Dual human milk oligosaccharide-fibre utilisation is a selection cue for the weaning gut microbiome Authors: Yunjeong So, Michael Jakob Pichler, Susanne Søndergaard Kappel, Chunsheng Jin, Carsten Eriksen, Ioanna Chatzigiannidou, Maria Haugaard Bohl Andersen, Vasileios Tsiamis, Marie Vestergaard Lukassen, Lea Eschen Skytthe, Susann Teneberg, Karsten Kristiansen, Susanne Brix, Lise Aunsholt, Maher Abou Hachem Published: 2026 ABSTRACT AbstractGut microbiome (GM) maturation in early life follows organised taxonomic successions, yet how the weaning diet impacts these trajectories remains underexplored. Here, we collected faecal samples at pre-, early and late weaning from seven mother-infant dyads forming the Milkome cohort, designed to evaluate the contribution of human milk oligosaccharides (HMOs) to GM maturation during weaning (NCT07026526). Surprisingly, all preweaning infant faecal consortia grew on multiple dietary fibres, consistent with the prevalence of fibre-degradation genes in their metagenomes. Utilisation of both HMOs and dietary fibres was discovered as a metabolic hallmark of the weaning GM, as supported by metagenomics and the growth of faecal consortia on HMOs, following their enrichment on fibres. The growth of a defined consortium on weaning-mimic substrates, further showed that distinct Clostridia simultaneously deploy HMO and fibre utilisation pathways, which confers competitive growth against HMO- or fibre-utilising bifidobacteria. Metagenomics, culturomics and HMO-utilisation profiles of 137 maternal isolates were concordant with retention of the HMO-utilisation capacity by the adult GM. Our findings highlight dual HMO-fibre utilisation as an unrecognised selection cue of core adult GM species during weaning, which outlines a plausible mechanism of GM maturation in early life and extends the importance of HMOs to the weaning transition.Subject terms: Microbiome, Glycobiology Introduction The gut microbiome (GM) assembly commences at birth1,2 and this community continues to mature towards an adult-like structure during the first 1–3 years of life3,4. The mature GM structure is relatively resilient over time, hampering further colonisation of new members5,6. Perturbations in the assembly and maturation of the GM are associated with serious inflammatory7–10, metabolic11,12, and neurological13,14 disorders. These links collectively underscore the critical and persistent impacts of the early life GM on the developmental biology and health trajectory of the host15. Diet is a key factor shaping the GM16. The onset of breastfeeding and weaning (solid food intake) are critical nutritional milestones that elicit GM shifts. The most abundant component in mother’s milk, after lactose and fat, is a mixture of complex carbohydrates, collectively called human milk oligosaccharides (HMOs)17. HMOs are non-digestible by infants, thereby serving as prime nutrients and a selection cue for the assembly of the neonatal GM18. Exclusively breastfed infants are predominantly colonised by HMO-utilising bifidobacteria19, whereas formula feeding favours different communities1,20,21. During weaning, infants’ diet gradually shifts from mother’s milk to solid food, rich in dietary fibres from grains, vegetables and fruits, marking a key GM maturation phase. Besides HMOs, mucin or other host-derived glycans22,23, dietary fibres are the main available nutrients for the GM during weaning24. Maturation of the GM in early life follows defined taxonomic trajectories, despite high interindividual variations in microbial composition and dietary profiles2–4. The expansion of Bacillota is followed by Bacteroidota1,2,25, with a decreased abundance of bifidobacterial HMO-utilising specialists. To date, breastfed infant cohorts are sampled by infant age1–3,26–30 and not nutritional status, precluding matching of weaning nutritional profiles to available longitudinal metagenomic data. Fluctuating HMO-dietary fibre gradients are a unique dietary signature of weaning, yet the mechanisms by which the weaning diet influences GM shifts remain underexplored. Here, we analysed the metagenomes of faecal samples and enriched communities from the MILKOME infant-mother dyad cohort, designed specifically to monitor complex carbohydrate utilisation profiles of infant GM at three weaning stages, in addition to the maternal GM. Remarkably, we show that the ability to utilise dietary fibres is acquired before the introduction of solid food. Interestingly, HMO utilisation capacity was maintained by both the late weaning infant and the mother GM. These findings unveil a plausible selection mechanism of core GM groups during weaning, mediated by the simultaneous exposure to HMOs and dietary fibre gradients. Results Pre-weaning microbiome harbours both HMO and dietary fibre utilisation potential To study the maturation of the infant GM during weaning, we established the Milkome infant-mother dyad cohort. Healthy mothers and their vaginally delivered and exclusively breastfed infants, who did not receive antibiotics or formula even during weaning, were recruited (Supplementary Table 1). Stool samples (n = 18) from seven mothers and infants were collected at pre-, early and late weaning, defined as 100%, 70% (7 out of 10) and 30% (3 out of 10) of meals being mother’s milk, respectively, with the rest being solid food (Fig. 1a). Sampling and analyses scheme of the Milkome cohort. a Samples were collected at pre-weaning (T1), early weaning (T2), and late weaning (T3) based on the mother’s milk to solid food meal ratio (%). b Study overview spanning sample collection, enrichment of microbial consortia from the faecal samples on different carbohydrates, including human milk oligosaccharides (HMOs), strain isolation, and whole metagenomic shotgun sequencing. Two modified yeast casitone fatty acid media (mYCFA, mYmCFA) were used to reduce background growth on the medium (see “Methods”). The figure was partially created in BioRender (Abou Hachem, M. (2026) https://BioRender.com/mt4g87u). For two dyads, either the late weaning or both the early and late weaning samples were not delivered (Fig. 1a). Freshly (< 1.5 h) collected faecal samples were equilibrated in an anaerobic chamber and enriched on selected carbohydrates. We performed metagenomics shotgun sequencing of infant faecal samples from the three weaning stages, mother faecal samples collected at early weaning of their infants, and faecal consortia enriched on carbohydrates to allow strain-level annotation (Fig. 1b). Actinomycetota members displayed highest relative abundance in 6 out of 7 infant pre-weaning faecal samples (Fig. 2a). Low levels of Bacteriodota at pre-weaning, which increased in late weaning samples concomitant with a reduction in Pseudomonadota, were consistent with previously reported trends1,3,27. The top abundant and prevalent Bifidobacterium species were HMO-utilisers, e.g., the specialist Bifidobacterium longum subsp. infantis (B. infantis)19,31, and Bifidobacterium bifidum that utilises mucin O-glycans and HMOs via extracellular glycosidases32,33. The abundance of B. infantis was at the cost of other Bifidobacterium species, including B. longum (Infants C and D at T1), B. bifidum (Infants D and E at T1) and the less prevalent B. pseudocatenulatum (Infant E at T1), all of which have been reported to utilise HMOs via different systems than B. infantis34 (Fig. 2b). These observations are consistent with the specialisation of B. infantis on HMOs, notably the major fucosyllactose fraction19,34 (Supplementary Figs. 1a, b and 2). B. adolescentis, not previously reported to utilise HMOs, was only observed in a single sample (Infant B at T2 and T3) (Fig. 2b), indicating poor competitiveness of this species, even at late weaning. Bacilli and Clostridia, besides Negativicutes, were also prevalent in the weaning samples, but their HMO utilisation machinery remains underexplored. These data emphasise the important contribution of HMO utilisation for the competitiveness of bifidobacteria at least until late weaning. Complex carbohydrate-utilisation potential of infant microbiome during weaning. a Phylum level profiles of faecal bacteria at pre-weaning (T1, n = 7), early weaning (T2, n = 6), and late weaning (T3, n = 5). Weaning progress in months illustrates the individual weaning rates. b Heatmap of changes in Bifidobacterium species. Other taxa are depicted at class level. c Shifts in carbohydrate active enzymes (CAZymes) targeting HMOs, HMOs/host glycans, O-glycans and glycosaminoglycans (GAGs). d Heatmap changes in different CAZyme families and top three abundant genes per substrate are shown. e Shifts in CAZymes targeting starch, xylan backbone, pectin backbone, β-glucans, including xyloglucans, and fibres used as food additives. f α-Diversity using the Shannon index (vegan35), highlighting diversification of species and CAZymes targeting starch, dietary fibres and food additives as weaning progresses. The P values in (c, e, and f) were calculated using two-sided Wilcoxon rank-sum test in R36. g Non-metric multidimensional scaling (NMDS) ordination of species composition calculated using Bray-Curtis dissimilarity (vegan). CAZyme associations with community structure were assessed using envfit (vegan, 999 permutations). Arrows indicate significant changes in CAZymes after multiple testing correction using the Benjamini–Hochberg false discovery rate (FDR) (P.adj < 0.05); those with P.adj < 0.015 are labelled. Colours denote substrate (dietary fibres: green, host glycans: blue, fibres used also as food additives: grey). *Mother of the infant had one antibiotics dose during birth. To analyse complex carbohydrate utilisation, we annotated carbohydrate-active enzyme (CAZyme)37,38 encoding genes in the metagenomic data. Specific CAZyme families and subfamilies were manually assigned as markers for selected fibres, HMOs and host-derived glycans (Supplementary Tables 2 and 3, Source Data), based on biochemically characterised homologues in each CAZyme family (www.cazy.org). Glycoside hydrolase (GH) families 11239 and 13640,41, which degrade the HMO type 1 building block lacto-N-biose (LNB) and the lacto-N-tetraose (LNT) backbone in HMOs, respectively, were selected as HMO utilisation markers (Fig. 2c, d and Supplementary Figs. 1c and 2). Additionally, HMO-active β-galactosidases (GH42)42, fucosidases (GH29 and GH95), sialidases (GH33) and lacto-N-biosidases of GH2037 were included (Supplementary Fig. 2, Supplementarty Results). The potential to utilise HMOs was persistent during weaning, as indicated by the abundance of genes encoding HMO-active CAZymes and non-significant changes between the weaning stages (Fig. 2c, d). The utilisation potential of host-derived glycosaminoglycans (GAGs) was low, suggestive of the poor competitiveness of GAG degraders, e.g., Bacteroides spp. at early weaning in the healthy infants43. Surprisingly, CAZyme genes targeting inulin, starch, xylan, pectin, and/or the food additive hydrocolloids β-mannan and gum Arabic (GA), which are also dietary fibres, were present already in pre-weaning samples (Fig. 2d–f), indicating that distinct fibre-utilising bacteria are acquired prior to introduction of solid food. The normalised gene abundance of pectin backbone, food additive fibres and GAG targeting CAZymes, increased significantly at late weaning compared to pre-weaning, suggesting a functional expansion of carbohydrate utilisation capacity along the dietary transition (Fig. 2e). Bacterial richness and evenness were only significantly different between pre-weaning and late weaning samples (Fig. 2f), consistent with the pivotal role of HMOs in shaping the GM, particularly at early weaning. A similar pattern was shown for starch, dietary fibres, including those used as food additives with significant changes observed at late weaning. Although the lack of information on onset and rate of weaning in published cohorts hampers direct comparison to our data, similar Shannon index trends in microbial composition were observed between our cohort and a previously reported large cohort3. The weaning gut microbiome utilises both plant dietary fibres and HMOs Next, we sought to examine the carbohydrate utilisation profiles of infant GM. We enriched faecal samples on a monosaccharide blend (glucose/galactose/fucose), HMOs, mucin, and dietary fibres abundantly present in cereal grains (arabinoxylan (AX)), vegetables and fruits (pectin (PEC); xyloglucan (XC)) (Supplementary Figs. 3 and 4 and Supplementary Tables 4–6), besides the natural food additive fibres galactomannan (GMN) and gum Arabic (GA). These enrichments were performed twice (enrichments I and II) (Fig. 3a). The first enrichment on fibre blends allowed rapid handling of fresh faecal samples. The second enrichment was performed using an optimised medium (Supplementary Table 4) on individual complex carbohydrate substrates. Growth of faecal consortia on HMOs, mucin and dietary fibres. a Workflow for faecal sample enrichment. Enrichment I was performed on a modified yeast extract-reduced medium (mYCFA) in technical duplicates, whereas enrichment II was conducted in a further optimised medium with reduced yeast extract and protein (Casitone) content (mYmCFA) in technical triplicates. b Changes of pH after a 24 h enrichment on HMOs, porcine gastric mucin, and dietary fibres. The consortia enriched on dietary fibres were subsequently re-grown on HMOs to evaluate dual HMO-fibre utilisation capacity. The P values were calculated using two-tailed Welch’s test with unequal variance and adjusted using Benjamini–Hochberg false discovery rate (FDR) method. Asterisks indicate FDR-adjusted P values: ns, Padj ≥ 0.05; * Padj < 0.05; ** Padj < 0.01; *** Padj < 0.001; **** Padj < 0.0001. Samples with Padj < 0.05 and pH decrease > 0.5 were outlined with a brown box. The means, standard deviations, and Padj values from three technical replicate growth experiments (n = 3) are provided (Source Data 1). Panel a is partially created in BioRender (Abou Hachem, M. (2026), https://BioRender.com/y395xm1). All the infant faecal samples displayed growth on HMOs in both enrichments, while growth on mucin was weak or insignificant in half of the 18 cultures (Fig. 3b and Supplementary Fig. 5). Interestingly, prevalent growth on dietary fibres was observed in pre-weaning samples, especially for AX, followed by PEC, GMN, XG, and GA (Fig. 3b and Supplementary Fig. 6a, b). Accordingly, all early and late infant weaning samples grew on AX and PEC (Fig. 3b and Supplementary Fig. 5). Growth of pre-weaning infant GM on dietary fibres has to our knowledge, hitherto, not been demonstrated. Then, we selected a Swedish mother-infant dyad cohort (n = 98)1 with available nutritional44 and metagenomic data to investigate the prevalence of fibre utilisation potential pre-weaning. We annotated both the HMO and fibre degradation CAZymes in exclusively breastfed infants sampled in the first week (n = 53) and at 4th month of life (n = 51), similarly to above. Most infants acquired HMO utilisation genes in the first week, with 30 out of 37 HMO and host-glycan utilisation genes being shared with our cohort (Supplementary Fig. 7a–c). The fibre utilisation CAZyme profiles from the Swedish cohort were more diverse; however, the metagenome of each sample (n = 104) encodes the potential to utilise at least one, but frequently multiple fibre types (Supplementary Fig. 7d–f), concordant to our cohort. Despite interindividual variations, median abundances of the selected CAZyme genes in the Swedish cohort corroborated the considerable fibre utilisation potential of the GM during exclusive breastfeeding (Supplementary Fig. 8). The growth on fibre by pre-weaning faecal consortia from all Milkome infants and the metagenomic analyses of both cohorts are consistent with the presence of distinct fibre utilising species in the GM of breastfed infants prior to weaning. The poor growth on mucin suggested that HMOs may sustain fibre-utilisers during exclusive breastfeeding. To test this, we examined the growth of the fibre-enriched consortia (enrichment II) on HMOs, which demonstrated the growth of most weaning consortia on HMOs (Fig. 3a, b and Supplementary Fig. 6b). These findings unravel the dual ability to utilise both HMOs and dietary fibres as a hitherto undescribed signature of the weaning GM. HMO degradation genes during weaning We explored the composition of the enriched consortia from the infant faecal samples on both HMOs and mucin. When present in the original faecal samples (12 out of 18), B. infantis dominated most of the HMO-enriched consortia, reaching similar or higher relative abundance than in the corresponding faecal samples across all time points (except Infant E at T3) (Fig. 4a). By contrast, B. infantis showed considerably lower relative abundance in consortia enriched on mucin, highlighting the specialisation of B. infantis on HMOs. By comparison, B. bifidum had generally a higher relative abundance on mucin than HMOs, reflecting its adaptation to mucin O-glycans. Interestingly, in the absence of B. infantis, HMO-enriched consortia were dominated by other Bifidobacterium species (e.g., B. catenulatum; Infant C at T1, B. pseudocatenulatum; Infant E at T1, or B. bifidum; Infant D at T2) or by Clostridium butyricum (Infant D and G at T1) and Ruminococcus gnavus (Infant G at T3), with Enterococcus spp. also relatively abundant, suggesting these taxa can utilise HMOs or their transformation products. Overlap in growth on HMOs and mucin was observed for Enterococcus species, potentially suggesting a less discriminative enzymatic machinery. The key butyrate producer45, Roseburia intestinalis increased in abundance in a late weaning sample enrichment on mucin (Infant A), consistent with its reported cross-feeding on mucolytic bacteria22. Infant faeces consortia enriched on HMOs and Mucin. a Microbial community composition in faeces and faecal consortia enriched on dietary fibres based on metagenomic analyses. Bubble size indicates relative abundance, and fill colour represents the log2 fold change relative to the original faecal sample. Only taxa with fold change ≥ 1 and relative abundance ≥ 10% in at least one enriched consortium are shown at species level for clarity, except the Enterococcus and Streptococcus genera and Negativicutes class. For included taxa, individual data points with relative abundance <0.05 are not displayed. b Selected HMO-degrading activities of glycoside hydrolase families: GH112, GH136, GH20, GH42, GH29, GH95, and GH33. c Abundance per sample of the HMO-utilisation genes, encoding the CAZymes in (b). HMO gene abundances were summed per sample, and differences between samples with and without Bifidobacterium infantis were assessed using unpaired two-sided Wilcoxon rank-sum tests. Group medians are indicated, and corresponding P values are shown. Genes exhibiting a ≥ twofold increase are indicated by a white asterisk (*). d The total HMO utilisation CAZyme (as in (b)) genes were quantified per condition (faeces, HMOs, and mucin). Each data point represents an individual donor (colour) and time point (geometric shape). Differences per condition were assessed using unpaired two-sided Wilcoxon rank-sum tests. Gene abundance was normalised using a TPM (transcript per million) approach with read counts divided by gene length and scaled to one million across genes. We also compared genes encoding HMO type 1 backbone degrading enzymes (GH112 and GH136), sialidases and fucosidases (GH33, GH29 and GH95), as well as hexosaminidases and lacto-N-biosidases (both from GH20, Fig. 4b and Supplementary Fig. 2). The selected genes were abundant in most samples at levels comparable to the faeces, underscoring a pivotal role of HMOs in GM selection even at late weaning (Fig. 4c, d). Multivariate analysis based on Bray-Curtis dissimilarity revealed no significant effects of donor identity, weaning stage, or maternal antibiotic exposure on abundance profiles of HMO utilisation genes across faecal samples and enriched consortia on HMOs or mucin (Supplementary Table 7). The presence or absence of B. infantis had no significant effect on the median abundance of HMO genes in the faecal samples and HMO-enriched consortia, consistent other HMO utilisers blooming in the absence of B. infantis. The significantly lower abundance in the mucin-enriched consortia validated that the selected CAZymes are preferentially associated with HMO utilisation during weaning (Fig. 4d). Fibre utilisation gene signatures in the weaning infant GM We analysed the fibre-enriched consortia to explore both the taxonomic groups and genes conferring fibre degradation potential during weaning. The inability of most B. infantis strains to utilise fibres was underscored by a lower relative abundance than the visualisation threshold (10%) in all enrichments (Fig. 5a). By contrast, B. longum was enriched in early and/or late weaning on AX (infants A, B and E) relative to the faecal samples (Fig. 5a), attesting the growth of distinct HMO-utilising bifidobacteria on plant fibres. Moreover, Enterococcus species and other Bacilli were also enriched on plant fibres, including those used as food additives (Fig. 5a and Supplementary Fig. 6c). Interestingly, the relative abundance of several Clostridia increased in the fibre enrichments relative to faecal samples, including Lachnospiraceae such as the butyrogenic R. intestinalis, a primary degrader of xylan and mannan46,47 as well as three Clostridium species and Agathobacter rectalis (Fig. 5a). Infant faecal consortia enriched on common dietary fibres. a Microbial community composition in faeces and faecal consortia enriched on dietary fibres based on metagenomic analyses. Bubble size indicates relative abundance, and fill colour represents the log2 fold change relative to the original faecal sample. Only taxa with fold change ≥ 1 and relative abundance ≥ 10% in at least one enriched consortium are shown at species level for clarity, except the Enterococcus genus and Negativicutes class. For included taxa, individual data points with relative abundance <0.05 are not displayed. b Abundance of selected CAZyme genes mediating the degradation of arabinoxylan (AX), pectin (PEC), or xyloglucan (XG) in faeces and respective fibre-enriched consortium. Differences in conditions (faeces vs. enrichment) were assessed using paired two-sided Wilcoxon signed-rank tests. Medians of each condition are indicated, and the corresponding P values are shown. Genes exhibiting a ≥ twofold increase are marked with an asterisk (*). Gene abundance was normalised using a TPM (transcript per million) approach with read counts divided by gene length and scaled to one million across genes. Next, we sought to rationalise the observed enrichment profiles on the selected fibres. Genes encoding fibre-specific CAZymes were identified in both the faecal samples and fibre-enriched consortia (Fig. 5b and Supplementary Fig. 6d–f). The enrichments generally increased the abundance of CAZyme markers for target fibres significantly as compared to the faecal samples, except for PEC and GA, likely due to the already high abundance of pectin degradation genes in distinct faecal consortia (Fig. 5b and Supplementary Figs. 6f). Interestingly, distinct endo-enzymes, critical for polysaccharide backbone degradation, were only observed after enrichment, for example, xylanase (GH10) in infant C and D at pre-weaning, pectin lyases in infant C at pre-weaning and mannanases in infant A and E at early weaning. Despite the general observed increase in genes encoding endo-enzymes, this was not significant at gene level after multiple test correction, except for certain xylanolytic enzymes (Supplementary Figs. 9 and 10). In summary, these analyses demonstrated that fibre-utilising GM species are already present during exclusive breastfeeding, plausibly owing to their HMO utilisation capacity. Expansion of these species and further selection of bacteria with similar catabolic profiles are likely promoted as weaning progresses. Dual HMO-fibre utiliser are competitive on fibre-HMO weaning mimic nutrients in vitro To assess the competitiveness of dual HMO-fibre utilisers on weaning mimic nutrients, we selected five well-characterised GM strains: B. infantis (HMO specialist), B. adolescentis (starch/fibre oligomer utiliser), and three dual HMO-fibre utilisers Roseburia species, viz. R. intestinalis, R. hominis and R. inulinivorans. To design the competion assay, we generated growth profiles of each strain on glucose, lactose, xylo- and manno-oligomers, as well as the fibres arabinoxylan (AX) and galactomannan (GMN), the latter also used as a common food additive (Supplementary Fig. 11). Equal proportions of each strain were grown on glucose, then the consortium was propagated in four consecutive growth steps on: HMOs, fibres (blend of AX and GMN) and alternating HMO/fibre conditions (Supplementary Fig. 12 and Supplementarty Results). A 16S rRNA gene amplification analysis revealed that the Bifidobacterium genus dominated the HMO consortium (~ 65%), but that Roseburia members also increased their relative abundance by about a third as compared to glucose (Supplementary Fig. 13). The same analysis showed that Roseburia members dominated (> 92%) the fibre consortia. Interestingly, the alternating HMO/fibre growth steps underlined the competitiveness of the Roseburia dual utilisers on both HMOs (45–49%) and fibre (94%) (Supplementary Fig. 13). Next, we grew the same consortium on HMOs, the AX/GMN fibre blend as above, or a mixture of HMOs and the fibres and performed a comparative metaproteomics analysis of the three nutritional conditions (Supplementarty Results). Proteomics-derived relative abundance data showed, consistent results with our cohort faecal enrichment on HMOs (Fig. 4a), that B. infantis was the dominant species in the HMO culture (~ 55%) (Supplementary Fig. 14). Roseburia dual fibre-HMO utilisers showed their competitiveness in both the HMO and HMO-fibre mixture conditions (relative abundance 33–41%) and were dominant in the fibre condition (98%) (Supplementary Fig. 14). Then, we analysed differentially regulated proteins in response to the different substrates (Supplementarty Results). The HMO utilisation markers (GH136 and GH112) from R. intestinalis and R. hominis were highly upregulated in the HMOs cultures, together with α-fucosidases (GH29 and GH95) (Fig. 6a, b and Supplementary Fig. 2, Source Data 3). α-Fucosidases (GH29 and GH95), a sialidase (GH33) and a GH2 β-galactosidase (lactase), all required for the depolymerisation of fucosylated- and sialylated-lactoses, were also upregulated by B. infantis, supporting the preference to these major HMOs. The Roseburia species upregulated their mannan-degrading enzymes on the fibre blend (Fig. 6a), revealing preference for mannan (GMN) over xylan (AX). Metaproteomics analysis of mini-consortium grown on either HMOs, fibre or a mix of both nutrients. The analysed metaproteomes were prepared from cultures of the mini-consortium comprising the HMO specialist Bifidobacterium infantis, the fibre oligomer utilising Bifidobacterium adolescentis, as well as dual HMO-fibre utilisers from the Roseburia genus following growth on HMOs, fibres (1:1 blend of arabinoxylan and galactomannan) or an equal mix of both substrate types. a–c are the volcano plots showing −log10(false discovery rate, FDR-adjusted P values) and log2(fold change) of the metaproteome comparisons between of the fibre vs. HMO, fibre vs. HMO-fibre mix, and HMOs vs. HMO-fibre mix, respectively. The colour and geometric shape indicate the CAZyme taxonomic origin and specificity, respectively. The CAZymes families targeting HMOs, arabinoxylan or galactomannan are labelled, in addition to the top 15 most upregulated CAZymes (ranked by |log₂(FC)| × −log₁₀(P adjusted). Thresholds for FDR-adjusted P and log2(FC) are indicated, with statistically upregulated and downregulated proteins in the right and left top quadrants, respectively. The data are from three biological replicates (n = 3). Strikingly, Roseburia members exhibited a similar upregulation pattern of HMO degradation enzymes on both the fibre-HMOs mix and the HMOs conditions (Fig. 6a, b), indicating that these species utilise HMOs both in the presence and absence of fibre. The modest upregulation of fibre utilisation enzymes in the fibre vs. fibre-HMOs mix consortia (Fig. 6b) indicated that the fibre degradation apparatus is similarly upregulated in the presence or absence of HMOs. Collectively, the above analyses provide compelling evidence that distinct dual utilisers, including some observed in our cohort samples, e.g., R. intestinalis, have evolved to competitively and simultaneously utilise both HMOs and fibres, a nutrient combination uniquely relevant to the weaning transition. Maternal GM grows on HMOs Selection of dual HMO-fibre utilisers during weaning implies the retention of HMO utilisation pathways in the mature GM of adults. To investigate this, we enriched the maternal faecal samples on HMOs. Enrichments on mucin aimed at assessing if growth on HMOs is driven by mucin degraders, due to the structural overlap of glycans from both substrates. All faecal samples sustained growth on HMOs. Surprisingly, growth on mucin was markedly weaker than on HMOs, except for two mothers (D and E) (Fig. 7a). The maternal faecal microbiome differed significantly (PERMANOVA, R² = 0.028, P = 0.0003) in overall community composition from both infant faecal samples and substrate-enriched consortia thereof (Supplementary Fig. 15a). Likewise, consortia enriched on HMOs (R² = 0.018, P = 0.0056) were significantly distinct from the other maternal sample types. High abundance of Bacillota and Bacteriodota and low Actinomycetota abundance was observed as opposed to the infant samples (Fig. 2a and Supplementary Fig. 15b). Bacteroides spp. were abundant in all mothers but were depleted after enrichments on HMOs and were markedly reduced in the mucin cultures (Fig. 7b and Supplementary Fig. 15b). A considerable increase of Actinomycetota, was observed in the HMO cultures. Bifidobacterium common HMO utilisers (except B. adolescentis in one sample), e.g., B. pseudocatenulatum, B. catenulatum, were markedly enriched (Fig. 7a). B. bifidum abundance increased in consortia enriched on HMOs and mucin in two samples (mothers D and E) (Fig. 7b). A diversity of Bacilli and Clostridia were also enriched on HMOs. Generally, enrichment of species was specific to either HMOs or mucin, although certain taxa were enriched on both substrates. These findings underscore that specific adult GM are adapted to preferential growth on HMOs rather than mucin O-glycans. Bacteroidota were not competitive on HMOs, consistent with their delayed expansion until solid food becomes the prime GM nutrient. Maternal faecal consortia enriched on HMOs and mucin. a Changes of pH after a 24 h enrichment on HMOs and mucin. P values were calculated using two-tailed Welch’s test (unequal variance assumed) and adjusted using Benjamini–Hochberg false discovery rate (FDR) method. Asterisks indicate FDR-adjusted P values: ns, P.adj ≥ 0.05; * P.adj < 0.05; ** P.adj < 0.01; *** P.adj < 0.001; **** P.adj < 0.0001. Samples with P.adj < 0.05 and pH decrease > 0.5 are framed in brown squares. The mean, standard deviation, and P.adj values are provided (Source Data 1). b Microbial community composition in faecal samples and enriched consortia based on metagenomic analyses. Bubble size and fill colour represent relative abundance and log2(fold change) relative to the original faecal sample, respectively. Only taxa with fold change ≥ 1 and relative abundance ≥ 1% in at least one enriched consortium are shown. For included taxa, individual data points with relative abundance <0.05 are not displayed. c Abundance of HMO degradation genes in faecal samples and consortia enriched on HMOs and mucin. The abundances of faeces and HMOs were compared using paired two-sided Wilcoxon signed-rank tests. Medians of each condition are indicated, and P value is shown. Genes exhibiting a ≥ twofold increase are indicated by a white asterisk (*). Gene abundance was normalised using a TPM (transcript per million) approach with read counts divided by gene length and scaled to one million across genes. d Selected metagenome assembled genomes (MAGs) from both infant and maternal faeces as well as the enriched consortia, from the top 20 most abundant species in infant and maternal faeces, including all from Bifidobacterium. Clostridium butyricum and Roseburia intestinalis were additionally included due to their high abundance in the enriched consortia. Top five abundant CAZyme families per substrate are shown. Then, we analysed HMO-degradation CAZyme markers in the maternal GM. The abundance of β-galactosidase (GH42) genes was increased in the HMO consortia, whereas counterparts encoding GH112 displaying dual activity on the HMO I block (LNB) and the mucin-derived galacto-N-biose (Gal-β1,3-GalNAc) were observed in consortia enriched on both HMOs and mucin (Fig. 7c). For a strain-level overview, we analysed the 214 high-quality MAGs ( > 90% completeness and <5% contamination), retrieved from infant and mother faeces as well as enriched consortia. Two MAGs from B. infantis and one B. bifidum encode mainly HMO-active enzymes (Fig. 7d and Supplementary Fig. 16a). Only few infant consortia enriched on dietary fibres contained B. infantis and B. bifidum at lower abundance than the corresponding faecal samples, consistent with the poor competitiveness of both taxa on plant fibres. By contrast, B. longum was observed in fibre enrichment of infant samples (Fig. 5a) and B. pseudocatenulatum was observed in maternal HMO enrichments, consistent with dual HMO-fibre utilisation of both taxa (Fig. 7b, d). The remaining Bifidobacterium MAGs showed mainly plant fibre utilisation potential, besides the presence of one or multiple GH42 genes, which may target either plant fibres or HMOs, e.g., LNT (Fig. 7d and Supplementary Figs. 15c and 16a). The MAGs of butyrate producers in maternal and infant samples, e.g., A. rectalis (Eubacterium rectalis) and R. intestinalis, encode both plant-derived fibres and HMOs/mucin utilisation genes (Fig. 7d and Supplementary Fig. 16a, b). The Bacillota enriched on HMOs include core GM Clostridium and Ruminococcus species, which both possess HMO and fibre degradation CAZymes (Fig. 7d). In summary, our work unravels the growth of distinct adult GM bacteria on HMOs, consistent with the metagenomic CAZyme analyses showing dual fibre-HMO utilisation profiles for prevalent GM species, consistent with the in vitro consortium competition and metaproteomics analyses. To investigate the growth on HMOs in more detail, we isolated 258 strains from maternal faecal samples, enriched on HMOs. Re-growth and re-plating resulted in 188 isolates showing significant growth (P < 0.01) relative to no-carbohydrate controls. Then, we grew 137 of these isolates on 1% (w/v) HMOs at 37 °C for 24 h (Supplementary Figs. 17 and 18a, b). The isolates clustered into five HMO groups, based on their HMO depletion profiles on 22 selected HMOs (Supplementary Figs. 18c, 19 and 20, and Supplementarty Results). Group I had the highest growth median, which was associated with the depletion of abundant undecorated LNT/LNnT (lacto-N-neotetraose) HMO backbones (Supplementary Figs. 1 and 20), including those branched with a β1,6-N-acetylglucosamine unit (Supplementary Fig. 19). By contrast, groups II and III lacked activity on LNT/LNnT, but removed α2,3- and/or α2,6-linked sialic acid (Neu5Ac) caps, correlating to lower growth median than group I (Supplementary Fig. 18c). In summary, the depletion of common HMO mainchains, correlated to efficient growth by the maternal isolates. Interestingly, the HMO utilisation markers (GH112 and GH136) were identified in bifidobacteria, but also in several Clostridia from the maternal GM. Due to the small size of the cohort, we wanted to assess if these markers are broadly found in core adult GM species. To this end, we analysed the taxonomic distribution of both families in a human GM catalogue harbouring 286,997 genomes48. Surprisingly, Actinomycetota (including Bifidobacterium) harboured the minority of both GH112 and GH136 sequences (Supplementary Fig. 21). Phylogenetic analysis demonstrated the broad distribution of both CAZyme families in core GM species (Supplementary Fig. 22a–d), whereas the conservation of active site residues indicated high selection pressure to maintain function in GH112 and GH136 (Supplementary Fig. 21e, f). Most GH112 (84 %) and GH136 (60 %) sequences were from Bacillota (Supplementary Fig. 22), consistent with the growth of Roseburia and Eubacterium species on HMOs mediated by these CAZymes41. The presence of both CAZyme families in Eubacteriales (e.g., Ruminococcus) and Bacilli (e.g., enterococci) was also in agreement with the enrichment of several species from these taxonomic groups on HMOs in both mother and infant samples (Figs. 4a and 6b, d, Supplementary Fig. 21, and Supplementary Table 8). These finding are consistent with the dual HMO-fibre utilisation profiles of core adult GM groups, further underscoring the role of HMOs, in addition to dietary fibres, as a selection cue during the weaning transition. Discussion Breastfeeding for 1–1.5 years has been common until shorter lactation was inferred in the 19th century Western societies, elevating child mortality rates49. The World Health Organization (WHO, https://www.who.int) recommends at least 6 months of exclusive breastfeeding, followed by a weaning period up to 2 years of age50, but less than half of the infants under 6 months are breastfed according to WHO. This study focuses uniquely on weaning by sampling three different stages, with pre-weaning as a reference. We demonstrate that fibre-utilising bacteria colonise infant guts before introduction of solid food. This notion is supported by the growth of faecal consortia from exclusively breastfed Milkome cohort infants on dietary fibres and conservation of fibre utilisation genes in pre-weaning infant GM from both Milkome and a large Swedish cohort1. Concordance between both cohorts (111 pre-weaning metagenomes) indicates the generality of these findings. The ability of fibre-utilisers to grow on HMOs, albeit less competitively than the specialist B. infantis, allows for their persistence at low abundance during exclusive breastfeeding. Growth on fibres and the significant increase in genes mediating fibre degradation in the fibre-enriched consortia provide strong evidence for the presence of fibre utilisers during exclusive breastfeeding. Weaning triggers a gradual and organised expansion of dual HMO-fibre utilisers, commensurate with increased solid food to mother’s milk ratio. This expansion is evidenced by the increase in fibre utilisation CAZymes associated with the changes in the community structure in late weaning (Fig. 2d–g), as well as the more prevalent growth on fibre in early and late weaning samples as compared to pre-weaning (Fig. 3b). This is also evident from the additional taxonomic groups, especially Clostridia, that become responsive to fibre enrichments (Fig. 5a). Further support to the nutritional adaptions is provided by the in vitro consortium competition assay. Notably, butyrogenic HMO-fibre utilisers target type 1 HMO backbones41, but not the major fucosylated or sialylated lactose fraction that B. infantis prefers, which explains the relative competitiveness of Roseburia dual utilisers on both HMOs and fibre against both B. infantis and B. adolescentis. This strategy has likely evolved to select for core GM species and suppress random colonisation by environmental microbes during the critical maturation transition. At weaning completion, the gut mucosal surfaces become densely colonised by a core GM community, which allows for more liberal and diet-driven diversification of infant GM. This proposed mechanism is in line with the dominance of HMO-utilising bifidobacteria in late weaning samples, key to the maturation of infant immune system via aromatic lactate metabolite secretion, e.g., by B. infantis51. Despite the abundance of HMOs in weaning diet, hitherto the maturation of the GM has been largely attributed to the introduction of solid food52–54. At present, bifidobacteria are classified either as adult or infant species and focus has been either on their HMO or fibre utilisation. Recently, utilisation of both HMOs and plant-derived fibres was reported in B. catenulatum subsp. kashiwanohense55, having low prevalence in infants and adults55. Similarly, a dual HMOs-fibre utilising B. longum clade was recently shown to be prevalent in weaning infants from Bangladesh, but not other geographic locations56. Here we emphasise that dual HMO-plant fibre utilisation applies to other prevalent bifidobacteria, e.g., B. longum, which possess CAZymes targeting both components of the weaning diet40,57. A hitherto not reported finding is the preponderance of HMO type 1 degrading enzymes of GH112 and GH136 in non-bifidobacterial core GM groups. Previously, we reported that primary fibre degraders from Roseburia and Eubacterium grow on the abundant HMO type 1 backbones41, mediated by both GH112 and GH136 enzymes, which also were deployed in the competition assay with B. infantis (Fig. 6). Another metabolic route involves cross-feeding on sialic acid from HMO (or mucin) degradation, via a sialic acid (Neu5Ac) importer present in certain Clostridia, e.g., Faecalibacterium and Ruminococcus58 species, both showing increased relative abundance in the mucin enrichment. Increased butyrate production by early colonising Clostridia during weaning has been proposed to be important for anti-inflammatory immune modulation59. By comparison, the poor competitiveness of Bacteriodes glycan utilisation generalists during growth on HMOs is consistent with their delayed expansion towards cessation of breastfeeding, as diverse solid foods dominate infant nutrition. The selection of dual HMO-fibre utilisers during weaning, notably Clostridia, implies some of these groups are likely retained in the adult GM. The growth of maternal faecal samples on HMOs and degradation of type 1 HMO blocks by maternal isolates are consistent with the persistence of distinct dual utilisers that plausibly retain fitness to the adult GM through fibre utilisation. The global analysis of HMO utilisation markers in the human GM reference catalogue and active site analyses of the HMO marker CAZymes further support the wide-spread capacity to utilise HMOs by human adult core Bacillota species. In conclusion, this study highlights the role of HMOs in the maturation of infant GM during weaning. The small size of the Milkome cohort and the lack of additional information to analyse confounding factors related to the weaning GM variation are clear limitations. Corroboration of our findings in larger cohorts, with additional sampling and more detailed information on diet and other relevant factors is required to shed more light on the expansion of dual utilisers in the weaning transition. In addition, extending these analyses to more geographically, ethnically and culturally diverse populations would be instrumental in highlighting potential differences in GM maturation. Nonetheless, the fresh faecal enrichment strategy across multiple weaning diet components, combined with biochemical insight on HMO/fibre utilisation, phylogenetic analyses based on the human GM catalogue and metagenomic data from a larger infant cohort, collectively support the broad relevance of the findings from this work. We reveal the presence of dual HMO and fibre-utilising bacteria before the introduction of solid food diet. This core weaning GM gradually blossoms and expands following the onset of weaning. The ability to utilise HMOs (specifically type 1 blocks) besides diverse fibres common in solid food is plausibly an important selection cue contributing to the organised colonisation and expansion of core GM taxa during weaning. A consequence of this previously unrecognised evolutionary adaptation is the persistence of the HMO utilisation trait in adult GM. Perturbations in the GM maturation orchestrated by the natural weaning diet is likely to have implications on the health outcomes of the host. Additional research is needed to shed light on how timing, rate of weaning, and nutrition of both the mother and infant-dyad affect GM maturation and potentially the health trajectory of the weaning infant. Methods The Milkome cohort The study, centred on the Milkome cohort, was approved by the Ethical Committee of the Capital Region of Copenhagen, Denmark (study registration ID H-21067700) and registered at Clinicaltrials (https://clinicaltrials.gov, registration ID NCT07026526). The cohort comprises seven mother-infant dyads who were recruited at Copenhagen University Hospital, Rigshospitalet (Copenhagen, Denmark) between August 2022 and January 2024, based on informed consent obtained from the mothers (maternal consent) and the parents (infant consent). Detailed information about volunteer inclusion and exclusion criteria are summarised in Supplementary Table 1. Sample collection Infant and maternal faeces were collected by participants at three distinct weaning stages: Pre-weaning (exclusive breastfeeding), early weaning (seven out of ten consecutive meals given to the infant within a period of 1–2 days, would be mother’s milk and three would be solid food), late weaning (three out of ten consecutive meals given to the infant within a period of 1–2 days would be mother’s milk and seven would be solid food). An overview of the sample collection scheme and analyses is reported in Fig. 1. Faecal sample collection was designed to maximise viability and recovery of the GM via (1) preserving an anaerobic headspace in sample containers, (2) cold storage and minimal time ( < 1.5 h) from collection to laboratory delivery, and (3) enrichment of faecal consortia from the freshly received samples. Faecal samples were collected using a commercial anaerobic microbiome collection kit (GutAlive, Gijón, Spain) and processed within 1.5 h of collection in an anaerobic chamber (Don Whitley, Bingley, UK or Coy Laboratory Products, Grass Lake, MI, US). Upon arrival, samples were equilibrated anaerobically at 37 °C for 30 min and thoroughly homogenised. One 0.5 g aliquot was immediately processed (vide infra), while the remaining aliquots were stored at −80 °C. Preparation of Human milk oligosaccharides (HMOs) and porcine gastric mucin growth substrates Pooled human milk samples (purchased or donated from Hvidovre hospital, Hvidovre, Denmark) were used to purify HMOs by solid phase extraction (SPE) as previously described60. In short, milk fat was removed by centrifugation (10,000 × g, 30 min, 4 °C), and proteins were precipitated by the addition of 3 volumes of ice-cold ethanol (99% v/v) and incubation for 18 h at 4 °C. Lactose in the protein-free supernatants was digested with a commercial ß-galactosidase from Kluyveromyces lactis (Sigma Aldrich) (20 U mL−1, 3 h at 37 °C). Then, residual lactose and monosaccharides were removed by SPE using an in-house column packed with a 100 × g of activated charcoal (20–40 mesh particle size, Sigma Aldrich). The column was activated with one column volume (CV) of 80% (v/v) acetonitrile (ACN) in milliQ water and equilibrated with two CVs of 4% (v/v) ACN in milliQ water (solution A), before samples were loaded. Residual lactose and monosaccharides were removed by washing with six CVs of solution A and HMOs were eluted with two CVs of 40% (v/v) ACN in water. A rotary evaporator was used to remove residual ACN before the HMOs were freeze-dried. The residual lactose concentration in the eluted HMOs was determined by high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD). For this, samples (10 µL) were injected into a 3 × 250 mm CarboPac PA200 column (Thermo Fisher Scientific) and a 3 × 50 mm CarboPac PA200 guard column (Thermo Fisher Scientific) installed on an ICS-5000 (Dionex) system for the separation. The bound saccharides were eluted with a stepwise linear gradient of sodium acetate: 0–7.5 min of 0–50 mM, 7.5–25 min of 50–150 mM and 25–35 min of 150–400 mM, at a flow rate of 0.35 mL min−1 and a mobile phase of constant 0.1 mM NaOH. Standards (0.01–0.5 mM) of lactose in milliQ water were used to quantify residual lactose as described above. The analysis was performed in independent triplicate experiments. The lactose-depleted HMO preparation (0.42 ± 0.02% (w/w) lactose in the HMO preparation, corresponding to 61 µM lactose (deemed insufficient to markedly impact the growth outcomes) in standard growth medium containing 0.5% (w/v) HMOs, was stored at −20 °C until use for bacterial growth, enrichment and isolation experiments. Porcine gastric mucin type III (PGM) was further purified as previously described61. In short, mucin was added to a final concentration of 2.5% (w/v) in phosphate-buffered saline, pH 7.4 (PBS) and stirred for 20 h at room temperature. Insoluble material was removed by centrifugation (10,000 × g, 30 min, 4 °C) and soluble mucin was precipitated from the supernatants by adding 3 volumes of ice-cold ethanol (99 % v/v) for 18 h at 4 °C. Next, purified soluble mucin was dialysed in milliQ using a SpectraPorTM 50 kDa molecular weight cut-off membrane (Fisher Scientific, Waltham, MA, USA), freeze-dried, and stored at −20 °C. Enrichment of faecal consortia on selected carbohydrate sources Initially, we tested the YCFA medium on the first received samples; however, high growth on the no carbohydrate-supplemented control prompted us modify the medium to reduce baseline growth. To this end, yeast extract content was reduced by 2.5-fold in the first enrichment (mYCFA). Fresh faecal samples (0.5 g) from mothers and infants were diluted (1:10 w/v) in mYCFA medium62 (0.1% (w/v) yeast, 1% (w/v) Casitone and 0.05% (w/v) PGM (Supplementary Table 3) and thoroughly homogenised for 5 min using a vortex mixer. This medium was additionally designed to mimic the in vivo presence of mucin, which may serve as an adhesion surface for distinct GM groups. Following sedimentation of insoluble particles, 400 µL of supernatant was transferred to a fresh tube and subsequently diluted three additional times 1:10 (v/v), resulting in a final dilution of 10−3 in mYCFA. The diluted samples were used to inoculate 6.5 mL mYCFA medium, supplemented with 0.5 % (w/v) carbohydrates (Supplementary Table 4) sterilised by filtration (0.5 µm filters) or standard autoclaving (for insoluble carbohydrates), to a final sample dilution of 10−5. The cultures were incubated in technical duplicates for 24 h at 37 °C in an anaerobic chamber (Don Whitley, Bingley, UK or Coy Laboratory Products, Grass Lake, MI, US) and bacterial growth was monitored by measuring OD600 and pH. Enriched bacterial consortia were supplemented with glycerol to a final concentration of 25% (v/v) before freezing at −80 °C. To further optimise the specificity of the carbohydrate enrichments and reduce potential protein-mediated growth by proteolytic bacteria, the consortia from the first enrichment were subjected to an additional enrichment round (Enrichment II) on the same carbohydrate sources, using a 50% reduced Casitone content (0.5% w/v) medium (mYmCFA) (enrichment II, Supplementary Tables 4–6). For this, pre-cultures were prepared by inoculating 1 mL mYmCFA medium containing 0.5% (w/v) corresponding carbohydrate (1% (w/v) for HMOs), with 40 µL of each cryopreserved enriched consortium (Fig. 1b). Pre-cultures were grown to mid-late exponential phase (OD600 0.5–0.9), then used to inoculate cultures for two experimental setups: (a) metagenomics shotgun sequencing and (b) carbohydrate utilisation profiles. For metagenomic shotgun sequencing, cultures were inoculated in either 10 mL of fresh mYmCFA with 1% (w/v) HMOs or 20 mL of fresh mYmCFA with 0.5% (w/v) mucin or polysaccharides to accommodate the lower growth mucin or polysaccharides (single experiment). These cultures were harvested after 24 h, except for those supplemented with HMOs, which were harvested after 8 h, due to more rapid growth. Following centrifugation (4000 × g, 15 min, 4 °C) cells were washed twice with ice-cold sterile PBS and then stored at −80 °C for DNA extraction. For carbohydrate utilisation profiling (Fig. 2), cultures were inoculated into 96-well microtiter plates in 280 µL of fresh mYmCFA containing 0.5% (w/v) carbohydrates (1% (w/v) for HMOs), each to a starting OD600 = 0.05. Growth, performed in technical triplicates, was monitored by OD600 using a BioTek Epoch 2 microtiter plate reader (Agilent, Santa Clara, CA, USA) and pH measurements after 24 h, which were then compared to non-inoculated media. Isolation and growth of bacterial strains For isolation, HMO-enriched consortia from maternal samples (sample: early weaning of the infant) were grown in 2 mL mYmCFA supplemented with 1% (w/v) HMOs to late exponential phase (OD600 0.7–0.9). Next, the enriched cultures were serially diluted 200-fold in max (1:10 steps (v/v)) in sterile mYmCFA to a final dilution of 10−3 and thereafter spread (20 µL) on mYmCFA agar plates containing 1% (w/v) HMOs. Plates were incubated anaerobically at 37 °C for 2–3 days. From each maternal sample, 50 single colonies were randomly picked and re-streaked twice on a fresh mYmCFA, supplemented with 1% (w/v) HMOs agar plate. The isolates were then transferred to liquid mYmCFA media supplemented with 0.5% (w/v) glucose, 0.5% (w/v) lactose and 0.5% (w/v) maltose, grown for 18 h (single experiment), and stored at −80 °C in a 25% (v/v) glycerol suspension. To explore the HMO utilisation capacity, cryopreserved isolates were grown in 1 mL mYmCFA supplemented with 0.5% (w/v) each of glucose, lactose and maltose to mid-late exponential phase (OD600 0.5–0.7). These pre-cultures were used to inoculate 200 µL mYmCFA medium with 1% (w/v) HMOs to a starting of OD600 of 0.05 and growth was followed by measuring OD600 at 0 and 24 h. Then, cells were pelleted by centrifugation (20,000 × g, 15 min, 4 °C) and supernatants were stored at −20 °C for LC-MS/MS glycomic analysis of HMO depletion profiles. Analysis of HMO utilisation profiles using LC-MS/MS For HMO utilisation profiling, supernatants from the best growing maternal isolates in mYmCFA medium with 1% (w/v) HMOs, based on OD600 values after 24 h, were analysed by LC-MS/MS as previously described in ref. 63. In short, samples of 40 µL culture supernatant containing 5 µg µL−1 HMOs were reduced with 0.5 M NaBH4 and 20 mM NaOH at 50 °C overnight and desalted using ZipTip C18 tips (Millipore) packed with a cation exchange resin (AG50WX8, Bio-Rad). After drying (SpeedVac), residual borate was removed by repetitive methanol additions and evaporations. Prior to mass spectrometry, samples were reconstituted in 100 µL milliQ water containing 20 pmol µL−1 3ʹ-sialyl-N-acetyllactosamine (Dextra Lab, UK) and analysed by liquid chromatograph-electrospray ionisation tandem mass spectrometry (LC-ESI/MS) using a 10 cm × 250 µm I.D. column, packed with 5 µm porous Hypercarb graphitised carbon particles (Thermo-Hypersil, Runcorn, UK). Glycans were eluted using a linear gradient of 0-80 % (v/v) acetonitrile in 10 mM NH4HCO3 over 46 min at a flow rate of 10 µL min−1. The samples were analysed in negative ion mode on an LTQ linear ion trap mass spectrometer (Thermo Electron), with an IonMax standard ESI source equipped with a stainless-steel needle kept at −3.5 kV. Compressed air was used as the nebuliser gas. The heated capillary was kept at 270 °C, and the capillary voltage was −50 kV. A full scan (m/z 380–2000, two micro scans, maximum 100 ms, target value of 30,000) was performed, followed by data-dependent MS2 scans (two micro scans, maximum 100 ms, target value of 10,000) with normalised collision energy of 35%, isolation window of 2.5 units, activation q = 0.25, and activation time 30 ms). The threshold for MS2 was set to 300 counts and data were processed using Xcalibur (v 2.0.7). The peak area (the area under the curve (AUC)) of each glycan structure was calculated using the Progenesis QI software (Nonlinear Dynamics, Waters Corp., Milford, MA, USA). The AUC of each structure was normalised to the total AUC and expressed as a percentage. The LC-ESI/MS raw data have been deposited in Glycopost under the accession number GPST000504. For the HMO utilisation analysis, a subset of 8 unsubstituted, 9 fucosylated, 5 sialylated HMOs, and 3 putative HMO degradation products were selected, and their relative abundances in individual samples after 24 h of growth were compared against ten independently analysed HMO controls (non-inoculated mYmCFA supplemented with 1% (w/v) HMOs). Changes in relative abundance profiles of the HMOs were analysed and clustered as a metric for similarities of utilisation and/or enzymatic modification amongst the individual isolates, hierarchical clustering was performed using the “group average” clustering method and Euclidean distance as the distance measure (OriginPro 2023 v10.0.0.154). Shotgun metagenomic sequencing, assembly, annotation and 16S rRNA analysis DNA from faeces and enriched cultures was extracted using PureLink Microbiome DNA purification Kit (Fisher Scientific) according to manufacturer’s manual. MGI DNBSEQ sequencing libraries were prepared using the MGIEasy Fast FS DNA library Prep set (MGI Tech Co., Ltd., Shenzhen, China) according to the manufacturer’s protocol. Libraries were sequenced on DNBSEQ-T7 (MGI Tech Co., Ltd.) using the DNBSEQ-T7RS High-throughput Sequencing Set (FCL PE150) (MGI Tech Co., Ltd.) to yield about 40 million paired-end reads per sample. Sample preparation and the whole metagenome shotgun sequencing were performed in Copenhagen. High-performance computing resources were used for all sequencing read analyses described in this section64. The quality of sequence reads was assessed using FastQC v0.11.965 and trimmed with Trimmomatic v0.3866 (parameters: leading 9, trailing 3, sliding window 4:15, minlen 36). Reads mapped to the human reference genome (GRCh38.p148) were removed using Bowtie2 v2.5.267 for stool samples. Microbial taxonomic profiles were predicted with MetaPhlAn 4 v4.1.068 using the Bifidobacterium longum subspecies-resolved database69, which converted to GTDB taxonomy r207. Taxa with relative abundance below 0.01% were excluded. Filtered reads were assembled using MEGAHIT v1.2.970 in default mode. Proteins were annotated on the assembled contigs with BAKTA v1.9.371, and carbohydrate-active enzymes (CAZymes)37 were predicted with dbCAN v4.1.438. To quantify CAZyme family and subfamilies abundances for each sample, filtered reads were aligned to coding sequences and contigs using BWA v0.7.1772. Mapped reads were sorted and converted to BAM files using SAMtools v1.2073. Read counts for proteins were calculated with BEDTools v2.30.074, with gene lengths extracted using SeqKit v2.4.075. These counts were normalised using a TPM (transcript per million) approach with read counts divided by gene length and scaled to one million across genes, using dbCAN-utils script from dbCAN v4.1.476. Aligned sequences of glycoside hydrolase family 42 (GH42) were extracted from metagenome-assembled genomes (MAGs) from maternal faeces and maternal consortia enriched on HMOs. Multiple specificities of GH42, including LNT-specific β-galactosidases active on Gal-GlcNAc linkages, β-galactosidases active on various Gal-Gal linkages found in plant-derived saccharides, α-l-arabinopyranosidases were annotated to maximise the accuracy of predictions in the lack of CAZy-defined subfamilies in GH42. A multiple sequence alignment and a phylogenetic tree were generated using MAFFT 777, and the tree was visualised using Interactive Tree of Life v678. Metagenomic bins were constructed using Vamb v.4.1.379 and SemiBin v.2.1.080, refined with DAS tool v.1.1.681 and dereplicated with CoverM v.0.7.082 at 98% average nucleotide identity. Bins with > 90% completeness and <5% contamination were selected (CheckM2 v.1.0.283). Taxonomic assignment of the bins was performed with GTDB-Tk v2.3.2 (r_220)84, and proteins and CAZymes were annotated with BAKTA v1.9.371 and dbCAN v4.1.476, respectively. An overview of the whole metagenome sequencing analysis is shown in Supplementary Fig. 23. For the 16S rRNA amplicon analysis of the synthetic in vitro consortium, Raw sequences were processed using QIIME 2 (v2026.1)85. Denoising of sequences was performed with DADA286 to generate amplicon sequence variants (ASVs). Taxonomy was assigned using a Naive-Bayes classifier trained on 16S rRNA reference sequences of the five strains, extracted from NCBI—Bacterial 16S Ribosomal RNA RefSeq Targeted Loci Project. ASVs were filtered to genus levels for downstream analysis. Plots were created in R v.4.5.1 using the ggplot2 package. In vitro competition assay of a synthetic five-strain consortium on an HMO, fibre, and alternating HMO-fibre medium Growth profiling of individual strains on relevant carbohydrate substrates Pre-cultures of the five human gut microbiome bacterial strains Bifidobacterium longum subsp. infantis ATCC 15697, Bifidobacterium adolescentis ATCC 15703, Roseburia intestinalis L1-82, Roseburia hominis A2-183 and Roseburia inulinivorans DSM 16841 were grown overnight in YCFA supplemented with 0.4% (w/v) glucose. Then, the overnight culture of each strain was inoculated in technical duplicates into sterile 96-well microtiter plates (200 µL) at a starting OD600=0.01 in YCFA containing 0.4% (w/v) glucose, lactose, xylooligosaccharides, arabinoxylan (AX), galactomannan (GMN), the major HMO 2ʹ-fucosyllactose (2ʹ-FL), and a complex mixture of HMOs from mothers’ milk. To compensate the partial loss of the major 2ʹ-FL during the HMO purification, this compound was supplemented to the HMO mixture to a concentration of 20% (w/v) to mimic natural HMO mixtures (see Supplementary Table 5 for a description of carbohydrate substrate). Growth was monitored for 24 h at 37 °C in an anaerobic chamber (Don Whitley, Bingley, UK or Coy Laboratory Products, Grass Lake, MI, USA) by measuring OD600 every 15 min using a BioTek Epoch 2 microtiter plate reader (Agilent, Santa Clara, CA, USA). Relative abundance of the synthetic consortium in the competition growth experiment Following growth on glucose, the five above strains were combined in equal proportions to form a defined consortium, which was grown in YCFA supplemented with 0.4% (w/v) glucose from a starting OD600=0.01 at 37 °C in an anaerobic chamber (Don Whitley) until entry into stationary phase. Then, four sequential growth cycles were performed in YCFA with either 0.4% (w/v) glucose, 0.4% (w/v) HMOs (supplemented with 20% w/v 2′-FL), 0.4% (w/v) dietary fibre blend of equal proportions of AX and GMN (Supplementary Table 5), or an alternating HMOs and the dietary fibre blend-supplemented medium. At the start of each cycle, cultures were diluted 100-fold into fresh YCFA representing the four growth conditions, with alternating-substrate cultures initially grown on HMOs to mimic the in vivo starting key nutrition for the GM. Based on growth experiments of the consortium (Supplementary Fig. 10). The first cycle was incubated for 14 h (early lag phase) before dilution into fresh fibre blend-supplemented medium for the alternating-substrate condition. After a 16 h growth (early lag phase) in the second cycle, the alternating-substrate culture was inoculated into the HMO medium for a 14 h third cycle, followed by a final dilution into fibre-containing medium and a fourth 16 h incubation. Control cultures grown exclusively on glucose, HMOs, or dietary fibres followed the same dilution schedule. The growth was performed in technical duplicates (n = 2). At the end of each cycle, 2 × 1 mL samples were collected. Cells were washed twice in sterile PBS and once in Tris-EDTA buffer, pH = 8.0, pelleted by centrifugation, and stored at −80 °C prior to DNA extraction using the PureLink Microbiome DNA Purification Kit (Fisher Scientific) according to the manufacturer’s instructions. The extracted DNA was used for 16S rRNA gene amplification, sequencing analysis. Following the final cycle, the cultures from each condition were diluted 100-fold into YCFA containing 0.4% (w/v) 2′-FL. Growth on 2′-FL was monitored for 24 h as described above, as a proxy for the relative abundance of B. infantis, the only species in the consortium that can grow on this substrate. The 16S rRNA gene amplification analysis was carried out at the DMAC core facility with single technical analysis (n = 1) form each of the technical duplicate cultures above (Technical University of Denmark, Department of Health Technology). In brief, PCR amplification of the V3–V4 hypervariable region of the 16S rRNA gene was carried out in a 50-μL final volume containing 10 ng microbial DNA, 2× KAPA HiFi Hot-Start ReadyMix (Roche, Basel, Switzerland), 16S rRNA gene primers carrying Illumina overhang adapter sequences. The thermal cycle consisted of 3 min at 95 °C, 30 cycles of 30 s at 95 °C, 30 s at 55 °C, and 30 s at 72 °C, and a final 5-min step at 72 °C. PCR products were purified with Agencourt AMPure XP magnetic beads (Beckman Coulter, Brea, CA, USA). Indexed libraries were prepared by limited-cycle PCR with Nextera technology and cleaned-up as described above. Libraries were normalised to 4 nM and pooled. The sample pool was denatured with 0.2 N NaOH and diluted to a final concentration of 8 pM with a 10% PhiX control. Sequencing was performed on an Illumina MiSeq platform using a Nano 2 × 250 bp paired-end protocol, according to the manufacturer’s instructions (Illumina, San Diego, CA, USA) Growth of the consortium on single and mixed substrates for metaproteomic analysis The five bacterial strains used in the consortium above were combined in equal proportions to form a defined consortium, similarly to above. Then, the consortium was diluted to OD600 = 0.02 in YCFA supplemented with 0.4% (w/v) glucose and grown for 8 h at 37 °C anaerobically as above to the late exponential phase OD600 = 0.7–0.8. Cultures were then diluted to a starting OD600 = 0.02 in fresh YCFA containing 0.4% (w/v) of (1) HMOs (with 20% (w/v) 2′-FL, (2) the AX-GMN dietary fibre blend used above, or (3) a mixed HMO-fibre blend containing 0.4% (w/v) of each substrate (total 0.8% (w/v). Three cultures (n = 3) per condition were propagated from the starting glucose culture. The Cultures were incubated at 37 °C until exponential phase monitored by measuring OD600 every 15 min in 200 µL volumes using a BioTek Epoch 2 microtiter plate reader (Agilent, Santa Clara, CA, USA). Cells were harvested by centrifugation, washed three times with sterile PBS, and stored at −20 °C overnight until processing in the metaproteomic analysis. Sample preparation for proteomics Lysis buffer (6 M guanidinium hydrochloride, 10 mM TCEP, 40 mM CAA, 50 mM HEPES, pH 8.5) was added to each pellet, and samples were boiled for 5 min at 95 °C, followed by sonication on high for 5 cycles of 60 s of sonication with 30-s pauses using a Bioruptor Pico sonicator (Diagenode). Then, the samples were centrifuged at 18,000 × g for 10 min, and supernatants were transferred to clean Eppendorf Protein LoBind tubes. Protein concentrations of the supernatants were determined using the BCA Rapid Gold assay (Thermo Scientific) and 20 µg of protein from each condition were taken forward for digestion. Volumes were normalised to 20 µL with lysis buffer, and samples were diluted threefold in digestion buffer (10% v/v acetonitrile in 50 mM HEPES). Four hundred nanograms of LysC protease (Wako) was added to each sample, followed by incubation for 4 h at 37 °C with shaking at 750 RPM on a Thermomixer (Eppendorf). Samples were then diluted to a final tenfold dilution in digestion buffer, and 200 ng of trypsin (MS-grade, Thermo Scientific) was added to each sample before incubation for 18 h and 30 min at 37 °C with shaking at 750 RPM. Samples were acidified by adding TFA to a final concentration of 1% (w/v), and the resulting peptides were desalted using a SOLA SPE plate (HRP 10 mg/2 mL, Thermo). Between each application, solvents were passed through the plate by centrifugation at 350 × g. For each sample, the filters were activated with 200 µL of 100% (v/v) methanol, followed by 200 µL of 80% (v/v) acetonitrile containing 0.1% (w/v) formic acid. The filters were equilibrated twice with 200 µL of 1% TFA (w/v) and 3% acetonitrile (v/v), after which the samples were loaded. Following two washes with 200 µL of 0.1% (w/v) formic acid, peptides were eluted into clean Eppendorf Protein LoBind tubes using 40% (v/v) acetonitrile containing 0.1% (w/v) formic acid. Eluted peptides were concentrated using a Savant SpeedVac SPD120 (Thermo Scientific) and reconstituted in 12 µL of solvent A (2% v/v acetonitrile, 1% w/v TFA), sonicated for 1 min in a water bath sonicator and centrifuged for 10 min at 18,000 × g. Peptide concentrations were determined by NanoDrop measurement using a DeNovix DS-11 FX+ spectrophotometer. MS analysis 500 ng of peptides were loaded onto an µPAC trapping column (Thermo Scientific, 3003438) connected in line with a 50 cm µPAC NEO column (Thermo Scientific, 7500467) using 100% solvent A at 750 bar on a Vanquish NEO UHPLC system (Thermo). The column oven was operated at 50 °C. Peptides were eluted over a 68-min gradient ranging from 4 to 45% solvent B (80% v/v ACN, 0.1% w/v formic acid) at a flow rate of 250 nL min−1. The Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific) was operated using a DIA method. Full MS spectra were acquired at a resolution of 120,000 with a normalised AGC target of 300% or maximum injection time set to auto, over a scan range of 400–1000 m/z. MS2 spectra were acquired at a resolution of 30,000 with an AGC target of 800% or maximum injection time set to auto and a normalised collision energy of 27. Isolation windows were set to 12 m/z with window placement optimization enabled, and the MS2 scan range was set to 145–1450 m/z. MS performance was verified for consistency using the Pierce™ HeLa Protein Digest Standard (Thermo Scientific), and chromatography was monitored to ensure reproducibility. Database search Raw files were analysed using Spectronaut™ (v17.4). UniProt reference database for the relevant taxa (taxon IDs: 391904, 367928, 536231, 585394 and 622312). Dynamic modifications were set to oxidation (M) and acetylation of protein N-termini. Cysteine carbamidomethylation was set as a static modification. All results were filtered to a 1% FDR, and protein quantification was performed at the MS2 level. Protein groups were inferred using IDPicker. Metaproteomic data analysis The MaxLFQ and iBAQ protein abundance values were exported from Spectronaut™ without applying additional normalisation or imputation. To estimate per-sample bacterial relative abundances, only features quantified in at least one replicate in all biological conditions (HMOs, HMOs + fibre, Fibre) were retained for analysis. iBAQ scores were then log₂-transformed, median-normalised, and mean-centered by subtracting the grand mean across all measurements. For each species within each sample, stoichiometry (%) was calculated as the sum of linear scale iBAQ intensities assigned to that species divided by the total iBAQ intensity of the corresponding sample. Significant differences in bacteria stoichiometry between conditions were assessed using two-sided Student’s t-tests. Prior to differential expression analysis, protein features were filtered to retain only those with at least one observed measurement across all samples. To correct for differences in species biomass shifts across conditions, protein-level MaxLFQ abundances were converted into within-species relative abundances by normalising to the total abundance of each species in each sample87. These relative abundances were then log₂-transformed and zero-centre normalised by median. To account for the expected absence of protein measurements for specific bacteria under conditions unfavourable for their growth, proteins missing across all triplicates within a given condition were imputed using a condition-wise downshifted normal distribution (downshift = 0.4, width = 0.3), analogous to the approach implemented in Perseus88. This imputation was applied exclusively for visualisation purposes. For differential expression analysis, linear models were fitted to protein expression data using the limma package89 (v3.64.3). Contrasts of interest were evaluated using empirical Bayes moderation, and the resulting P values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR). A global FDR-adjusted P value (q value) threshold of 0.05 and log₂ fold-change cutoffs of ± 0.585 were applied to define significantly up- and down-regulated protein features. Prediction of CAZyme activity was performed on differentially expressed protein sequences using dbCAN338, with HMMER searches against the dbCAN and dbCAN-sub databases and DIAMOND searches against the CAZy database, applying the default filtering parameters. All downstream analyses were performed using R v4.5.136. Sequence handling was conducted with the Biostrings package v2.76.090, while data visualisation utilised ggplot2 v4.0.091, patchwork v1.3.292, and ggpubr v0.6.293. Microbial diversity analyses All analyses were performed using R v4.3.236. The Shannon index, calculated using the vegan package v2.6-1035, was used as a measure of α-diversity, which reflects the diversity within a single sample by combining richness (the number of species) and evenness (the relative distribution of species). Differences in the Shannon index for both taxonomic diversity and polysaccharide-utilising CAZymes between weaning stages (pre-weaning, early weaning, and late weaning) were compared using a paired Wilcoxon test. Non-metric multidimensional scaling (NMDS) a method for visualising β-diversity (differences in community composition between samples), was conducted on Bray–Curtis dissimilarities of the relative abundance of microbial species, with selected CAZymes as environmental factors. Species-environment (CAZymes) relationships were assessed using envfit (vegan) with 999 permutations, and significant vectors (P < 0.05) were visualised. Public infant cohort data analyses Sequence raw data accession IDs for a Swedish infant-mother cohort1 was retrieved from the curatedMetagnomicData package v.3.16.144, selecting only samples from exclusively breastfed infants delivered vaginally with no reported antibiotic use. The corresponding raw data were downloaded from Sequence Read Archive (SRA) of the National Center for Biotechnology Information (NCBI), using the NCBI SRA Toolkit v.3.0.094, processed and analysed using the same pipeline as applied to the Milkome cohort (Supplementary Fig. 24), as described above. Search for HMO utilising genes in human gut microbiome (HGM) catalogue To investigate the prevalence of HMO-utilising genes in bacteria of the GM, we identified glycoside hydrolase family 112 (GH112) and GH136-like proteins within the HGM protein catalogue, consisting of 170,602,708 proteins clustered at 100% amino acid similarity (v2.0, downloaded in December 2021) from 286,997 genomes48. Hidden Markov model (HMM) profiles for GH112 and GH136, sourced from dbCAN-HMMdb-V11 (downloaded in August 2022)76, were used to identify GH112- and GH136-like proteins via the hmmsearch function from HMMER95 with default settings. Matching sequences were then retrieved, and their coverage relative to the query profiles was calculated. Curation threshold was set as E-value cutoff ≤ 0.001, coverage thresholds of ≥ 80% (GH112) and ≥ 60% (GH136), and maximum and minimum size of 600 and 300 amino acids, respectively. A multiple sequence alignment and phylogenetic trees were generated using MAFFT 777, and the tree was visualised using Interactive Tree of Life v678. Active site conservations of retrieved protein sequences were mapped on the identified active sites of the Bifidobacterium longum JCM 1217 (GH112, PDB ID: 2zuv[A])96 and Eubacterium ramulus ATCC 29099 (GH136, PDB ID: 6KQS[A])41 and illustrated with WebLogo 397. The abundance of GH112- and/or GH136-encoding genes was calculated by mapping the representative proteins from the catalogue to their corresponding clustered proteins and their genomic origins using a provided mapping file48. Statistics, reproducibility and visualisation All statistical analyses and visualisation were performed using R v4.3.236. The study pseudo-anonymous design precluded collection of personal data beyond the stage of weaning, which was included in statistical analyses. Given the exploratory nature of the study, the stringent requirement for fresh sample collection for culture, no statistical method was used to predetermine sample size, which was limited by the recruitment within the available time for the study. The growth of the first enrichment, conducted individually for each donor, was compared to no-carbohydrate-added controls and carbohydrate-supplemented cultures, with data pooled across different weaning stages and maternal samples. Data were pooled across donors at each time point and analysed using generalised linear mixed models implemented in glmmTMB v1.1.1098, with random effects accounting for donor identity, substrate supplementation of the medium, and duplicates at each time point. For enrichment II, growth was compared to no-carbohydrate-added controls and to carbohydrate-supplemented cultures, analysing data per time point for each donor individually using two-tailed t-tests with unequal variances. Differences in the abundance of normalised CAZymes in infant faeces at each time point, stratified by substrate type (e.g., HMOs, host glycans, xylan backbone, and pectin backbone), were assessed using the Wilcoxon rank-sum test implemented in R v4.3.236. Comparisons between faeces and enrichment for dietary fibres (AX, PEC, and XC) and the food additive fibres (GMN and GA) were performed using paired Wilcoxon signed-rank tests36, whereas comparison among faeces, HMOs and mucin were conducted using unpaired Wilcoxon rank-sum tests36 due to unequal number of samples in HMOs and mucin. The data were visualised using ggpubr v0.6.093, which builds on ggplot2 v4.0.191. Individual CAZyme genes associated with dietary fibre, including common food additive utilisation, were used to calculate fold changes between faeces and enrichment. Statistical significance was evaluated using a two-sided Wilcoxon signed-rank test against a null median of zero (μ = 0), and fold-change distributions were visualised per CAZyme using ggplot2 v4.0.191. To assess whether the medians of normalised fibre utilisation genes in exclusively breastfed infants from Swedish cohort1, were greater than zero, P values were calculated using a one-sided Wilcoxon signed-rank test36. Differences in community composition across timepoints within infant faecal samples were evaluated using PERMANOVA using adonis2 from vegan package35 based on Bray-Curtis dissimilarity with 9999 permutations. Additionally, group-specific differences in faecal and enriched consortia from mothers and infants were assessed via PERMANOVA35, with each age and substrate group compared against all others. Plots were generated with ggplot2 v3.5.291 and heatmaps by pheatmap v.1.0.1299. Carbohydrate utilisation profiles of high-quality MAGs were visualised using ComplexUpset v1.3.3100. Ethical considerations Volunteers were enrolled based on informed consent and the study was approved by the ethical committee of the Capital Region under the approval of the Capital Region ethical committee (Study registration ID: H-21067700 and registered in Clinicaltrials.gov under accession ID: NCT07026526). The study comprises a mother-infant dyad. Therefore, all adult cohort volunteers are females, while no restrictions on the sex of infants were applied. Infant sex is not disclosed due to the pseudo-anonymous design. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source data Acknowledgements We express our deep gratitude to all the infants and mothers for their participation in this study. We are grateful to the DTU Computing Center for providing high-performance computing resources. We thank Maria Louise Leth for her contribution in translating the Milkome cohort protocol to Danish and the undergraduate students Sofus Boisen, Nicoline Bølling Bjørnskov Jensen, Barbara Borcic, and Lucija Backov for their work within the project. We want to express our gratitude to Associate. Prof. Manimozhiyan Arumugam and Dr. Camila Alvarez-Silva for their kind help at the early stage of the project. We also would like to thank to Senior Researcher Dr. Jesper Holck for help with the initial HMOs glycomic analysis. We are thankful to Associate. Prof. Mikael Lenz Strube and Lauge Alfasten for testing long-read sequencing. We also thanks Maike Wennekers Nielsen for help with sample preparation and MS workflow for the metaproteomics experiment. Author contributions M.A.H., Y.S. and M.J.P conceptualised this work. L.A. and S.S.K. led cohort recruitment. Y.S. and M.J.P. developed methodology on the microbial work and M.J.P. performed the HMO preparation. C.J. and S.T. provided methodology and infrastructure for the glycomic analyses and K.K. provided methodology and infrastructure for the metagenomic shotgun sequencing. Y.S. and M.J.P. conducted the investigations. Y.S. carried out most of the analyses and visualisations with contribution from M.J.P., C.E., I.C., and S.B. M.J.P. performed the maternal strain isolation, growth, as well as HMO utilisation profile analysis and visualisations. Synthetic consortium experiments were carried out by M.H.B.A. M.V.L. contributed to the metaproteomic MS analysis, and V.T. conducted the downstream metaproteomic analysis with contribution from M.H.B.A., who visualised the synthetic consortium experiments and metaproteomics volcano plots. L.E.S. analysed and visualised the 16S analysis. M.A.H. acquired funding and carried out the project administration. M.A.H. and Y.S. wrote the first draft of the manuscript with input from M.J.P. and C.J., who wrote the sections related to HMO utilisation by mother. All authors contributed to the manuscript editing and approved the final manuscript version. Peer review Peer review information Nature Communications thanks Ranko Gacesa and the other, anonymous, reviewer for their contribution to the peer review of this work. A peer review file is available. Funding The study was funded by a Research Fund Denmark, Natural Sciences (FNU Project 2 grant no. 1026-00386B) to M.A.H. C.J was partially employed on a Novo Nordic Foundation Interdisciplinary Synergy Programme (grant no. NNF22OC0077684). Data availability The metagenomic data from the Milkome cohort have been deposited in the European Nucleotide Archive (ENA) and are publicly available under the project accession number PRJEB97976. The 16S rRNA sequencing data for the in vitro consortium were deposited under the accession PRJEB107780 (Secondary Accession is ERP188667). The glycomics MS raw files have been deposited in the GlycoPOST database under the ID GPST000504. The proteomic data has been deposited in the PRIDE depository under the accession: PXD074160. The Supplementary Information file and the Source Data 1 (Main manuscript), Source Data 2 (Supplementary) as well as Source Data 3 (Proteomics experiment) are available online. Source data are provided with this paper. Code availability Codes used for the study are available at https://github.com/yunso-bio/MILKOME and are linked to Zenodo 10.5281/zenodo.19832054 [https://zenodo.org/records/19832054]. The scripts used for the MS-based proteomics analysis and the relevant figures are available at https://github.com/DTU-Proteomics-Core/2041_HMO_gut_consortium_proteomics and are linked to Zenodo 10.5281/zenodo.19882716 [https://zenodo.org/records/19882716]. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-73297-5. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source data Data Availability Statement The metagenomic data from the Milkome cohort have been deposited in the European Nucleotide Archive (ENA) and are publicly available under the project accession number PRJEB97976. The 16S rRNA sequencing data for the in vitro consortium were deposited under the accession PRJEB107780 (Secondary Accession is ERP188667). The glycomics MS raw files have been deposited in the GlycoPOST database under the ID GPST000504. The proteomic data has been deposited in the PRIDE depository under the accession: PXD074160. The Supplementary Information file and the Source Data 1 (Main manuscript), Source Data 2 (Supplementary) as well as Source Data 3 (Proteomics experiment) are available online. Source data are provided with this paper. Codes used for the study are available at https://github.com/yunso-bio/MILKOME and are linked to Zenodo 10.5281/zenodo.19832054 [https://zenodo.org/records/19832054]. The scripts used for the MS-based proteomics analysis and the relevant figures are available at https://github.com/DTU-Proteomics-Core/2041_HMO_gut_consortium_proteomics and are linked to Zenodo 10.5281/zenodo.19882716 [https://zenodo.org/records/19882716].