlying mechanisms that preserves its continuity across generations remain poorly understood.

Birth and early postnatal life impose one of the most stringent ecological bottlenecks in microbial inheritance. The gastrointestinal tract, where the symbiotic microbiome resides at the highest density and diversity, faces dramatic and continuous shifts in oxygen level, nutrient composition, and immune activities, immediately after birth. Only microorganisms that can sense and adapt to these dynamic physiological changes can establish durable colonization. For the host, the same time window determines lifelong metabolic and immunological trajectories.^8–10 For the first 100 weeks of human life, dynamic restructuring of the infant gut microbiota over several taxa, such as Bifidobacterium and Bacteroides, has been well described.^11,12 Although the initial microbial community is strongly influenced by delivery mode,⁸ subsequent microbiota maturation is further shaped by human-to-human transmission after birth,¹³ host physiology, and dietary transitions throughout early life.⁵ Nevertheless, the specific molecular effectors and mechanisms that drive these developmental transitions remain poorly understood.

At this critical developmental juncture, Bacteroides fragilis, a prominent human symbiont, displays a distinct colonization pattern in early life that distinguishes it from other prevalent Bacteroides species.¹² To identify species-specific molecular effectors that make this species unique, we performed genome-wide fitness profiling during neonatal colonization. This analysis uncovered an oxygen-responsive glycolipid, alpha-galactosylceramide (BfaGC), that connects key biophysical properties of bacterial membranes to adaptation within the microaerobic neonatal gut. Notably, BfaGC biosynthesis peaks during early life, coinciding with a critical window of host mucosal immune development. Our findings reveal a species-specific strategy through which a microbial metabolite coordinates bacterial fitness with host developmental physiology, providing a molecular framework for understanding how symbiotic continuity is maintained across generation.

Results

Genetic determinants of bacterial colonization in the neonatal gut

Longitudinal human microbiome studies reveal that Bacteroides fragilis exhibits a distinct early-life temporal pattern within the infant Bacteroidales community, with a transient yet significant enrichment during early life (Fig. 1A).^12,14,15 This observation was reproducible in gnotobiotic mice, harboring a defined 10-member Bacteroidales consortium (Fig 1B). After initial colonization, B. fragilis abundance in the adult gut was rapidly decreased and stabilized at ~2%; when transmitted to offspring, B. fragilis abundance expanded significantly during the pre-weaning stage (up to ~40% at postnatal day 14), suggesting that this species, but not other Bacteroidales, possesses host age-specific fitness traits uniquely tuned to the neonatal environment.

Fig. 1. Genetic determinants of Bacteroides fragilis colonization in the neonatal gut.

(A) Longitudinal relative abundance of the order Bacteroidales, Bacteroides fragilis, and B. fragilis abundance within Bacteroidales in the gut microbiome of human infants from 3 cohorts. (B) Longitudinal relative abundance of a defined consortium of 10 Bacteroidales species (B. fragilis, Bacteroides caccae, Bacteroides thetaiotaomicron, Bacteroides uniformis, Bacteroides stercoris, Phocaeicola vulgatus, Phocaeicola dorei, Bacteroides ovatus, Segatella copri, Parabacteroides distasonis), colonized in germ-free Swiss-webster mice. Numbers above bars indicate mean B. fragilis relative abundance, with SD shown in gray. Data represent mean of n=6–7 mice per time point. (C) Transposon sequencing (Tn-seq) workflow for identifying genes essential for colonization. A high-density Tn-mutant library of B. fragilis (~160,000 mutants) was introduced into germ-free pregnant mice. The initial mutant pool and populations recovered from dam and pup mice were then analyzed via insertion sequencing. Genes with significantly decreased insertion frequency under a given condition were classified as beneficial for colonization in that environment. (D) Classification of beneficial genes based on Tn-seq revealed condition-specific requirements for B. fragilis colonization in the adult and neonatal gut. (E) KEGG pathway enrichment analysis highlighting distinct metabolic and stress-response pathways engaged at each developmental stage. (F) Identification of cydAB, encoding cytochrome bd oxidase, as a key determinant of B. fragilis survival in the neonatal gut, further validated by competitive colonization between wild-type and its isogenic Dcyd mutant B. fragilis in neonatal and adult mice. Shown are the means and SD. ***p<0.005 by Student’s t test (F). See also Figures S1–S3 and Table S1 and S2.

To define the genetic determinants of early-life colonization, we conducted genome-wide transposon sequencing (Tn-seq) using a high-density B. fragilis mutant library (>160,000 isogenic mutants) (Figs. 1C and S1A).^16,17 The library was introduced into pregnant germ-free mice, enabling natural vertical transmission and stage-specific selection in vivo. Mutant coverage was extensive, with 80.8% of open reading frames disrupted under nutrient-rich conditions (Fig. S1B). Following in vivo selection, insertion diversity declined sharply—from 160,965 unique insertion sites in vitro to 43,816 in adults and 12,110 in neonates (Fig. S1C). This pattern revealed a pronounced genetic bottleneck during neonatal colonization. Using zero-inflated negative binomial models, we identified 963 and 973 genes required for B. fragilis fitness in the adult and neonatal gut, respectively (Figs. 1D and S1D; Table S1). In adults, enriched pathways included carbohydrate and amino acid metabolism, consistent with nutrient autonomy as a key fitness trait. O-antigen biosynthesis, particularly polysaccharide A (PSA) synthesis, highlighted a dual role of extracellular polysaccharides in bacterial survival and host interaction (Figs. 1E and S2; Table S2).¹⁸

Among the enriched pathways in the neonatal gut (Fig. 1E), oxidative phosphorylation was prominently represented, including cydAB, which encodes cytochrome bd oxidase, a terminal oxidase in aerobic respiration. Consistent with the Tn-seq result, a competition assay confirmed that cydAB was indispensable for neonatal colonization (Fig. 1F).^19,20 Deletion of cydAB caused a severe colonization defect (competitive index = 0.038 in neonates vs. 0.337 in adults), revealing that B. fragilis, although a strict anaerobe, transiently exploits oxygen-dependent metabolism during early colonization. While cytochrome bd-mediated respiration supports adaptation near the intestinal epithelium in adults,^21,22 its essentiality in neonates underscores a heightened reliance on this pathway before full anaerobiosis is established.²³ The conservation of cydAB across Bacteroidales species further emphasizes its role as a core adaptation to fluctuating oxygen environments.²⁰ These respiratory adaptations coincided with enrichment of oxidative-stress response genes, as six of nine reactive oxygen species (ROS)-scavenging enzyme mutants exhibited neonatal-specific fitness defects (Fig. S3A), underscoring the central role of aerotolerance in early colonization.^20,24 In addition, glyoxylate/dicarboxylate and propanoate metabolism were enriched (Figs. 1E and S3B), suggesting functions in NADH generation and redox balancing under transient oxygen exposure, despite the absence of a canonical glyoxylate shunt.^25–27 Collectively, these findings show that B. fragilis orchestrates aerobic respiration, oxidative-stress defenses, and auxiliary metabolic circuits to endure transient oxygenation in the neonatal gut, thereby revealing a multifaceted strategy for early-life colonization success.

BfaGC is a necessary and sufficient fitness factor of early gut colonization, enhancing aerotolerance of B. fragilis

To identify B. fragilis-specific determinants of neonatal colonization, we compared Tn-seq hits across ten Bacteroidales genomes used in the colonization experiment (Figs. 1B and 2A; Table S3). Of 575 B. fragilis-specific genes (BLAST search, >80% query converage, E-value < 1e⁻²⁰; ~12.1% of the genome), 98 were essential for colonization in neonates. Given the transiently microaerobic environment of the neonatal gut, we next focused on neonatal fitness genes responsive to oxygen exposure. Cross-referencing our Tn-seq datasets with published transcriptomes (Table S1)²⁶ revealed only three of 98 neonatal-specific genes were upregulated by oxygen, notably including agcT (BF638R_3264), encoding a ceramide alpha-galactosyltransferase^28–30 responsible for the biosynthesis of the unique alpha-galactosylceramide (BfaGC) of B. fragilis. Sphingolipids constitute major membrane lipids in Bacteroidales, required for membrane integrity and stress adaptation in the gut.³¹ Notably, core ceramide biosynthetic genes (spt, cerR, and cerS) are essential for adult colonization (Fig. 2B),³² whereas agcT is uniquely required in neonates, indicating that ceramide galactosylation to form aGC confers a context-specific advantage beyond the structural function of its precursor.

Fig. 2. BfaGC enhances B. fragilis colonization in the neonatal gut by promoting aerotolerance.

(A) Among B. fragilis-specific genes, those beneficial in the neonatal gut are indicated in pale red, and a subset previously shown to be oxygen-induced is marked in dark brown. agcT, encoding ceramide alpha-galactosyltransferase, is one of only three genes belonging to both categories. (B) Relative fitness of transposon-insertion mutant of core ceramide synthesis genes (spt, cerR, cerS) and agcT was determined compared to WT in both adult and neonatal gut. (C) Germ-free Swiss-Webster mice were orally gavaged with a 1:1 mixture of wild-type and DagcT mutant B. fragilis, and bacterial abundances were measured in neonatal and adult mice at the indicated time points. For postnatal colonization, germ-free neonates were gavaged with the mixture at postnatal day 0 (p0) or p7, and bacterial abundance were measured 1 week after colonization. (D) Competitive colonization assay of Bacteroides thetaiotaomicron strains lacking (wildtype; agcT⁻), or overexpressing agcT (agcT⁺). Competitive fitness of agcT was calculated by denominating the abundance of agcT⁺ strain by that of agcT⁻ strain. (E) Germ-free pregnant mice were co-colonized with a 1:1 mixture of wildtype B. fragilis and its DagcT mutant, in the presence or absence of either wildtype E. coli or an E. coli DhemA mutant. Shown are the means and SD. *p<0.05, ***p<0.005 by one-way ANOVA (C and E) or by Student’s t test (D). See also Figures S4–S8 and Table S3.

To determine the functional importance of BfaGC, we co-introduced a 1:1 mixture of wild-type and ΔagcT B. fragilis into germ-free mice (Fig. 2C). Deletion of agcT abolished BfaGC synthesis without affecting bacterial growth or upstream ceramide production (Figs. S4A and S4B). While both strains persisted comparably in adults, the ΔagcT mutant exhibited a sevenfold reduction in neonatal colonization following natural vertical transmission, consistent with Tn-seq predictions. This defect was evident when colonization was initiated postnatally at day 0 (p0), but less significant when colonization was initiated at p7, indicating that BfaGC is specifically required during the earliest neonatal window of colonization. In parallel, we engineered an aGC-deficient species, Bacteroides thetaiotaomicron, to heterologously express agcT to produce aGC (Figs. S4C and S4D). In competitive colonization assays, the agcT-expressing strain consistently outcompeted its parental counterpart in neonatal mice, demonstrating that BfaGC production alone enhances early-life colonization (Figs. 2D and S4E).

Previous transcriptomic analyses indicated that agcT expression increases upon oxygen exposure.²⁶ Given the enrichment of oxidative-stress response pathways during neonatal colonization and the established role of sphingolipids in stress adaptation,³¹ we hypothesized that BfaGC enhances aerotolerance of B. fragilis. Consistent with this, both agcT expression and BfaGC levels increased after short-term oxygen exposure in vitro (Fig. S5A). While wild-type and ΔagcT strains grew similarly under anaerobic and microaerobic conditions, ΔagcT strains displayed markedly reduced viability following acute aerobic challenge, which was fully restored by agcT complementation (Figs. S6A and S6B). Likewise, in a 1:1 in vitro competition assay, WT and ΔagcT showed no fitness difference under anaerobic conditions, whereas ΔagcT displayed a marked competitive defect during aerobic exposure (Fig. S6C). In contrast, wild-type and ΔagcT strains showed no consistent difference across other stress conditions, including starvation, low pH, antibiotics, and membrane stress (Fig. S7). These results demonstrate that BfaGC primarily confers oxygen tolerance, supporting B. fragilis fitness under early-life oxygen stress.

To test whether BfaGC promotes early gut colonization by enhancing adaptation to transient oxygen exposure in vivo, we conducted co-colonization assays with Escherichia coli, a facultative anaerobe that consumes oxygen in the gut (Figs. 2E and S8). The presence of E. coli attenuated the fitness difference between wild-type and ΔagcT strains while maintaining the total B. fragilis population. This effect was lost when co-colonized with E. coli ΔhemA, a strain deficient in heme biosynthesis and incapable of aerobic respiration,³³ indicating that BfaGC-mediated fitness advantage depends on environmental oxygen levels. Together, these results establish that BfaGC promotes B. fragilis persistence in the neonatal gut by enhancing tolerance to transient oxygen exposure.

BfaGC stabilizes membrane integrity to support aerobic respiration

The neonatal gut is a metabolically dynamic niche characterized by transient oxygen exposure and the predominance of facultative anaerobes. Given that ceramide constitutes a major structural lipid in Bacteroidales membranes, we hypothesized that its glycosylation into BfaGC remodels membrane properties to promote colonization in this partially oxygenated milieu. Because aerobic respiration depends on membrane-bound complexes, we investigated whether BfaGC contributes to respiratory adaptation and fitness. Oxygen consumption assays showed that the ΔagcT mutant exhibited reduced aerobic respiration relative to wild-type B. fragilis, whereas the Δcyd strain, lacking cytochrome bd oxidase, displayed no detectable oxygen consumption (Figs. 3A and S9A). Upon brief oxygen exposure, the ΔagcT mutant showed an increased ATP/ADP ratio and elevated acetate production, suggesting a continued reliance on low-yield fermentative metabolism despite oxygen availability (Figs. S9B and S9C). This overflow metabolism indicates insufficient coupling of metabolic flux into the electron transport chain (ETC).

Fig. 3. BfaGC enhances B. fragilis aerobic respiration by stabilizing membrane potential and proton retention.

(A) Normalized oxygen consumption rate of WT and DagcT under microaerobic condition. (B) Growth rates of WT and DagcT was measured in PYG medium in the presence of indicated amounts of CCCP under anaerobic and microaerobic condition. (C) Membrane potential measurements using the DiSC3(5) fluorescence assay. Bacterial cells were treated with DiSC₃(5) at 5 min and fluorescence was monitored for 40 min with excitation wavelength of 651 nm and emission wavelength of 675 nm. (D) Proton permeability was measured using small unilamellar vesicles (SUVs) composed of lipid extracts from wildtype, DagcT, and DagcT supplemented with synthetic BfaGC (2 ng or 5 ng). SUVs were loaded with the pH-sensitive fluorescent dye pyranine (2 mM) and equilibrated in a neutral buffer (pH 7.0). To induce proton influx, SUVs were transferred to an acidic buffer (pH 4.5), and intravesicular pH changes were monitored over 30 min. Fluorescence was measured at excitation wavelengths of 410 nm and 454 nm, with emission at 510 nm. Intravesicular pH values were calculated using a standard calibration curve based on the fluorescence ratio (410/454 nm). The proton influx rates were determined by performing linear regression on time-resolved values obtained through inverse logarithmic transformation of measured pH. (E) Competitive colonization assay of Dcyd and Dcyd DagcT B. fragilis in neonatal mice. Germ-free mice were co-colonized with either a 1:1 mixture of WT and DagcT or a mixture of Dcyd and Dcyd DagcT. Competitive fitness of the agcT gene was determined by calculating the ratio of DagcT to WT (DagcT/WT) or D cyd DagcT to Dcyd (Dcyd DagcT/Dcyd). Shown are the means and SD. *p<0.05, **p<0.001, ***p<0.005 by one-way ANOVA (B and E) or by Student’s t test (A and C). See also Figures S9–S11.

Tn-seq analysis suggested a reconfiguration of respiratory chains, with B. fragilis utilizing sodium-coupled complexes (Nqr, V-type) in adults but favoring proton-dependent systems (Nuo, F-type ATPase) in neonates (Figs. S10A and S10B).³⁴ Transcriptomic data support this transition, showing repression of nqr operon and concurrent induction of cytochrome bd under oxygen exposure (Fig. S10C).²⁶ Because BfaGC is oxygen-inducible, we hypothesized that it supports this energetic shift by enhancing proton retention and sustaining PMF. To test this, we challenged wild-type and ΔagcT strains with the protonophore CCCP. Under anaerobic conditions, both strains were insensitive, whereas under microaerobic conditions the ΔagcT mutant exhibited heightened CCCP sensitivity (Fig. 3B). Consistently, membrane potential was significantly reduced in the ΔagcT mutant relative to wild-type (Fig. 3C), identifying BfaGC as a key determinant of PMF maintenance. Although the PMF is generated by ETC, BfaGC did not alter ETC complex activity (Figs. S9D and S9E).

Efficient PMF generation requires membranes with low proton permeability, minimizing “proton hopping” via hydrogen-bonded water wires and thereby preventing energy dissipation.^35,36 Glycolipids such as digalactosyldiacylglycerol can reduce proton permeability.³⁷ Building on this principle, we tested whether BfaGC directly enhances proton exclusion. Small unilamellar vesicles (SUVs) prepared from wild-type and ΔagcT lipid extracts revealed markedly higher proton influx in ΔagcT-derived SUVs relative to wild-type SUVs (Fig. 3D). Incorporation of synthetic aGC into ΔagcT-derived liposomes restored membrane proton impermeability to wild-type levels, demonstrating that BfaGC directly reinforces membrane proton retention. This effect was recapitulated in fully synthetic lipid vesicles (Figs. S11A–C), confirming that physiological aGC/ceramide ratios are sufficient to modulate proton flux.

Finally, to determine whether the colonization advantage conferred by BfaGC depends on aerobic respiration, we compared the fitness of Δcyd and Δcyd ΔagcT mutants, both lacking cytochrome bd oxidase, under microaerobic conditions. The two strains exhibited comparable viability (Fig. S11D), indicating that the contribution of BfaGC to aerotolerance requires a functional respiratory chain. Consistently, in neonatal colonization assays, the ΔagcT mutant showed a marked fitness defect relative to wild-type, whereas no difference was observed between Δcyd and Δcyd ΔagcT strains (Fig. 3E). These results demonstrate that the benefit of BfaGC to neonatal colonization is contingent on respiratory metabolism, mechanistically linking membrane bioenergetics to oxygen adaptation.

Glycosylation likely alters the interfacial behavior of ceramide rather than simply increasing headgroup size. Because proton permeation across lipid bilayers is highly sensitive to interfacial hydration and packing,³⁸ BfaGC may strengthen headgroup organization in a manner that renders the membrane interface less permissive to proton leakage. This model is consistent with studies linking membrane glycosylation to altered bilayer packing and dynamics,^39,40 as well as reduced proton permeability in galactolipid-enriched membranes.³⁷

Aeroprotective benefit of BfaGC ties host-microbiota mutualism in the neonatal gut

Previous studies demonstrated that B. fragilis, but not a ΔagcT mutant lacking BfaGC, modulates colonic NKT cell homeostasis, consistent with oral administration of synthetic BfaGC recapitulating the same immunomodulatory effect.^30,41–43 This effect is confined to the neonatal period, coinciding with the transient window of oxygen exposure and peak B. fragilis colonization.⁴⁴ Given that BfaGC is also a well-characterized CD1d ligand regulating NKT-cell development and function, these findings suggest that neonatal oxygen availability may shape both bacterial fitness and the production of host-facing immunological signals.

To assess this hypothesis, we employed a stringent-response-deficient B. fragilis mutant (ppGpp⁰), which fails to activate oxidative stress-responsive transcriptional programs, including oxygen-induced BfaGC synthesis (Figs. S12A and S12B).^45–47 In monocolonized neonates, the ppGpp⁰ strain did not induce BfaGC production (Fig. 4A) despite achieving colonization levels equivalent to the wild type (Fig. 4B). Consistent with its defect in BfaGC induction, ppGpp⁰-colonized pups did not normalize colonic NKT cell frequencies, placing impaired neonatal BfaGC induction, rather than colonization per se, as the most direct correlate of the immune modulation (Fig. 4C).

Fig. 4. Oxygen-induced BfaGC coordinates neonatal immune imprinting and enables enterotoxigenic niche expansion.

(A) Quantification of aGC levels in wild-type and ppGpp⁰ mutant B. fragilis strains under neonatal (at postnatal day 7 and 14) and adult gut conditions (at 6- and 12-weeks old). (B) B. fragilis burden in the colon of neonates (postnatal day 14) and adults (week 6) monocolonized with either wild-type and ppGpp⁰ mutant strains. (C) Flow cytometry analysis of colonic NKT cell frequency (as a proportion of CD45⁺ cells) in germ-free (GF) mice and gnotobiotic mice monocolonized with either wild-type B. fragilis (WT) or ppGpp⁰ mutant at week 8. Shown are the means and SD. (D) The proportion of aGC to ceramide (C34:0) was measured by LC-MS analysis of samples collected from seven regions of the GI tract in B. fragilis-monocolonized mice. (E) The competitive fitness of the agcT gene across the seven regions was determined through co-colonization experiments with wild-type B. fragilis and its isogenic DagcT mutant. (F) Experimental scheme of enterotoxigenic B. fragilis (ETBF) infection. Mice were pretreated with an antibiotic cocktail (vancomycin, gentamicin, and ampicillin) in drinking water from day −4 for 3 days, followed by gentamicin alone on day −1. On day 0, mice were orally gavaged with 10⁹ cells of B. fragilis strains; vector control (WT/pNBU2), Bft-expressing wild type (WT/Bft), DagcT/Bft, or Dcyd/Bft, and colon and cecal samples were collected on day 7. (G) B. fragilis burden in the colon at day 7. (H) Colon length measured at day 7 post-infection. Shown are the means and SD. *p<0.05, **p<0.001, ***p<0.005 by two-way ANOVA (A), or by one-way ANOVA (C, G, and H). See also Figure S12.

These results, along with the characteristic abundance of B. fragilis in neonates (Fig. 1A), imply that the host gut is exposed to significantly higher levels of BfaGC in early life. The neonatal stage is the critical window for colonic NKT cell proliferation, temporally regulated by BfaGC.^42,43 Transient induction of BfaGC during the neonatal period may calibrate NKT numbers, by regulating their proliferative nature. In this view, BfaGC functions not only as a metabolic adaptation but also as a commensal developmental cue that shapes early immune maturation.

BfaGC-dependent aerotolerance enables niche acquisition of B. fragilis at both physiological and pathological conditions in adult hosts.

Besides its community-level adaptation synchronized with the host development, gastrointestinal microbiota also shows significant differences caused by local oxygen levels, in both physiological (e.g., specific locations in the GI tract)^48,49 and pathological (e.g., infection or inflammation)^50,51 contexts. We further examined whether BfaGC can provide fitness in spatially oxygenated niches, beyond early life. Similar to the early gut, BfaGC production is significantly induced in the small intestine than the cecum and colon (Fig. 4D). In accordance, BfaGC-dependent fitness peaks in the ileum (Fig. 4E). This spatial signature indicates that the BfaGC enables B. fragilis to adapt to localized oxygenated niches.

In parallel to its mutualistic natures to the host, B. fragilis also harbors an enterotoxigenic lineage (ETBF), arising through horizontal acquisition of bft (a zinc-dependent metalloprotease).⁵² Such horizontal transfer events recur only within B. fragilis among Bacteroidales.^52,53 Bft expression enables B. fragilis to invade the lamina propria during host development.⁵⁴ Given that host tissue is more oxygenated than the gut lumen, and most ETBF isolates retain BfaGC (Fig. S12C), species-specific aerobic adaptation may have predisposed B. fragilis to the emergence of enterotoxigenic lineages. To validate this, antibiotics-pretreated mice were orally inoculated with B. fragilis strains: wild-type harboring vector (WT/pNBU2; vector control), Bft-expressing wild-type strain (WT/Bft), and Bft-expressing mutants lacking either agcT (ΔagcT/Bft) or cyd (Δcyd/Bft) (Fig. 4F). Quantitative culture analysis of luminal contents showed that Bft expression increased bacterial burden in both colon and cecum; however, this colonization advantage was abolished in both ΔagcT/Bft and Δcyd/Bft strains (Figs. 4G and S12D). Bft expression levels were comparable across strains, confirming that the colonization differences are metabolic rather than expression-driven (Fig. S12E). Consistent with this colonization pattern, WT/Bft infection was accompanied by intestinal shortening relative to the vector control and the phenotype was attenuated in both mutant backgrounds (Fig. 4H). Together, these results identify BfaGC-dependent aerobic respiration as the basis for niche expansion in adult host, in parallel with its role in early life.

Divergent sphingolipid biosynthesis pathways allow specialized or universal fitness strategies among Bacteroidales

Sphingolipid biosynthesis is a conserved feature of Bacteroidales, yet only a subset of species, including B. fragilis, produce aGC.³⁰ Most others instead synthesize phosphoinositol dihydroceramide (PI-Cer), a chemically distinct glycosphingolipid.⁵⁵ Our 10-species synthetic microbiome also shows that B. fragilis is the only species bearing agcT (hence becomes the sole producer of aGC), where the other 9 species commonly produce PI-Cer (Fig. 5A). This phylogenetic split within a single bacterial order poses a key evolutionary question: what selective pressure preserved aGC in a few lineages while favoring PI-Cer across the majority?

Fig. 5. BfaGC mediates temporal niche acquisition of Bacteroidales in the neonatal gut.

(A) The synthesis of ceramide derivatives across different Bacteroidales species. The synthesis of alpha-galactosylceramide (aGC) is restricted to B. fragilis, whereas other species predominantly synthesize phosphoinositol dihydroceramide (PI-Cer). Bacteroides salyersiae uniquely synthesizes both aGC and PI-Cer. (B) Competition colonization assay between wildtype B. thetaiotaomicron and PI-Cer-deficient mutant (DPICS) strains in adult mice. Ther abundance was monitored for 10 days after co-colonization by measuring colony-forming units (CFUs). (C) Extracted ion chromatograms (XICs) of Ceramide (34:0), aGC (34:0), and PI-Cer (34:0) from B. salyersiae wild-type (WT), DagcT, and DPICS strains. WT produces both aGC and PI-Cer, whereas DagcT lacks aGC and DPICS lacks PI-Cer. (D) Schematic of tripartite competition assay comparing B. salyersiae WT (aGC⁺/PI-Cer⁺), DagcT (aGC⁻/PI-Cer⁺), and DPICS (aGC⁺/PI-Cer⁻) under distinct host context. (E) Relative abundance of each strain in adult and neonatal guts, quantified by qRT-PCR using strain-specific primers. (F) Aggregate abundance of aGC⁺ strain (WT+DPICS) was compared between adult (week 1) and neonates (p7) (right). (G) aGC abundance was compared between WT B. salyersiae and its isogenic DPICS bacterial culture. (H) Growth of B. salyersiae WT, DagcT, and DPICS strains in the presence of increasing concentrations of colistin. Growth was measured as final OD600 relative to untreated controls. (I) Human microbiome proxy analysis of age-dependent sphingolipid-associated patterns. Species carrying agcT were used as a proxy for aGC-producing taxa, and species carrying PICS-associated genes were used as a proxy for PI-Cer-associated taxa. (J) comparative genomic survey of sphingolipid biosynthetic pathways across Bacteroidales. SPT-positive genomes were classified based on the presence of absence of agcT and PICS-associated genes. Shown are the means and SD. *p<0.05, **p<0.001 by Student’s t test (F and G). See also Figures. S13 and S14.

To directly probe the selective pressures governing aGC and PI-Cer biosynthesis, we compared their fitness contributions across distinct gut environments. In the adult gut, B. thetaiotaomicron lacking PI-Cer (PI-Cer synthase-deficient ΔPICS) exhibited a marked fitness defect (Fig. 5B), whereas B. fragilis ΔagcT remained unaffected (Fig. 2A), indicating that PI-Cer supports long-term persistence under stable anaerobic conditions, serving as a generalist lipid. Conversely, the ΔagcT mutant displayed a pronounced colonization defect in neonates, confirming that aGC functions as a temporal specialist required for early-life colonization. Together, these complementary roles suggest that Bacteroidales functionally partition sphingolipid utilization across developmental time: the stable adult gut favors PI-Cer-mediated homeostatic persistence, whereas the transient neonatal environment selects for aGC-driven niche acquisition.

To further dissect this trade-off within a single native context, we employed two independent systems to compare the fitness provided by different sphingolipids. Firstly, we directly compared the fitness of PI-Cer- and aGC-producing isogenic strains in B. thetaiotaomicron, using BtWT (PI-Cer⁺/aGC⁻), BtΔPICS (PI-Cer⁻/aGC⁻), and BtΔPICS/agcT (PI-Cer⁻/aGC⁺) strains (Fig. S13). Consistent with the necessity of PI-Cer (Fig. 5B), BtWT quickly dominated the adult gut, across the ileum, cecum, proximal and distal colon. Of note, expansion of aGC-producing strain was observed even in the absence of native PI-Cer, confirming the role of aGC as a fitness factor in early life. Secondly, we also generated an isogenic single-gene knockouts of Bacteroides salyersiae (Bs), which naturally produces both aGC and PI-Cer (Figs. 5A, 5C, and S14A).

Tripartite competition comparable to that of B. thetaiotaomicron system with a single-gene deletions has been established, using BsWT (PI-Cer⁺/aGC⁺), BsΔagcT (PI-Cer⁺/aGC⁻), and BsΔPICS (PI-Cer⁻/aGC⁺) (Figs. 5D, 5E, and S14B). In the adult gut, BsΔagcT strain rapidly outcompeted both BsWT and BsΔPICS, demonstrating a clear selective advantage for PI-Cer under anaerobic conditions as well as dispensability (or, surprisingly, negative fitness) of aGC. In contrast, under neonatal hosts, both aGC-producing strains significantly expanded from the baseline abundance in dams, supporting aGC-mediated competitive advantage during early colonization, even though BsΔagcT remained the dominant strain overall. Notably, the relative increase in BsΔPICS abundance was larger than that of BsWT during neonatal transition (10.6 fold vs 2.9 fold, Fig. 5F), which is possibly related to its higher aGC production (Fig. 5G). The earlier peak abundance of aGC-producing strains (before p7) in the B. salyersiae model likely reflects its simpler competitive context, which directly resolves sphingolipid-dependent fitness, whereas the timing of B. fragilis expansion in the 10-member consortium (peaked at p14) is shaped by additional neonatal fitness traits beyond aGC. These findings indicate that aGC and PI-Cer confer stage-specific fitness advantages, reflecting divergent selective pressures in neonatal versus adult gut environments. Moreover, direct comparison between wild-type and PI-Cer-only, and aGC-only strains suggests reciprocal interference between these pathways: aGC diminishes PI-Cer-mediated persistence in the adult gut, whereas PI-Cer may partially constrain aGC-associated fitness during early-life colonization. This provides a mechanistic explanation for the evolutionary segregation of these sphingolipid pathways among Bacteroidales.

To further probe this distinction, we compared WT, ΔagcT, and ΔPICS under additional stress conditions and found that ΔPICS showed increased susceptibility to colistin, an antimicrobial peptide (Fig. 5H). This is consistent with prior work linking inositol lipid deficiency in Bacteroides to altered antimicrobial peptide resistance and membrane-associated traits.⁵⁵ On the other hand, PI-Cer-deficient liposome did not show a detectable change in proton permeability (Fig. S11D), suggesting that PI-Cer supports a membrane function distinct from the oxygen-responsive, proton-retentive role of aGC. In line with this distinction, fecal lipid analysis from wild-type B. salyersiae-monocolonized mice showed induction of aGC early in life, whereas PI-Cer became relatively more abundant later (Fig. S14C).

A similar pattern was evident in proxy analysis in the human infant microbiome data. agcT-harboring Bacteroidales species were enriched earlier in life and declined over time, whereas PICS-associated taxa increased progressively with age (Fig. 5I). This age-dependent asymmetry was reinforced at the genomic level across Bacteroidales (Fig. 5J), where the broad distribution of PI-Cer-associated pathways contrasted with the restriction of agcT to a much smaller subset of genomes, consistent with PI-Cer as a broadly retained membrane component and aGC as a more specialized component to early-life. These findings collectively show that although aGC and PI-Cer represent alternative solutions to distinct ecological pressures, this divergence reflects a broader principle: host-associated microbes evolve molecular specializations that track the temporal and spatial architecture of the gut environment.

Discussion

Often claimed as ‘the second genome’, bacterial symbionts must sustain inheritability across generations, by enduring rapid physiological shifts and unexpected pathological challenges in their host. Our study uncovers previously unrecognized strategies that enable Bacteroides fragilis to thrive in the gut at early host life. Although strict anaerobes are classically thought to rely solely on fermentation or anaerobic respiration, we show that B. fragilis can activate oxygen-dependent respiration. A central discovery of this work is the identification of BfaGC as a distinct molecular component of such transition. BfaGC, previously identified as a host immunomodulatory factor, orchestrates this bioenegetic shift: from sodium-motive force (SMF)-driven respiration to PMF-driven respiration, optimizing energy efficiency under limited resources.

Because proton-based respiration is inherently prone to leak-driven energy loss, efficient PMF maintenance requires a proton-tight membrane.^35,56 BfaGC stabilizes membrane potential and minimizes proton leakage, thereby sustaining PMF and enhancing aerobic respiratory efficiency.

More broadly, oxygen-responsive remodeling of membrane architecture appears to represent a conserved strategy for sustaining bacterial bioenergetics, under changing environment, including cardiolipin-mediated rearrangement of respiratory complexes,⁵⁷ viscosity-driven acceleration of electron flow in E. coli,⁵⁸ and oxygen-dependent intracytoplasmic membrane biogenesis in purple bacteria.⁵⁹ The biophysical role of BfaGC fits this paradigm, underscoring the broader principle that membrane composition dynamically tunes bacterial bioenergetics in response to environmental stress. This observation is consistent with broader evolutionary models proposing that shifts in environmental oxygen availability influenced transitions between sodium- and proton-dependent bioenergetic systems.

In parallel, our previous and current findings further recognize BfaGC not only as an oxygen-responsive membrane adaptation that promotes neonatal colonization, but also as a developmental signal that links bacterial physiological state to colonic NKT cell maturation. The oxygen-induced nature of this pathway suggests that microbial adaptation to the neonatal gut environment can directly influence the abundance of a host-facing immunomodulatory lipid. This coordination provides a molecular example of developmental host-microbe alignment, in which a microbial fitness trait coincides with a key window of host immune maturation.

On the other hand, for a persistent gut symbiont such as B. fragilis, exposure to oxygen is not confined to the neonatal period. Spatial profiling of GI tract suggests that BfaGC may also facilitate colonizing partially oxygenated niches in adults. While this trait likely supports lifelong presence of B. fragilis, it may have also contributed to the emergence of ETBF, a variant adapted to exploit oxygen-exposed mucosal environments.^52,54,60 These contrasting impacts highlight that adaptation is not inherently beneficial to the host, warranting further investigation into distinct spatiotemporal mechanisms.

Finally, aerotolerance conferred by BfaGC places B. fragilis in a unique competitive position among the Bacteroidales. While most members of this order produce PI-Cer to persist in the strictly anaerobic gut,⁵⁵ B. fragilis lacks this lipid despite sharing the common ceramide precursor. This divergence implies a selection between niche-specific ‘specialist’ and ‘generalist’ strategies. Our results with B. salyersiae support this view: constitutively producing both molecules incurs a fitness cost, suggesting an evolutionary cost-benefit trade-off. In conclusion, our findings highlight how a single lipid molecule orchestrates temporal niche acquisition, spatial persistence, and evolutionary diversification. Ultimately, this underscores a mechanism driving dynamic, yet stable, host-microbiota symbiosis.

Limitations of the Study

This study has several limitations. Firstly, our Tn-seq screen was designed to capture fitness determinants across maternal colonization, vertical transmission, and neonatal establishment. This design preserves the physiological route of early-life colonization, but necessarily introduces sequential in vivo selection. As a result, mutants depleted during adult colonization may be difficult to assess independently during the neonatal stage. This consideration is particularly relevant for genes with broad colonization effects, and neonatal Tn-seq scores for such genes should be interpreted in the context of prior adult selection.

Secondly, regarding quantification of the bacterial presence in the gastrointestinal tract, our analyses primarily quantified bacterial burdens in luminal contents from the colon and other locations. In ETBF experiments, this approach provides a direct measure of Bft-associated luminal expansion. Further spatial imaging studies are warranted to assess the contribution of BfaGC to the bacteria’s capacity to acquire the lamina propria niche during early development.

Resource availability

Lead Contact

Request for further information and resources should be directed to and will be fulfilled by the lead contact, Sungwhan F. Oh (soh2@bwh.harvard.edu).

Materials Availability

Materials and further information are available from the lead contact upon completing a materials transfer agreement.

Data and code Availability

Data files for transposon sequencing and bacterial lipidomics analysis are publicly available at the Harvard Dataverse using the following Unpublished Dataset Preview URL : https://dataverse.harvard.edu/previewurl.xhtml?token=59417da4-c3b3-48cb-b7ee-d0a8b073774f

STAR Methods

1. EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

Germ-free and gnotobiotic mouse experiments were performed using age-matched Swiss Webster mice maintained in the Harvard Center for Comparative Medicine under protocols approved (IS187–09). All animal procedures were conducted in accordance with institutional and national guidelines for the care and use of laboratory animals. Mice used in this study were 0–12 weeks of age, including neonatal pups. Animals were housed under a 12 h light/12 h dark cycle at 21 and 50% humidity. Germ-free mice were bred and maintained in inflatable plastic isolators. For gnotobiotic experiments, breeding pairs were orally gavaged with the indicated bacterial strains and maintained in either isolators or ISOcage Ps to generate monocolonized or co-colonized offspring for analysis. Sterility of germ-free and gnotobiotic isolators was routinely monitored by aerobic and anaerobic culture of stool samples.

Both male and female animals were used. Because most neonatal experiments were performed using litter-based designs to capture vertical transmission and early-life colonization, sex was not analyzed as a biological variable in this study. This may limit generalization of the findings across sexes.

Bacteria

The study used human gut-associated Bacteroidales strains, including Bacteroides fragilis NCTC9343 and 638R, Bacteroides thetaiotaomicron VPI-5482, Bacteroides salyersiae DSM-18765, Bacteroides caccae ATCC 43185, Bacteroides uniformis ATCC 8492, Bacteroides stercoris DSM 19555, Phocaeicola vulgatus ATCC 8482, Phocaeicola dorei DSM 17855, Bacteroides ovatus ATCC 8483, Prevotella copri DSM 18205, Parabacteroides distasonis ATCC 8503, and Escherichia coli MG1655, as indicated for individual experiments. Gene deletion and overexpression strains generated in this study included B. fragilis ΔcydAB, ppGpp⁰ mutants and Bft-expressing strains; B. thetaiotaomicron ΔPICS and agcT-expressing strains; B. salyersiae ΔagcT and ΔPICS strains; and E. coli ΔhemA setrain.

Frozen bacterial stocks were maintained at −80°C. For routine culture, bacteria were streaked on Brain Heart Infusion agar plates and single colonies were inoculated into deoxygenated peptone-yeast extract-glucose (PYG) broth in an anaerobic chamber. PYG medium contained 2% proteose peptone, 0.5% yeast extract, 0.5% NaCl, 0.5% glucose, 0.5% K₂HPO₄, 0.05% L-cysteine, 5 mg/L hemin, and 2.5 mg/L vitamin K₁. For metabolomics experiments bacteria were grown in minimal medium supplemented with the indicated carbon sources. Where indicated, antibiotics were used at the following concentration: ampicillin (100 μg/mL) for E. coli, erythromycin (10 μg/ml), gentamicin (20 μg/ml), and anhydrotetracycline at the indicated concentrations for Bacteroides strains. Unless otherwise specified, all Bacteroides cultures were grown under anaerobic condition at 37°C.

2. METHOD DETAILS

Construction of Bacteroides knockout/overexpression mutants

Primers used for this study were described in Supplementary table 4. ΔcydAB ((ΔBF9343_1854/1855) and ppGpp⁰ (ΔBF9343_0658, BF9343_2136) mutant B. fragilis, ΔPICS (ΔBT_1522) B. thetaiotaomicron, and ΔagcT or ΔPICS Bacteroides salyersiae were constructed using the pLGB13 suicide vector.⁶¹ For overexpression, the agcT gene, relA gene, or bft gene was cloned into the pNBU2_erm-TetR-P1T_DP-GH023 vector under either their native promoter or the TetR promoter (anhydrotetracycline-inducible).⁶²

Construction of transposon mutant library

The screen was performed in B. fragilis 638R because this strain is more amenable to large-scale genetic manipulation and transposon library construction. NCTC 9343 was used for targeted follow-up assays because it is a widely used experimental background for B. fragilis biology. Transposon mutagenesis was performed via conjugation between B. fragilis 638R recipient cells and Escherichia coli S17–1 λpir donor cells harboring the transposon-containing plasmid pSAM-Bt.⁶³ For this process, 2 mL of overnight-cultured B. fragilis was inoculated into 800 mL of fresh PYG medium, while the donor E. coli was grown in 50 mL of LB medium supplemented with ampicillin. The recipient and donor cells were cultured until reaching exponential growth phases (B. fragilis OD₆₀₀ 0.05–0.1, E. coli OD₆₀₀ ~0.5), and subsequently mixed at a ratio of approximately 10:1 based on optical density. The cell suspension was harvested by centrifugation at 8,000 × g for 5 minutes and resuspended in ~5 mL of fresh PYG medium. The resulting cell mixture was spotted onto BHI agar plates and incubated aerobically for 16 hours. Post-incubation, the spots were scraped off the plates, resuspended in 5 mL of PYG medium, and spread onto BHI agar plates containing erythromycin and gentamicin to counter-select for B. fragilis transconjugants harboring the transposon. The plates were incubated under anaerobic conditions for 2–3 days. To estimate the total size of the mutant library, aliquots of the conjugated cell suspension were serially diluted, plated on selective agar, and colony-forming units (CFUs) were enumerated. The total library size was determined based on CFU calculations. The resultant transconjugants were resuspended in PYG medium containing 15% glycerol and immediately stored at −80°C. For subsequent experiments, individual aliquots were thawed and cultured as needed.

In vivo selection of transposon mutant library

The transposon mutant library was cultured anaerobically in fresh PYG medium for 2 hours, corresponding to approximately 2–3 replication cycles, to minimize potential growth rate biases among mutants. The library was grown to an optical density (OD₆₀₀) of 1.0. Aliquots of 200 μL, containing approximately 2 × 10⁸ mutant cells, were prepared for oral gavage. Pregnant germ-free Swiss-Webster female mice were administered the mutant library via oral gavage at embryonic days 7–9 (E7–9). Fecal samples were collected from the adult female mice on the day of delivery, as well as from their offspring on postnatal day 10.

Transposon sequencing using INseq method

Following selection, genomic DNA was isolated from the mutant population. The extracted gDNA was digested with the restriction enzyme MmeI, which specifically cleaves both termini of the transposon cassette. The digestion reaction was carried out at 37°C for 2 hours and subsequently heat-inactivated at 65°C for 20 minutes. The digested DNA was resolved via agarose gel electrophoresis, and a fragment approximately 1.4 kb in size—corresponding to the transposon cassette—was excised based on a molecular size marker. DNA was purified from the gel using the Monarch Gel Extraction Kit, with final elution in nuclease-free water. Adaptor ligation was performed by incubating the purified DNA with a 0.5 μM adaptor containing sample-specific barcode, and T4 DNA ligase at 20°C for 2 hours. The ligated DNA fragments were PCR-amplified using Lib_5 and Lib_3 primers. The PCR protocol included an initial denaturation step at 98°C for 30 seconds, followed by as few amplification cycles as necessary to remain within the linear range of amplification (98°C for 10 seconds, 63°C for 10 seconds, and 72°C for 10 seconds; 20 cycles), with a final hold at 4°C. The amplified PCR products were purified from agarose gels, and their concentration and integrity were assessed using the Agilent TapeStation High Sensitivity D5000 assay. Sequencing libraries were prepared and analyzed using the Illumina NextSeq 500 platform.

Data Processing and Statistical Analysis

Quality control, adapter sequence trimming, and removal of transposon sequences from the reads were performed using Trimmomatic.⁶⁴ Processed reads were aligned to the B. fragilis 638R reference genome using Bowtie2 v2.4.2 with default parameters.⁶⁵ Mapped reads were quantified by FeatureCounts,⁶⁶ assigning read counts to corresponding open reading frames (ORFs). Differential abundance analysis of read counts was conducted using either DESeq2⁶⁷ or zero-inflated negative binomial (ZINB) models, depending on the experimental group. An enrichment analysis of functional terms for genes of interest was carried out using KEGG gene set enrichment analysis.⁶⁸

To identify essential genes for in vitro growth, insertion sites from sequencing data were mapped to annotated gene regions. Insertion events were categorized by matching their positions to the start and end coordinates of ORFs. Unmatched insertion sites were excluded from further analysis. Read counts for each gene were aggregated and analyzed using a Bayesian modeling approach.⁶⁹ The posterior probabilities for transposon insertion at each gene were estimated using PyMC (v4). For each gene, insertion probability was modeled using a uniform Beta (1,1) prior, and observed read counts were assumed to follow a Binomial likelihood parameterized by the gene-specific insertion opportunity and probability of insertion. Posterior inference was performed using the No-U-Turn Sampler (NUTS) with default PyMC settings. All MCMC chains demonstrated satisfactory convergence (≈ 1.0). Gene-level posterior probabilities were then log-transformed, yielding a distinctly bimodal distribution (Extended data 1b), with the lower peak corresponding to essential genes (low probabilities of transposon insertions) and the upper peak representing non-essential genes. To objectively separate these two groups, a two-component Gaussian mixture model (GMM) was fitted to the distribution, and the cutoff was determined at the intersection point of the two Gaussian components, where the posterior probabilities of belonging to either category are equal. To further enhance stringency, only genes that additionally exhibited a high component membership probability (P > 0.95) in the essential cluster were ultimately classified as essential.

For identifying genes essential for colonization in the adult gut, DESeq2 was employed to analyze differential abundance of read counts between samples collected from the initial mutant library and those recovered from adult mouse fecal samples. Read count normalization was performed by calculating size factors based on the total reads per sample. Results were adjusted for multiple testing using the Benjamini-Hochberg method, and genes with an adjusted p-value <0.05 were considered significantly enriched.

In the neonatal gut, ZINB models were used to account for excess zeros and overdispersion in read count data. Samples from adult and neonatal mice were compared using a generalized linear mixed-effects model (glmmTMB package). Gene-wise estimates and p-values for group-specific differences were extracted from the model. False discovery rates (FDR) were calculated to identify genes with significant changes in fitness, using an FDR cutoff of 0.05. Functional enrichment analysis was then performed on significant gene sets to identify critical pathways contributing to bacterial survival in the neonatal environment.

Competitive colonization assay

The colonization fitness of different B. fragilis, B. thetaiotaomicron strains was assessed by a competitive colonization assay. Erythromycin-sensitive and erythromycin-resistant strains (engineered via pNBU2 cassette insertion) were cultured in liquid PYG medium until they reached exponential growth phase (OD600 ~1.0). Equal volumes of each strain, adjusted to identical optical densities, were mixed prior to oral gavage into germ-free Swiss Webster mice. A fraction of the input mixture (used for gavage) and stool samples collected at defined time points were plated on BHI agar with and without erythromycin to distinguish between the two strains. Colony-forming units (CFUs) were enumerated, and the relative proportions of the two strains in each sample were calculated. The normalized ratio of the strains in stool samples was determined by comparing their proportions to those of the initial inoculum.

For tripartite competition assays with B. salyersiae, the proportions of individual strains were determined using quantitative real-time PCR (qRT-PCR). Stool samples were collected at defined time points, including the initial inoculum, and genomic DNA (gDNA) was extracted. Strain-specific primers were employed to quantify the abundance of each strain, enabling accurate calculation of their relative proportions within the microbial community.

RNA extraction and quantitative real-time reverse transcription PCR (qRT-PCR)

RNA was extracted from the bacterial cells cultured under the indicated condition using the Direct-zol RNA Miniprep kits (Zymo research) following the manufacturer’s instructions. Total RNA (1,000 ng) from each sample was used to synthesize cDNA using the qScript cDNA Synthesis Kit (Quantabio). Twenty-fold diluted cDNA was subjected to real-time PCR amplification using the iTaq Universal SYBR Green One-step Kit (BioRad) with specific primers in a CFX96^™ Real-Time System (BioRad, Hercules, CA, USA).

Colonic NKT cell analysis

Germ-free (GF) mice and mice monocolonized from birth with either WT B. fragilis or its isogenic ppGpp⁰ mutant were euthanized for analysis. The large intestines were excised, with surrounding adipose tissue carefully removed. The intestines were opened longitudinally, thoroughly cleared of fecal content, sectioned into 1-inch segments, and subjected to two washes in HBSS containing 2 mM EDTA for 30 minutes at 37°C. Following epithelial cell removal, the tissues were rinsed briefly in HBSS, then incubated in RPMI 1640 containing 10% FBS, collagenase type VIII (1 mg ml⁻¹), and DNase I (0.1 mg ml⁻¹) for 45 minutes at 37°C with continuous agitation. The resultant digested tissues were resuspended in FACS buffer (PBS with 2% FBS and 1 mM EDTA), filtered sequentially through 70 μm and 40 μm strainers, and prepared for subsequent flow cytometry analysis.

For flow cytometry staining, the following reagents were employed: APC-labelled mCD1d tetramer (unloaded or PBS-57 loaded; 1:500; NIH Tetramer Core Facility), anti-mouse MHC class II-PE/Cy7 (1:1000), TCRβ–PE (1:300), CD45–PerCP–Cy5.5 (1:300; Biolegend), and a cell viability dye (Fixable Viability Dye eFluor 780, 1:1,000; ThermoFisher). Samples were stained for 30 minutes at 4°C, followed by washing with cold FACS buffer. Flow cytometry was conducted using a BD FACSFortessa system (BD Biosciences), with gating performed based on forward and side scatter, singlet population, and viability. The frequencies of TCRβ⁺CD1d tetramer⁺ cells within the gated total CD45⁺ population were quantified as the target population. Data analysis and quantification were performed using FlowJo V10 software (BD Biosciences).

Lipidomics and Metabolomics analyses

Sample Preparation

For lipidome analysis, bacterial cell pellets (harvested from 1 ml of OD 1.0 cultures) or mouse fecal and colonic content samples (~20 mg) were extracted using an one phase extraction with 100% MeOH (for bacterial cells) and the Matyash method (for mouse samples; MeOH:MTBE:Water = 3:10:2.5). Perdeuterated (d₃₅) β-galactosylceramide (Matreya) was included as an internal standard for normalization.

For cellular metabolite analysis, bacterial cultures grown in 50 ml of minimal media supplemented with 0.5% glucose were harvested, and metabolites were extracted using cold extraction buffer (MeOH:Acetonitrile:Water = 40:40:20 with 0.1% formic acid) at −20°C overnight. The extracts were centrifuged at 10,000 × g for 10 min, dried using a SpeedVac at 4°C, and reconstituted in 60 μl of MeOH for analysis.For short-chain fatty acid (SCFA) analysis, the spent media were subjected to chemical derivatization using 3-nitrophenylhydrazine (3-NPH). Briefly, 30 μl of medium was mixed with 30 μl of 5 μg ml⁻¹ internal standard (d₄-acetic acid, d₅-propionic acid, d₇-butyric acid in 100% acetonitrile), followed by the sequential addition of 20 μl of 200 mM 3-NPH in 50% acetonitrile and 20 μl of a mixture containing 120 mM N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride and 6% pyridine in 50% acetonitrile. After incubation at 40°C overnight, the reaction was chilled on ice, diluted with 100 μl of water, and subsequently diluted 10-fold in 25% acetonitrile prior to LC-MS analysis.

UHPLC-MS and MS/MS Conditions

Mass spectrometry analyses were performed using an ultra-high-performance liquid chromatography-tandem mass spectrometry (UHPLC–MS/MS) system (Thermo Scientific Vanquish RP-UPLC coupled with a Q Exactive Orbitrap). The analysis was conducted in both negative and positive ion modes with the following parameters: Spray voltage: 3.25 kV; Sheath gas: 40 AU; Auxiliary gas: 8 AU; Capillary temperature: 350°C; Auxiliary gas heater temperature: 350°C; S-lens RF level: 65.0 AU; Mean collision energy: 22.5 AU.

For lipidomics analysis, chromatographic separation was performed using a YMC-Triart C8 column (2.1 mm × 100 mm × 3 μm) under the following gradient elution conditions: (0–2 min) 50% 2-propanol, 10% acetonitrile, 2.5% formic acid, (2–5 min) Linear increase to 85% 2-propanol, 10% acetonitrile, 2.5% formic acid, (5–12 min) Isocratic hold, (12–20 min) Immediate re-equilibration to the initial conditions and hold. The column was maintained at 40°C, with a flow rate of 200 μL min⁻¹. High-resolution mass spectrometry (MS1 scan resolution = 70,000 at m/z 200) was used in combination with top 10 data-dependent acquisition (DDA) at R = 17,500, using an isolation window of 1.0 m/z. Alternatively, a parallel reaction monitoring (PRM) approach was employed to selectively target ceramides and their derivatives, including aGC and PI-cer.

For cellular metabolite analysis, chromatographic separation was performed using an Acquity UPLC BEH Amide XP column (2.1 mm × 150 mm, 1.7 μm) under the following gradient conditions: (0–2 min) 95% acetonitrile with 10 mM ammonium acetate and 10 mM ammonium hydroxide (isocratic), (2–15 min) linear decrease to 20% acetonitrile with the same buffer composition, (15–16 min) isocratic hold, (16–20 min) return to initial conditions, followed by (20–30 min) re-equilibration. High-resolution mass spectrometry (MS1 scan resolution = 120,000 at m/z 200) was performed over an m/z range of 70–1,000.

For SCFA analysis, chromatographic separation was performed using an Agilent Zorbax C18 column (4.6 mm × 75 mm, 1.8 μm) at 40°C with a flow rate of 0.6 mL min⁻¹. The gradient elution conditions were as follows: (0–3 min) 25% acetonitrile with 0.05% formic acid (isocratic), (3–8 min) linear increase to 97.5% acetonitrile with 0.05% formic acid, (8–14 min) isocratic hold, and (14–20 min) re-equilibration to initial conditions. High-resolution mass spectrometry (MS1 scan resolution = 70,000 at m/z 200) was performed over an m/z range of 100–700, and targeted detection of 3-NPH-derivatized SCFAs and internal standards was achieved using parallel reaction monitoring (PRM) at R = 17,500.

Data analysis

Data acquisition was performed using Xcalibur 4.0 (Thermo Fisher Scientific). Relative quantification for all LC-MS-based analyses—including lipidomics, cellular metabolomics, and SCFA profiling—was carried out by integrating MS1 peak areas. For lipidomic and metabolomic datasets, values were normalized to the recovery of internal standards and either sample mass (for fecal and intestinal contents) or optical density (OD) of bacterial cultures. For SCFA measurements, quantification was normalized to the volume of collected culture supernatant, using stable isotope-labeled standards for calibration.

Determination of oxygen consumption rates of B. fragilis

Oxygen consumption rates (OCRs) of wild-type, ΔagcT, and Δcyd mutant B. fragilis cells cultured under microaerobic conditions were measured using the Extracellular Oxygen Consumption Assay (ab197243, Abcam), following the manufacturer’s instructions with minor modifications. Bacterial cultures were harvested, immediately resuspended in minimal medium supplemented with 0.5% glucose, and diluted to an OD₆₀₀ of 1.0. Two hundred microliters of the cell suspension were aliquoted into black-walled 96-well plates, followed by the addition of 15 μL of the extracellular O₂ consumption reagent. Two drops (~100 μL) of pre-warmed high-sensitivity mineral oil were added to each well to prevent oxygen diffusion. Fluorescence was monitored every 2 minutes for 2 hours using an excitation wavelength of 380 nm and an emission wavelength of 650 nm. Cells not treated with the dye were used as a negative control, and fluorescence values were normalized by subtracting the signal of dye-free samples. Cell-free medium (either aerobically incubated or deoxygenated) was used as an additional control to gauge baseline oxygen levels. Oxygen consumption rate (OCR) was determined by performing linear regression on the fluorescence intensity values over time. The slope of the linear portion of the time vs fluorescence plot (expressed as RFU min⁻¹) was used as a measure of OCR for each replicate well.

Liposome Preparation and Proton Permeability Measurement of Liposomes

Wild-type B. fragilis and its isogenic ΔagcT mutant were cultured under microaerobic conditions until reaching an optical density of ~0.5. Lipids were extracted from 10 mL of bacterial culture using the Matyash method (MeOH:MTBE:Water=3:10:2.5). The extracted lipids, with or without supplementation of synthetic BfaGC (17:0/17:0) (SB2211)⁴² at the indicated concentrations, were evaporated to dryness under a nitrogen gas stream in a glass test tube.

In parallel, liposomes were also generated using defined synthetic lipid mixtures composed of 1,2-dioleoyl-sn-glycero-3-phosphatidylcholine (DOPC) (Cayman, 15098), C16 Dihydroceramide (d18:0/16:0) (Avanti Research, 860634), and C16 Galactosyl(α) Dihydroceramide (d18:0/16:0) (Avanti Research, 860730). Each lipid was prepared as a 2 mM stock solution, and 30 μL of DOPC and 20 μL of a ceramide/aGC mixture (in the indicated molar ratio) were combined and dried under nitrogen prior to rehydration and vesicle formation.

To generate liposome suspensions, the dried lipid film was then rehydrated with 300 μL of buffer solution (2 mM trisodium 8-hydroxypyrene-1,3,6-trisulfonate (pyranine) (ThermoFisher Scientific, L11252), 5 mM KH₂PO₄, 100 mM KCl, adjusted to pH 7.0 with KOH) at 50°C for 10 minutes. The rehydrated lipid mixture was sonicated for 20 minutes using an ultrasonic cleaner.

To remove disordered liposomes, the suspension was centrifuged at 5000 × g for 10 minutes, followed by dialysis against pyranine-free buffer (5 mM KH₂PO₄, 100 mM KCl, adjusted to pH 7.0 with KOH) to eliminate any extraliposomal pyranine.

Proton permeability of liposomes was assessed as previously described with modifications. Each liposome suspension (50 μL) was mixed with 150 μL of acidic buffer (5 mM KH₂PO₄, 100 mM KCl, adjusted to pH 4.5 with HCl). Fluorescence emission at 510 nm was monitored over 30 minutes using a fluorescence spectrophotometer. The internal pH of the liposomes was determined based on the logarithmic ratio of fluorescence emission intensities at excitation wavelengths of 410 nm and 454 nm: log (I_{410 nm}/ I_{454 nm}). To construct the calibration curve, the fluorescence ratios were plotted against the pH of standard solutions (33 μM pyranine; 5 mM KH₂PO₄; 100 mM KCl) across a pH range of 5 to 7.5.

Determination of membrane potential dynamics in B. fragilis

Wild-type and ΔagcT mutant B. fragilis cells were cultured under microaerobic conditions until reaching the logarithmic growth phase. The cultures were then diluted to an OD₆₀₀ of 0.1 in growth medium supplemented with 0.5 mg mL⁻¹ fatty acid-free bovine serum albumin (BSA). Cells (200 μl) were transferred to black polystyrene 96-well plates, and their autofluorescence was recorded for up to 5 minutes. DiSC₃(5), dissolved in DMSO, was subsequently added to a final concentration of 2 μM (1% DMSO), and fluorescence quenching was monitored until a stable baseline was established. Fluorescence intensity was measured every minute, with vigorous shaking between readouts, using an excitation wavelength of 651 nm and an emission wavelength of 675 nm. All media, plates, and instruments were pre-warmed to 37°C before use.

Determination of the activity of electron transport chain complexes

To assess the activity of respiratory complexes, inside-out membrane vesicles were prepared from Bacteroides fragilis strains, including wildtype, ΔagcT, and the indicated mutants. Briefly, bacterial cells were converted into spheroplasts by incubation in 200 mM Tris-HCl (pH 8.0), 2 mM EDTA, and 30% sucrose containing 0.1 mg mL⁻¹ lysozyme. Following centrifugation at 5,000 × g for 10 minutes, the spheroplasts were resuspended in 100 mM HEPES-KOH (pH 7.5), 50 mM KCl, 10 mM MgSO₄, 2 mM dithiothreitol (DTT), and 0.5 mM phenylmethanesulfonyl fluoride. The suspension was sonicated, and cell debris was removed by centrifugation at 5,000 × g for 10 minutes. Membrane vesicles were collected by ultracentrifugation at 120,000 × g for 30 minutes and resuspended in 25 mM HEPES-KOH buffer (pH 7.5).

Cytochrome oxidase activity was quantified by monitoring the oxidation of N,N,N′,N′-tetramethyl-p-phenylenediamine (TMPD) in the presence of 0.2 mM ascorbate. Absorbance at 611 nm was measured every 10 seconds over 10 minutes using 5 μg of membrane protein in a reaction containing 5 mM TMPD. NADH dehydrogenase activity was assessed via the reduction of 2,6-dichlorophenolindophenol (DCIP) at 600 nm, with absorbance measured over a 30-minute interval. Reactions contained 50 μM DCIP, 125 μM NADH, and 1 μg of membrane vesicles. Enzyme activity was expressed as absorbance change per minute per microgram of membrane protein.

Enterotoxigenic B. fragilis infection

Three-week-old C57BL/6J (Jackson laboratory) were pretreated with an oral antibiotic cocktail in drinking water (0.5 g L⁻¹ vancomycin, 0.5 g L⁻¹ gentamicin, and 1g L⁻¹ ampicillin) for 3 days (day −4 to −1), followed by 0.2 g L⁻¹gentamicin alone on day −1. On day 0, mice were orally gavaged with 10⁹ cells of B. fragilis strains and monitored for 7 days. At day 7, colon and cecal tissues were collected. Group assignment and clinical scoring were blinded.

Outcomes included colon length, histopathology (H&E, scoring systems), B. fragilis burden (determined by plating onto BHI agar plates with gentamicin and erythromycin) and total bacterial burden (Brucella agar plates). Bft expression level was compared among B. fragilis strains expressing HisBft by western blot using anti-His antibody.

Metagenomic analysis

Species-level profiles from multiple early-life human gut metagenomic cohorts, including TEDDY,¹² DIABIMMUNE,¹⁵ and Baumann-Dudenhoeffer et al. (2018),¹⁴ were used for cohort-level analysis. For bacterial species identified in these profiles, reference genomes were downloaded using the NCBI Datasets tool. Homologues of sphingolipid-related genes were searched in these reference genomes using DIAMOND BLASTp with a cutoff of >40% protein sequence identity. Species carrying each gene of interest were annotated as gene-encoding species, and their relative abundance were estimated from the corresponding species-level metagenomic profiles. Age-dependence changes in the abundance and taxonomic contribution of gene-encoding species were analyzed and visualized using locally estimated scatterplot smoothing (LOESS) implemented in the ggplot2 package in R.

Pimonidazole staining and quantification of intestinal hypoxia

To assess intestinal oxygenation in neonatal mice, pimonidazole staining was performed using the Hypoxyprobe^™-1 Kit (Hypoxyprobe, Inc.). Germ-free pregnant mice were monocolonized with WT E. coli and ΔhemA E. coli, and their offspring were analyzed at postnatal day 4 (p4). Pimonidazole was administered by intraperitoneal injection at 50 mg/kg body weight. Mice were euthanized 1 h after pimonidazole administration, and colon tissues were collected.

Tissue were fixed in 4% paraformaldehyde for 24 h at 4°C and embedded in paraffin. Sections of 5 μm thickness were prepared for immunostaining. Tissue sections were blocked in PBS-T containing 2% BSA for 2 h at 4°C and incubated with rat anti-pimonidazole primary antibody included in the Hypoxyprobe^™-1 Kit (1:50) overnight at 4°C. Sections were then washed and incubated with Alexa Fluor 594-conjugated goat anti-rat IgG H&L secondary antibody (Abcam, ab150160; 1:200) overnight at 4°C. Nuclei were counterstained with DAPI. Slides were mounted with SlowFade Glass Soft-Set Antifade Mountant and imaged on a Leica Stellaris 8 Falcon confocal system mounted on a Leica inverted DMi8 microscope using Las X acquisition software, with identical acquisition settings across all groups.

Pimonidazole signal was quantified using ImageJ. The luminal surface and mucosal region were defined based on DAPI staining and tissue morphology. For each field of view, a line was drawn from the luminal surface inward across the mucosa, and pimonidazole fluorescence intensity was measured within a 20-pixel-wide band along this line after background subtraction. Mean fluorescence intensity was calculated for each biological sample and used for statistical analysis. Representative images were selected from n=6 biological samples per group.

3. QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis

Statistical analyses were performed using GraphPad Prism 10 and R v4.4.1 unless otherwise noted. For comparisons between two groups, two-tailed unpaired Student’s t-tests were used. For multiple group comparisons, one-way or two-way analysis of variance (ANOVA) was applied, followed by Tukey’s post hoc test where appropriate. For transposon sequencing (Tn-seq) data, gene fitness in the adult gut was assessed using the Wald test within the DESeq2 framework, with multiple testing correction using the Benjamini–Hochberg procedure. In the neonatal gut, due to sparse counts, fitness estimation was performed using a zero-inflated negative binomial (ZINB) model, and significance was determined by Wald z-test. All statistical tests were two-tailed unless specified, and significance thresholds were defined as follows: *p-value < 0.05, **p-value < 0.01, ***p-value < 0.001. P-values greater than 0.05 were annotated.

4. ADDITIONAL RESOURCES

Data files for transposon sequencing and bacterial lipidomics analysis can be downloaded from the Harvard Dataverse using the following Unpublished Dataset Preview URL : https://dataverse.harvard.edu/previewurl.xhtml?token=59417da4-c3b3-48cb-b7ee-d0a8b073774f

Supplementary Material

2

Table S1. Genome-wide identification of fitness genes by Tn-seq in adult and neonatal mice, related to Figure 1.

3

Table S2. Enriched KEGG pathways identified by gene set enrichment analysis, related to Figure 1.

4

Table S3. Orthologous genes of Bacteroides fragilis 638R identified in nine Bacteroidales species, related to Figure 2.

5

Table S4. Primers used in this study, related to STAR Methods.

Document S1. Figures S1-S14

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CD1d tetramer, unloaded	NIH Tetramer Core Facility	Contract No. 75N93020D00005
CD1d tetramer, PBS-57 loaded	NIH Tetramer Core Facility	Contract No. 75N93020D00005
Anti-mouse TCRβ-PE	Biolegend	Cat# 109207; RRID: AB_313430
Anti-mouse CD45-PerCP-Cy5.5	Biolegend	Cat# 103132; RRID: AB_893340
Anti-mouse MHC class II-PE/Cy7	Biolegend	Cat# 107629; RRID: AB_2290801
Fixable viability dye eFluor 780	Thermo Fisher Scientific	Cat# 65-0865-14; RRID: AB_1272173
Anti-6X His tag® antibody [HIS.H8]	Abcam	Cat# ab18184; RRID: AB_444306
Goat Anti-Mouse IgG H&L (HRP)	Abcam	Cat# ab205719; RRID: AB_2755049
Bacterial and virus strains
Bacteroides fragilis NCTC 9343	Cato and Johnson (ref)	Bacteroides fragilis NCTC 9343
Bacteroides fragilis 638R	Privitera et al. (ref)	Bacteroides fragilis 638R
Bacteroides fragilis ΔagcT	Yoo et al., 2025	Bacteroides fragilis ΔagcT
Bacteroides fragilis ΔcydAB	This study	Bacteroides fragilis ΔcydAB
Bacteroides fragilis ΔcydAB ΔagcT	This study	Bacteroides fragilis ΔcydAB ΔagcT
Bacteroides fragilis ppGpp0	This study	Bacteroides fragilis ppGpp0
Bacteroides fragilis ppGpp0/pNBU2-relA	This study	Bacteroides fragilis ppGpp0/pNBU2-relA
Bacteroides fragilis WT/pNBU2	This study	Bacteroides fragilis WT/pNBU2
Bacteroides fragilis WT/Bft	This study	Bacteroides fragilis WT/Bft
Bacteroides fragilis ΔagcT/Bft	This study	Bacteroides fragilis ΔagcT/Bft
Bacteroides fragilis Δcyd/Bft	This study	Bacteroides fragilis Δcyd/Bft
Bacteroides fragilis WT/His-Bft	This study	Bacteroides fragilis WT/His-Bft
Bacteroides fragilis ΔagcT/HisBft	This study	Bacteroides fragilis ΔagcT/HisBft
Bacteroides fragilis ΔcydAB/HisBft	This study	Bacteroides fragilis ΔcydAB/HisBft
Bacteroides thetaiotaomicron VPI-5482	ATCC	Bacteroides thetaiotaomicron VPI-5482
Bacteroides thetaiotaomicron ΔPICS	This study	Bacteroides thetaiotaomicron ΔPICS
Bacteroides thetaiotaomicron ΔPICS/agcT	This study	Bacteroides thetaiotaomicron ΔPICS/agcT
Bacteroides salyersiae DSM 18765	DSMZ	Bacteroides salyersiae DSM 18765
Bacteroides salyersiae ΔagcT	This study	Bacteroides salyersiae ΔagcT
Bacteroides salyersiae ΔPICS	This study	Bacteroides salyersiae ΔPICS
Bacteroides caccae ATCC 43185	ATCC	Bacteroides caccae ATCC 43185
Bacteroides uniformis ATCC 8492	ATCC	Bacteroides uniformis ATCC 8492
Bacteroides stercoris DSM19555	DSMZ	Bacteroides stercoris DSM19555
Phocaeicola vulgatus ATCC8482	ATCC	Phocaeicola vulgatus ATCC8482
Phocaeicola dorei DSM 17855	DSMZ	Phocaeicola dorei DSM 17855
Bacteroides ovatus ATCC8483	ATCC	Bacteroides ovatus ATCC8483
Segatella copri DSM 18205	DSMZ	Segatella copri DSM 18205
Parabacteroides distasonis ATCC 8503	ATCC	Parabacteroides distasonis ATCC 8503
Escherichia coli MG1655	Laboratory stock	Escherichia coli MG1655
Escherichia coli ΔhemA	This study	Escherichia coli ΔhemA
Escherichia coli S17-1 λpir	Laboratory stock	Escherichia coli S17-1 λpir
Chemicals, peptides, and recombinant proteins
Brain Heart Infusion agar	Hardy Diagnostics	Cat# C5120
Brain Heart Infusion broth	Hardy Diagnostics	Cat# C5141
Proteose peptone	Hardy Diagnostics	Cat# C6641
Yeast extract	Hardy Diagnostics	Cat# C7341
Sodium chloride	Sigma-Aldrich	Cat# S9625
Glucose	VWR chemicals	Cat# 0188
Potassium phosphate dibasic	VWR chemicals	Cat# 26930.260
L-cysteine	Sigma-Aldrich	Cat# 168149
Hemin	Sigma-Aldrich	Cat# H9039
Vitamin K1	Sigma-Aldrich	Cat# V3501
Ammonium sulfate	VWR chemicals	Cat# 85509.290
Sodium carbonate	Sigma-Aldrich	Cat# 900505
Potassium phosphate monobasic	VWR chemicals	Cat# 0781
Calcium chloride dihydrate	Sigma-Aldrich	Cat# 223506
Magnesium chloride hexahydrate	Sigma-Aldrich	Cat# M2670
Manganese chloride tetrahydrate	Sigma-Aldrich	Cat# M5005
Cobalt chloride hexahydrate	Sigma-Aldrich	Cat# 255599
Iron sulfate heptahydrate	Sigma-Aldrich	Cat# 215422
Vitamin B12	Sigma-Aldrich	Cat# V2876
Ampicillin	Gold biotechnology	Cat# A-301-5
Chloramphenicol	Sigma-Aldrich	Cat# C0378
Erythromycin	Sigma-Aldrich	Cat# E5389
Gentamicin	Gold biotechnology	Cat# 50-153-2782
Anhydrotetracycline	Thermo scientific Chemicals	Cat# J66688
Serine hydroxamate	MedchemExpress	#HY-131894
Vancomycin	Gold biotechnology	Cat# 50-153-2848
Kanamycin	Sigma-Aldrich	Cat# K4000
Streptomycin	Sigma-Aldrich	Cat# S9137
Carbonyl Cyanide m-Chlorophenylhydrazone (CCCP)	Sigma-Aldrich	Cat# C2759
Colistin sulfate	Sigma-Aldrich	Cat# PHR-1605
Ethylenediaminetetraacetic acid disodium (EDTA)	MedchemExpress	Cat# HY-W018746
DiSC3(5)	Invitrogen	Cat# D306
Trisodium 8-hydroxypyrene-1,3,6-trisulfonate (Pyranine)	Thermo scientific Chemicals	Cat# L11252.14
1,2-Dioleoyl-sn-glycero-3-PC (DOPC)	Cayman chemical	Cat# 15098
Synthetic alpha-galactosylceramide (17:0/17:0)	Oh et al. (ref)	Cat# TBD
C16 Dihydroceramide (d18:0/16:0) used for synthetic SUVs	Avanti Research	Cat# 860634
C16 Galactosyl (α) Dihydroceramide (d18:0/16:0) used for synthetic SUVs	Avanti Research	Cat# 860730
Perdeuterated d35 β-galactosylceramide internal standard	Cayman chemical	Cat# 24467
METHYL T-BUTYL ETHER (MTBE)	Fisher scientific	Cat# E1274
Methanol	Sigma-Aldrich	Cat# MX0486
Acetonitrile	Sigma-Aldrich	Cat# AX0156
Isopropanol / 2-propanol	Solstice	Cat# 34965
Formic acid	Fisher scientific	Cat# A117-05
3-nitrophenylhydrazine	Sigma-Aldrich	Cat# 21804
d4-acetic acid	Cambridge Isotope Laboratories	CAT# DLM-12
d5-propionic acid	Cambridge Isotope Laboratories	CAT# DLM-1919
d7-butyric acid	Cambridge Isotope Laboratories	CAT# DLM-1508
Critical commercial assays
Direct-zol RNA Miniprep Kit	Zymo Research	Cat# R2050
qScript cDNA Synthesis Kit	Quantabio	Cat# 95047
iTaq Universal SYBR Green Supermix / One-Step Kit	Bio-Rad Laboratories	Cat# 1725121
QIAamp PowerFecal Pro DNA Kit	QIAGEN	Cat# 51804
Genomic DNA extraction kit	Invitrogen	Cat# K182000
Extracellular oxygen consumption assay kit	Cayman chemical	Cat# 502880
Deposited data
Tn-seq data	Harvard Dataverse	URL (see STAR Methods)
Bacterial lipidomics data	Harvard Dataverse	URL (see STAR Methods)
TEDDY cohort data	TEDDY repository (dbGaP)	dbGaP Study accession: phs001443
DIABIMMUNE infant microbiome data	Kostic et al., 2015 / curatedMetagenomicData	DIABIMMUNE
Baumann-Dudenhoeffer et al. 2018 infant microbiome data	Baumann-Dudenhoeffer et al., 2018	Baumann-Dudenhoeffer et al
UHGG genome catalog	MGnify	UHGG
Sund et al. oxygen transcriptome dataset	Sund et al., 2008	Oxygen-responsive transcriptome data
Experimental models: Organisms/strains
Germ-free Swiss Webster mice	Harvard Gnotobiotic Core Facility	SwissWebster
Gnotobiotic Swiss Webster mice	This study	Generated from germ-free Swiss Webster mice
SPF C57BL/6J mice	Jackson laboratory	Strain #: 000664, RRID:IMSR_JAX:000664
Recombinant DNA
pLGB13 suicide vector	Garcia-Bayona and Comstock, 2019	pLGB13
pNBU2_erm-TetR-P1T_DP-GH023	Lim et al., 2017	pNBU2
pSAM-Bt transposon plasmid	Goodman et al., 2009	pSAM-Bt
pNBU2-agcT	This study	pNBU2-agcT
pNBU2-relA	This study	pNBU2-relA
pNBU2-bft	This study	pNBU2-bft
pNBU2-HisBft	This study	pNBU2-HisBft
Software and algorithms
R	R Project for Statistical Computing	v4.6.0
RStudio	Posit	Rstudio Desktop v2026.05.0
tidyverse	R package	v.2.0.0
ggplot2	R package	v4.0.3
DESeq2	Bioconductor/R	v1.52.0
edgeR	Bioconductor/R	v4.10.1
BLAST	NCBI	V.2.17.0
FlowJo	BD Biosciences	v10.10.1
GraphPad Prism	GraphPad Software	v11.0.2
Xcalibur	Thermo Fisher Scientific	v4.8
MS-DIAL	RIKEN	v5.5.260323
CFX Manager	Bio-Rad	CFX Maestro v2.3
Bowtie2	Open source	v2.5.5
SAMtools	Open source	v1.23.1
bedtools	Open source	v2.31.1
fastp	Open source	v1.1.0
curatedMetagenomicData	Bioconductor	v3.20.0
Other
BD FACSFortessa	BD Biosciences	Instrument
CFX96 Real-Time PCR Detection System	Bio-Rad	Instrument
Thermo Scientific Vanquish RP-UHPLC	Thermo Fisher Scientific	Instrument
Q Exactive Orbitrap mass spectrometer	Thermo Fisher Scientific	Instrument
YMC-Triart C8 column	YMC	Cat# TO12S03-10Q1WT
Harvard Gnotobiotic Core Facility	Harvard Medical School	Animal resource

Highlights.

• BfaGC promotes B. fragilis colonization of the oxygen-exposed neonatal gut

• B. fragilis controls proton leakage with BfaGC to sustain respiratory bioenergetics

• BfaGC may help expanding bacterial niche and shaping host immune development

• Two classes of sphingolipids confer specialized versus broad fitness in the host