B. bifidum was isolated from stool samples and preserved in the culture collection of the Probiogenomics Laboratory (University of Parma, Italy). Stool samples were diluted in physiological solution and plated on Man-Rogosa-Sharpe (MRS) agar (Scharlau Chemie, Spain) containing 0.05% (w/v) L-cysteine hydrochloride and mupirocin (50 mg/L) and incubated in a specialized chamber (Concept 400; Ruskin) under anaerobic conditions (2.99% H₂, 17.01% CO₂, and 80% N₂) at 37 °C for 48 h. For subsequent assays, B. bifidum cells were grown under the same anaerobic conditions in MRS broth. The genome of six B. bifidum strains were decoded in the framework of this study and deposited in GenBank as strains 81B, 98B, 184B, 352B, 361B, and 2350B, under NCBI accession numbers GCF_055790765.1, GCF_055790725.1, GCF_055790805.1, GCF_055790825.1, GCF_055790785.1, and GCF_055790745.1, respectively.

Genome assemblies were generated from sequencing data of B. bifidum isolates in the Laboratory of Probiogenomics collection and from public shotgun metagenomic data of the human gut microbiota. Locally isolated reads were assembled with MEGAnnotator2, while new metagenomes were obtained from public repositories using the METAnnotatorX2 pipeline. SPAdes version 4.3.0 was used in both pipelines, with either “--isolate” or “--meta” invoked, depending on the sample. For both software tools, the default parameters were utilized^[35-37].

The B. bifidum dataset comprised all publicly available B. bifidum genomes retrieved from NCBI, together with an additional 1,106 MAGs reconstructed from 86 projects. Finally, six new genome assemblies derived from bacterial isolates from the Probiogenomics Laboratory collection were added [Supplementary Table 1]. The quality of each genome was evaluated using CheckM2, based on completeness and contamination levels^[38]. Genomes with completeness below 95% and/or contamination above 5% were discarded. Dereplication of the dataset was performed using dRep, with a 99.99% threshold to remove genomic redundancy^[39]. The final B. bifidum dataset includes 1,351 unique genomes [Supplementary Table 1]. We also downloaded all high-confidence genome sequences of HBS and PBS from NCBI^[40] [Supplementary Table 2].

The significant divergence in codon usage bias (CUB) and GC content was analyzed to identify highly confident horizontal transfer gene (hcHTG) between clusters of orthologous genes (COGs) from the RefSeq NCBI genomes of B. bifidum^[41]. The value of CUB was influenced by: Relative Synonymous Codon Usage (RSCU), which is the codon usage compared to the theoretical maximum, Effective Number of Codons (ENC), indicating the specificity required by the gene in the codon usage, the Codon Adaptation Index (CAI), which considers the codon usage compared to highly expressed reference genes, and GC content (expressed as a within-genome z-score). Genes in which all four parameters deviated from the 90% of the genome gene values (above the 90th percentile for RSCU, ENC and GC content and below the 10th percentile for CAI, respectively) were therefore classified as high-confidence horizontally transferred genes (hcHTGs). Percentile thresholds were computed independently for each genome. Across all analyzed B. bifidum genomes, the average threshold values were 0.49 ± 0.01 for the CAI 10th percentile, 34.26 ± 0.39 for the 90th percentile of ENC, 1.28 ± 0.05 for the GC content z-score 90th percentile, and 0.987 ± 0.00 for the RSCU 90th percentile.

Each COG formed by hcHTG was classified as a putative acquired COG. To identify putative COG donors, a BLASTP search was performed against the NCBI amino acid database, excluding matches to B. bifidum and Bifidobacterium sp. proteins. To compare the number of hcHTG events across the groups (B. bifidum, HBS, and PBS), a Kruskal-Wallis test, followed by pairwise Mann-Whitney U tests with Benjamini-Hochberg False Discovery Rate (FDR) correction, was performed. Furthermore, the internal distribution variability and homoscedasticity within each group were assessed and compared using the Fligner-Killeen test, also with FDR correction.

CRISPR arrays and Cas gene clusters were predicted in each of the 1,351 B. bifidum genomes individually using CRISPRCasFinder v4.3.2^[42], run locally in “General” detection mode with the associated Cas-gene detection module. Each predicted CRISPR array was assigned a CRISPRCasFinder evidence level from 1 to 4, reflecting the confidence of the repeat-spacer architecture. All spacer sequences extracted from the predicted arrays were pooled and compared using the blastn-short task against the INPHARED phage genome database^[42]. Matches were retained as significant when the percentage identity was ≥ 90%, the query coverage was ≥90%, and the e-value was ≤ 1e^-5. For each validated hit, we collected the phage genome taxonomy, the corresponding CRISPR array, the CRISPRCasFinder evidence level of the source spacer, and the identity of the matched phage genome.

KEGG (Kyoto Encyclopedia of Genes and Genomes) codes for each COG^[43] were predicted using eggnog-mapper v2^[44]. These annotations were used as input to the KEMET software^[45], which enables evaluation of the completeness of metabolic pathways for each genome. For both software, the default settings were used. Furthermore, the dbCAN3 software^[46] was used to assess the ability to utilize different carbon sources by identifying enzyme classes from the CAZy database^[47] involved in carbohydrate metabolism.

Gene annotations were performed by PROKKA^[48] with default settings. Pangenome studies were performed using the software Roary^[49], allowing the identification of the bifidobacterial core genome (genes present in 99% to 100% of the genomes in our dataset), soft-core genome (genes present in 95% to 99% of all genomes in the dataset), and accessory genome (genes present in less than 95% of all genomes analyzed). Genes’ distribution within Roary’s COGs was based on an identity threshold that varied by dataset (95% for single-species datasets and 80% for multispecies datasets).

The pangenome state was calculated using a custom Python 3 script that processed Roary’s output. The average pangenome progression of the dataset was computed from 1,000 possible permutations. For gene products not clearly annotated by Prokka, such as hypothetical proteins, a protein domain detection was performed with InterProScan using the Pfam 37.2 database^[50,51].

To identify potential operons comprising genes unique to B. bifidum, the PRL2010 strain upstream regions (nucleotides from position -200 to +70) of the unique B. bifidum COGs were analyzed. These regions were searched for RNA polymerase binding sites (RPBs) and transcription factor binding sites (TFBs), and their energetic and structural properties (ESp) were evaluated.

A reference database of RPB sequences was assembled from RNA-seq data reported in previous studies conducted on different strains of the genus Bifidobacterium^[52-56]. For each reference sequence, all possible degenerate variants were generated by systematically replacing one or more nucleotide positions with an N wildcard, producing an exhaustive set of full-length and degenerate query patterns. These patterns were used to scan the upstream regions of all PROKKA-annotated genes for sequence matches. For the purposes of the present study, only exact matches (i.e., matches with zero substitutions) were retained as positive RPB calls. As a qualitative validation step, the matching procedure was applied to the well-characterized B. breve strain UCC2003^[52], confirming that the pipeline correctly identified the expected RPB sequences in the upstream regions of this strain, consistent with previously reported transcriptional data.

A custom database of TFBs was constructed using all available sequences deposited in BacRegDB^[57], and then used as a reference in BLASTN analyses of the upstream region sequences. The average ESp value for each dinucleotide was collected from previous studies and public online databases^[58-61]. These values were used to calculate a continuous ESp score from 0 to 10 for each gene’s upstream region, integrating a parameter-priority-weighted score, a cluster-compressed redundancy score, and a Z-score-based distance-from-mean score. The distribution of continuous ESp scores was analyzed using an elbow-method approach to identify natural breakpoints separating distinct score ranges. The analysis was performed independently on the upstream regions of Escherichia coli (E. coli) and B. bifidum. As a qualitative validation step, the scoring was applied to a set of E. coli K12 upstream regions with known promoter activity, including constitutive promoter sequences from the iGEM catalog^[62]. The use of E. coli K12 enabled a qualitative evaluation of the pipeline by identifying ESp score ranges associated with strong, weak, or absent promoters, as defined and characterized in vitro. Similar ranges were identified in B. bifidum PRL2010, allowing the assignment of discrete values reflecting the putative propensity for double-strand unwinding: 0.99, 0.66, or 0.33 depending on the degree of this propensity, or 0 otherwise.

At each upstream region, a putative operon value (POV) was assigned as the sum of the presence of hypothetical RPBs (1/0), TFBs (0.5/0), and the discrete value of the ESp.

Phylogenetic analyses were performed using RAxML^[63] on the Roary output for all three genomic datasets. The Roary gene presence-absence matrix for each dataset was integrated with its corresponding core-genome phylogenetic tree to examine patterns of gain and loss of function (GoF/LoF). A Wagner parsimony model was applied to each Roary COG to evaluate GoF/LoF and the statistical significance of each event^[64]. Subsequently, the influence of the presence of GoF/LoF genes on genome distribution was also evaluated by Principal Coordinates Analysis (PCoA) based on Jaccard distances calculated from the presence matrix of the genes of interest.

B. bifidum PRL2010 was selected as the model strain for transcriptomic profiling due to its extensive genomic and functional characterization, including detailed annotation of the host-glycan degradation genes, and because it is the most documented strain belonging to this species in terms of genetics and interaction with the human gut microbiota^[4].

On this basis, transcriptomic data for B. bifidum PRL2010 grown on a panel of host-derived glycans representative of the major structural classes of HMO^[65] and of the intestinal mucosal environment^[66], were retrieved from a previously published study, including 2’-fucosyllactose (2FL), 3’-fucosyllactose (3FL), 3’-sialyllactose (3SL), 6’-sialyllactose (6SL), disialyllacto-N-tetraose (DSLNT), lactosamine, lacto-N-neotetraose (LNnT), lacto-N-tetraose (LNT), and mucin [Supplementary Table 3]. All glycans were used at a final concentration of 1% (wt/vol), as previously described^[2]. Furthermore, transcriptomic data using GlcNAc were newly generated in the present study.

B. bifidum PRL2010 was inoculated into de Man-Rogosa-Sharpe (MRS) (Scharlau Chemie, Barcelona, Spain) medium, where the carbohydrate source was represented by either 1% GlcNAc (wt/vol) (Carbosynth Ltd, UK) or glucose, as a control (Sigma-Aldrich), adjusting the culture to a final optical density at 600 nm (OD600) of 0.2. Cultures were incubated at 37 °C under anaerobic conditions (2.99% H₂, 17.01% CO₂, and 80% N) for 8 h. All experiments were performed in triplicate. Following incubation, cultures were centrifuged (7,000 rpm for 15 min) to separate the cell pellet from the supernatant. All aliquots were stored at -80 °C until RNA extraction.

Total RNA from each bacterial culture was isolated as previously described^[67]. Briefly, cell pellets were initially suspended in 1 mL of QIAZOL lysis reagent (Qiagen, United Kingdom) and transferred into tubes preloaded with 0.8 g of 106 μm glass beads (Sigma-Aldrich). Mechanical cell disruption was performed using a bead beater, with alternating cycles of 2 min of vigorous agitation followed by 2 min of rest on ice. After lysis, samples were centrifuged at 12,000 rpm for 15 min, and the aqueous phase, containing the RNA, was carefully recovered. RNA purification was then completed using the RNeasy Mini Kit (Qiagen, Germany) following the manufacturer’s guidelines.

Total RNA was isolated from cultures at the exponential growth phase (OD_600nm between 0.6 and 0.8). RNA quality and integrity were assessed using a TapeStation 2200 (Agilent Technologies, USA), while concentration was determined via a Picodrop microliter spectrophotometer (Picodrop, UK). Whole-transcriptome libraries were prepared using the Illumina Stranded Total RNA Prep, Ligation with Ribo-Zero Plus (Illumina, USA), following the manufacturer’s instructions, and a specific depletion step utilizing the Ribo-Zero Plus Microbiome probe pool to remove ribosomal RNA (rRNA). Fragmented and adapter-ligated cDNA libraries were sequenced on an Illumina platform.

Raw reads were quality-filtered and aligned to the B. bifidum PRL2010 reference genome using METAnnotatoX2 RNA-specific tool^[36]. Quantification of unique reads mapping to Open Reading Frames was performed using HTSeq^[68]. Differential gene expression analysis between GlcNAc-grown cells and glucose-grown cells was conducted with edgeR v4^[69].

B. bifidum PRL2010, B. bifidum LMG 11041, B. bifidum 184B, Bifidobacterium callitrichos DSM 23973, B. biavatii DSM 23969, and B. leontopitheci LMG 31471 were cultivated in de Man-Rogosa-Sharpe (MRS) medium supplemented with 0.05% L-cysteine hydrochloride in an anaerobic chamber (Concept 400, Ruskinn) at 37 °C. In this context, B. bifidum 184B was selected due to the higher ANI divergence from the type strain LMG 11041 and the reference strain PRL2010. After overnight growth, bacterial cultures were diluted, when necessary, to reach an initial OD600 of 0.2, washed in PBS, and resuspended in a previously formulated carbohydrate-free defined medium containing (per liter of distilled water) 4.0 g of sodium acetate; 1.0 g of tri‐ammonium citrate; 2.0 g of KH₂PO₄; 2.0 g of K₂HPO₄; 0.5 g of MgSO₄; 0.05 g of MnSO₄; 0.02 g of FeSO₄; 0.2 g of CaCl₂; 20 mg of adenine; 40 mg of xanthine; 0.4 g of cysteine; 0.3 g of aspartic acid; 0.3 g of glutamic acid; 0.2 g of each the following amino acids: alanine, arginine, glycine, histidine, isoleucine, leucine, lysine, methionine, phenylalanine, proline, serine, threonine, tryptophan, tyrosine and valine; 0.5 g of orotic acid; 0.5 mg of p‐aminobenzoic acid; 0.5 mg of folic acid, 2.0 mg of nicotinic acid; 2.0 mg of Ca‐pantothenate; 1.0 mg of biotin; 2.0 mg of pyridoxal; 2.0 mg of riboflavin; and 1.0 mg of vitamin B12 and 0.02 mg of uracil 0.02 mg. Subsequently, the medium was supplemented with either 1% (w/v) glucose (control condition), 1% GlcNAc, or 1% N-acetylgalactosamine (GalNAc). The medium was sterilized by filtration (0.22 μm). Cells were incubated anaerobically at 37°C for 6 h. All cultures were grown in triplicate.

Bacterial RNA was used to analyze the licC gene expression by quantitative real-time PCR (qPCR). Specifically, 500 ng of total RNA from each sample was reverse transcribed to cDNA using the iScript Select cDNA Synthesis Kit (Bio-Rad Laboratories) under the following thermal cycling conditions: 5 min at 25 °C, 30 min at 42 °C, and 5 min at 85 °C. At the end of the process, 25 ng/µL cDNA aliquots were amplified in a total reaction volume of 10 µL using the qPCRBIO SYBR Green Mix Lo-Rox (PCR Biosystems) and the forward and reverse primers (5 pmol each). The primers used for licC amplification were: licC_group2_forward 5’- ACTGCTTCTGCAAGTTCGG -3’ and licC_group2_reverse 5’- GGATGAGCATGATCGGGTT -3’, licC_LMG 31471_forward 5’- GCTCTGGTCGCCTTCTTCA -3’ and lic_LMG 31471_reverse 5’- GACGACGATGGCGGTAATC -3’, licC_50T_forward 5’- GGCATCTTCAACCAGACCT -3’ and licC_50T_reverse 5’- CTTGATGCAGGCGGTGTAGA -3’.

Expression levels were normalized relative to housekeeping genes as previously described in different studies^[70,71]. Real-time PCR was performed using the CFX96 system (Bio-Rad, CA, USA). Fluorescence was monitored at the end of each extension step. Melting curve analysis was performed at the end of each amplification cycle. Data analysis was performed using the relative standard curve method^[72]. The expression values were calculated as the ratio between the abundance of the target mRNA under each test condition and its abundance in the corresponding glucose-grown control.

To comprehensively characterize the B. bifidum taxon, an extensive collection of genomic sequences was curated from public repositories, complemented by newly DNA-sequenced genomes of isolated B. bifidum strains, and by the assembly of new MAGs. After genome quality control and dereplication using CheckM2 and drep^[38,39], the final B. bifidum dataset encompassed 1,351 unique high-quality sequences, including 156 genomes from cultured isolates and 1,195 MAGs [Supplementary Table 1]. The inclusion of MAGs enabled exhaustive screening of the species’ genetic diversity, resulting in a closed pangenome [Figure 1A], as previously accomplished for B. adolescentis and B. longum^[33,34].

In contrast, analysis of the datasets generated in this project revealed that the other HBS pangenomes remain open [Supplementary Figure 1]. Reconstructing the HBS pangenomes enabled a comparative assessment of their genomic features relative to those of B. bifidum, providing insights into their genetic diversity. In the comparison among the HBS and PBS [Supplementary Table 2], B. bifidum had the smallest genome size and the fewest coding DNA sequences, followed by B. adolescentis and B. pseudocatenulatum. Overall, the genomic features of B. bifidum show mean values closer to those of HBS than PBS, except for GC content, which remains above 60% as in other PBS.

ANI analysis revealed a mean of 98.91 ± 0.16% [Figure 1B], confirming remarkable genetic homogeneity and the absence of subspecies taxonomic divisions, in contrast to other HBS such as B. catenulatum and B. longum [Supplementary Figure 2]. However, we identified a specific subset of 22 B. bifidum genomes, 19 of which were from rural populations in Malawi and Mozambique, showing ANI values below 98.5%. Notably, phylogenomic analysis did not support the presence of a distinct lineage composed of these genomes. In the ANI-based phylogenomic tree, the 22 genomes occupy isolated positions rather than forming a distinct branch [Supplementary Figure 3]. This topology suggests that their divergence does not define a subpopulation shaped by the same selective pressures but rather reflects variation in accessory gene content within this geographical region. This characteristic is consistent with the higher biodiversity observed in non-Westernized settings, where more heterogeneous dietary and environmental exposures may allow a broader range of accessory gene content to persist than is usually observed in genomes from industrialized populations^[73-75]. Similar patterns, in which non-Westernized or rural cohorts exhibit greater strain-level and accessory genome diversity than Westernized populations, have been reported within the genus Bifidobacterium^[34,76].

This localized ANI divergence may be due to 46 nearly exclusive clusters of orthologous genes (COGs), 20 of which are completely absent from the rest of the dataset [Supplementary Table 4]. These genes were organized into seven putative genomic islands or functional units, suggesting a specialized gene toolkit for the rural gut environment. Functional annotation revealed that several of these genes encode glycosyltransferases (GT) from the GT1 and GT2 families and components of the Wzy-dependent pathway, which are likely involved in specific cell wall and exopolysaccharide modification. Furthermore, two genomic islands harbored functional domains involved in the degradation and import of fucosylated glycans and complex hexosamines. The limited presence of these functions in this subset of genomes may reflect niche-specific adaptation in non-Westernized populations, despite the overall genomic stability of B. bifidum.

The pronounced genetic homogeneity of B. bifidum, as indicated by ANI values, is further supported by the lack of significant variation in metabolic gene content across the pangenome of this taxon. Specifically, analysis of the glycobiome provided no evidence for subclusters with distinct carbohydrate-degrading capacities, suggesting a highly conserved metabolic core.

By evaluating the GoF/LoF mechanisms within the dataset, we identified 208 COGs with significant GoF and 245 COGs with significant LoF [Supplementary Table 5], which effectively separated the dataset into four distinct groups [Figure 2A and Supplementary Figure 4].

This variation is largely driven by the mobilome (25.4% of COGs), particularly prophage-related elements in COGs group 1, and restriction-modification systems in COGs group 2 [Figure 2B]. In the absence of characteristic plasmids in B. bifidum, these clusters likely represent episome-derived elements integrated into the chromosome^[78-81].

To assess the distribution of CRISPR-Cas systems across the species, we screened the 1,351 B. bifidum genomes using CRISPR-CasFinder. CRISPR loci were identified in 16.95 % of genomes harboring a high-confidence Cas gene cluster, i.e., 84 type I-C, one type class I-E, 139 type class II-A, and five type II-C systems. Type I-C clusters consistently encoded the cas3, cas5, cas8c, cas7, cas4, cas1 and cas2 genes, while class II-A clusters encoded cas9, cas1 and csn2. BLASTN comparison of the predicted spacer repertoire against the NCBI phage genome database detected prophage-derived spacer matches in 195 genomes, 186 of which specifically targeted Bifidobacterium phage Bbif-1 (GQ141189), identifying it as the predominant prophage target of CRISPR-Cas-mediated immunity in this species [Supplementary Table 6].

Given the impact of the mobilome on species biodiversity, the propensity of B. bifidum for horizontal gene transfer (HGT) was evaluated. Evaluation of the CUB and the GC content of each gene suggested that B. bifidum exhibits a different HGT rate than HBS [Figure 3A], potentially reflecting a mechanism conserved across PBS.

On average, we identified 5 ± 4 putatively acquired genes in B. bifidum, compared to 7 ± 4 in other PBS [Figure 3B and Supplementary Table 7]. Statistical analysis of the hcHTG variance across genomes in the three groups revealed that the internal variability of the B. bifidum dataset was more similar to that observed in the PBS than to that in the HBS [Supplementary Figure 5]. Based on amino acid sequence similarities, hypothetical donors include HBS (B. pseudocatenulatum, B. longum, B. adolescentis), Bifidobacterium callitrichos, a species associated with the gut microbiota of the non-human primate Callithrix jacchus, followed by Blautia spp., and Collinsella spp. [Figure 3C and Supplementary Table 8].

Results of the intraspecific comparison suggested that B. bifidum has reached broad stability in its core genome while retaining sufficient plasticity to acquire accessory genes or variants, particularly in those genomes associated with non-Westernized rural populations.

Given the highest HGT rate observed in B. bifidum compared to HBS, we performed an interspecific comparison with HBS and PBS to identify unique traits acquired during adaptation to the human gut. A comparative analysis of metabolic networks showed that B. bifidum was significantly enriched in genetic modules associated with carbohydrate uptake and consumption, energy production, and vitamin metabolism [Supplementary Figure 6].

Large-scale functional screening [Figure 4] identified 17 KEGG functions exclusively present in B. bifidum relative to all other HBS.

A significant portion of these unique traits was dedicated to cell wall biosynthesis and environmental interaction, including enzymes for teichoic acid synthesis (K18704, K21285), exopolysaccharide transport (K09690), and cell wall modification or anchoring (K03591, K13733, K03749). The analysis also highlighted species-specific metabolic components, including subunits of two different Phosphotransferase Systems (PTS) (K02759, K02798), immunomodulatory proteins (K08652), and transcriptional regulators associated with glycosyl hydrolases (K03489). Furthermore, B. bifidum possesses a unique Acetyl-CoA regeneration system encoded by KEGG orthologs K00161, K00162, K21416, and K21417^[82].

Consistent with this metabolic profile, B. bifidum harbored a highly specialized glycobiome, with a substantial fraction of predicted enzymes to target host glycans. B. bifidum dedicates 43% of its 48 highly conserved glycosyl hydrolases (GH) to host-derived glycans, such as HMOs and mucin [Figure 5A and Supplementary Table 9].

This accounts for 71% of the total host-glycan-targeting GH families identified across the HBS pangenome, significantly exceeding the group average (32 ± 6%). Comparative analysis identified eight GH families absent or rare in other HBS, namely, GH43 subfamily 24, GH51 subfamily 9, GH84, GH89, GH110, GH123, GH136, and GH177, and six unique COGs combining a GH domain with galactose high-affinity carbohydrate-binding module 32 (CBM32)^[83]. While all these glycobiome features were found with variable degrees in PBS [Figure 5B], B. bifidum displays the highest complexity and richness in enzymes dedicated to host-derived substrates. These results suggest that the evolutionary success of B. bifidum in colonizing the infant gut relied on a coordinated expansion of its degradative and transport machineries, acting as coupled systems for the simultaneous breakdown and uptake of host-derived glycans.

To delineate the unique genetic blueprint of B. bifidum, we compared its pangenome with that of the other six HBS, identifying 667 COGs unique to B. bifidum soft-core [Supplementary Table 10]. This extensive repertoire of B. bifidum-specific COGs is primarily dedicated to cellular metabolism (34.8%), membrane transport (12.8%), and transcriptional regulation (9.7%). Specifically, 25 unique COGs are implicated in the import and processing of human milk sugars^[84], while others are associated with host-immune interactions^[85] and osmotic and pH regulation^[86], i.e., functions that support survival in the infant gut^[23,87,88]. In addition, 37 COGs harbored unique functional domain annotations within the HBS pangenome. Among these specific protein signatures, in addition to amino acid and cofactor metabolism, redox and transcriptional regulation, there were COGs associated with two PTS systems (K02759, K02798) and teichoic acid biosynthesis (K18704, K21285), previously highlighted in the metabolic comparative analysis.

Genomic organization of the B. bifidum reference strain PRL2010^[89] using a Putative Operon Value (POV), predicted 40 operons comprising three or more unique COGs [Supplementary Table 11]. Fundamental processes, such as nucleotide and cofactor metabolism, were constitutively expressed under strong promoters (POV from 2.16 to a maximum of 2.49), whereas operons linked to niche adaptation, such as those for cellular response mechanisms to stress, exhibited POV values between 1.83 and 2.16 [Figure 6A].

Of the 37 COGs with unique functional annotations, only those with PTS systems and teichoic acid biosynthesis were arranged in three putative genetic operons [Figure 6B]. Unlike the other putative genetic operons, these exhibit a POV lower than 1, suggesting that their activation responds to specific conditions that allow the transcription of these functional islands.

The distinct evolutionary trajectory of B. bifidum is further underscored by the identification of 76 COGs exhibiting significant GoF events relative to both HBS and PBS [Supplementary Table 12]. Notably, Operons 36 and 40, reported in Figure 6B, represent substantial evolutionary acquisitions, encoding specialized PTS systems absent in other HBS. More specifically, operon 40 encodes a PTS system whose three subunits exhibit high amino acid identity (up to 96.7%) with the closest PBS, B. jacchi, and B. leontopitheci, as well as high identity to Bombiscardovia coagulans (77.5%) and Gardnerella spp. (83%-85%). This high level of conservation across genera suggests gene acquisition from primate-associated bifidobacteria and related Bifidobacteriaceae. Likewise, operon 36 encodes a 4.A.3 family PTS^[90], a GH1 6-phospho-beta-glucosidase (bglA), and a gntR/yydK-like regulator [Figure 6B]. This gene cluster shares 80.4% sequence identity with B. callitrichos and exhibits broader homology with more distantly related anaerobic genera, such as Clostridium (57.6%), Eubacterium (57.3%), and Carnobacterium (59.3%). By comparing the amino acid sequences encoded by the genes of the PTS systems with those of the PBS, we obtained a division of the species cluster. Specifically, i.e., B. leontopitheci, B. amazonense, B. biavatii, B. miconis, B. ramosum, and B. aerophilum, showed significant matches with an average identity of 53.8 ± 22.2, whereas no significant matches were identified for the remaining species. These patterns suggest a more complex or ancestral acquisition event. Overall, these results indicate that B. bifidum has enhanced its metabolic repertoire by acquiring high-affinity transport systems from a wide phylogenetic spectrum, thereby gaining a significant fitness advantage in colonizing the human intestinal environment.

To validate the functional relevance of the identified evolutionary gains, we analyzed the transcriptional profile of B. bifidum PRL2010 grown on different host-glycan substrates, such as^[2] [Supplementary Table 3]. In addition, we conducted targeted follow-up transcriptomic assays to support and extend earlier observations. Our analysis revealed a coordinated activation of species-specific genes [Figure 7], including unique PTS systems and six GHs carrying an associated CBM32 module, which were absent in the other HBS [Figure 5A].

Specifically, the PTS system encoded by operon 36 and several CBM32-containing GHs showed a significant transcriptional response to mucin compared with glucose. Two genes encoding subunits of the latter PTS exhibited the strongest induction (log₂FC > 7), which correlated with significant up-regulation of mucin-targeting enzymes, such as CBM32-containing GH29 and GH20 [Figure 7]. This coordinated activation was also observed when B. bifidum PRL2010 was grown on GlcNAc as the sole carbon source. In this condition, only 20 genes of B. bifidum PRL2010 were significantly up-regulated (log₂FC > 1), including the two PTS subunits of operon 36. Notably, the core catabolic enzymes nagA and nagB remained constitutively active in both glucose and GlcNAc, confirming that the pathway is functionally operative for direct metabolism rather than solely for cell wall structural biosynthesis.

A similar trend of coordinated response was observed for the second species-specific PTS system encoded by operon 40 and the broader CBM32-containing GH repertoire in response to HMOs. The PTS system encoded by operon 40 behaved differently from the mucin-specific PTS of operon 36, showing a broader and more consistent expression pattern across the different host-glycans tested. This effect was most evident with fucosylated and sialylated HMOs, while no significant overexpression was observed on GlcNAc. We observed a similar behavior in several CBM32-containing GHs [Figure 7A], which were strongly induced by 2FL, 3FL, and LNT, suggesting a key role in the utilization of human milk glycans.

To validate transcriptional induction of the species-specific PTS machinery, we performed RT-qPCR analysis of the licC homolog across three B. bifidum strains (PRL2010, 184B, and LMG 11041) and three PBS (B. callitrichos DSM 23973, B. biavatii DSM 23969, B. leontopitheci LMG 31471) grown on GlcNAc or GalNAc. Each condition, including the glucose reference, was assayed in triplicate. All three B. bifidum strains showed consistent upregulation of the licC homolog on GlcNAc when normalized on the glucose condition (FCs from 1.6 to 2.9), whereas induction was weak or absent in the PBS taxa tested (FCs from 0.6 to 1.4), supporting a B. bifidum-specific transcriptional response to this substrate. In contrast, induction by GalNAc was modest and inconsistent across all strains and species tested (FC ranging from 0.8 to 1.9), indicating that this substrate is not a strong inducer of the licC homolog.

Taken together, these findings demonstrate that B. bifidum possesses a specialized, coordinated genetic machinery and unique PTS transporters that enable efficient exploitation of the diverse glycan landscape of the neonatal gut.

The extensive addition of MAGs enabled the pangenome to approach closure and allowed us to represent intraspecific variability associated with specific geographic areas. Nevertheless, specific regions, such as South America, remain underrepresented due to the limited availability of gut-derived genomes from unexplored parts of the globe. In addition, because host metadata, such as age and sex, were not publicly available for most of the MAGs, we couldn’t identify a clear, strong correlation between genomic features and host characteristics. Genome-wide transcriptomic profiling was performed on a single strain of B. bifidum PRL2010 under standard laboratory conditions. Given the intraspecific divergence identified in this study, this profiling may not fully capture the transcriptional landscape of the whole species. RT-qPCR validation of the licC homolog partially mitigated this limitation by extending the analysis to additional B. bifidum strains and PBS taxa, but a genome-wide transcriptomic characterization of a broader strain panel, particularly of the most divergent genomic subset identified, would be needed to fully resolve the extent of intraspecific transcriptional variability. Finally, in the context of a large-scale MAG-based approach, residual assembly artifacts cannot be entirely excluded despite the application of stringent quality-filtering criteria.

In conclusion, this study provides the most comprehensive genomic-scale view of B. bifidum to date, revealing limited intraspecific genomic diversity alongside a marked functional specialization toward the human gut environment. Comparative genomic and transcriptomic analyses uncovered a pronounced specialization toward host-derived glycan utilization, including an expanded set of glycoside hydrolases and unique phosphotransferase systems that likely contribute to the ecological success of this species in early life. Despite its overall genomic stability, the identification of localized divergence in strains from non-Westernized populations highlights the persistence of niche-specific evolutionary adaptations. Collectively, these findings define B. bifidum as a metabolically specialized cornerstone of the neonatal gut microbiota and provide a robust foundation for future studies on its evolutionary dynamics and host-associated functions.