A Hadza-enriched Prevotella/Segatella xyloglucanase shows sequence conservation and functional specialization Authors: Marie Louise Jespersen, Kamilla Kjærgaard Munk, Sune Fjermedal, Bo Pilgaard, Anne S Meyer, Frank M Aarestrup, Saria Otani Published: 2026 ABSTRACT ABSTRACTBacteria can adapt to their environment through changes in their genetic material. A large proportion of gut bacteria are shaped by host-specific diet, including complex carbohydrates. The bacterial abundance, genetic content within the same bacterial species, and sequence-level variation in genes encoding similar carbohydrate-processing enzymes may therefore vary across hosts with different diets. We previously found that the abundance of diet-degrading genes varies between hominid host populations from Tanzania. We therefore hypothesized that, in addition to these abundance differences, selective pressure could act on individual gene sequences. Here, we investigated Tanzanian hominid gut microbiome differences at the taxonomic, genetic, structural, and functional levels. We analyzed 15,146 metagenome-assembled genomes (MAGs) spanning 1563 species and identified one species with striking host-associated separation. In particular, sequence variation in a xyloglucanase-encoding gene correlated strongly with the host population. This gene was highly conserved in the Hadza population, suggesting a role in the processing of diet-associated polysaccharides. Sequence differences and structural modeling revealed amino acid substitutions near the catalytic site, and biochemical assays using xyloglucan showed that representative variants differed in activity under identical assay conditions. Collectively, our findings suggest that host lifestyle and diet contribute to population-associated sequence variation in genes encoding enzymes involved in degrading polysaccharides. Introduction High gut microbiome diversity is increasingly recognized as a hallmark of human health. One example of this diversity is the gut microbial communities of the Hadza, an indigenous hunter-gatherer population in Tanzania.1–3 Compared to individuals living in more urban societies, the Hadza exhibit significantly greater taxonomic richness and diversity in their gut microbiome.1–3 This disparity has been attributed to the gradual loss of co-diversified gut bacterial lineages in more industrialized populations, likely driven by lifestyle shifts associated with urbanization, including changes in diet to be low in complex carbohydrates, increased use of medications such as antibiotics, and increased sanitation.2,4,5 Especially the Western diet, i.e. the North American and European diet characterized by high levels of processed foods, refined sugars, and fats, combined with low dietary fiber content, has also been implicated in the decline of gut microbiome diversity.6,7 In addition to their high taxonomic diversity, Hadza gut microbiomes are dominated by other bacterial species than those commonly found in individuals with more urban lifestyle.1,2 For example, members of the genus Prevotella are more frequently detected in the Hadza gut microbiome.1,8 Certain Prevotella species exhibit notable genomic plasticity, with substantial strain-level variation observed within a single species.9 Such intra-species diversity has been associated with host lifestyle factors, including diet and environment.8,10 In particular, the host diet has been shown to influence variation within Segatella copri (formerly Prevotella copri), which is associated with metabolic functions such as complex carbohydrate digestion and drug metabolism.11 The gut microbiome provides a complementary set of genomes encoding enzymes capable of breaking down otherwise indigestible substrates of the host diet, such as dietary fibers. However, in the absence of this functional requirement, for example, owing to a low-fiber diet, it may be lost through functional adaptations or the complete loss of specific bacterial species. Recent studies have identified such differences in microbiome function between various hosts, e.g., humans and other hominids, as well as between African and European human populations.7,10 The Hadza diet is fiber-rich, consisting mainly of seasonal fruit, baobab, tubers, and hunted wildlife.12 Correspondingly, the Hadza gut microbiome exhibits greater dietary gene diversity compared to that of individuals from Western populations.1,2,13,14 This elevated functional potential likely reflects microbial adaptation to the complex and fiber-rich diet of the Hadza population. Beyond genetic diversity, the structure of the Hadza microbiome is also shaped by social and environmental factors. Recent findings suggest that bacterial strains may be shared horizontally among individuals within the same Hadza community, likely facilitated by close social interactions and shared living conditions.2 These lifestyle and dietary variations between indigenous Hadza and non-indigenous human populations represent different ecological niches within the gut, which could be reflected in distinct adaptations and evolution of the gut bacteria between these host populations. The lifestyle differences among hosts may drive differences not only in bacterial taxonomic composition between host populations but also in genomic variation within shared species. Such intra-species host adaptations could be found as species specialization, reflected in differential functional capacity within the same bacterial species, introduced by the loss or uptake of new genes or the adaptation of existing ones. In fact, it has previously been found that Horizontal Gene Transfer (HGT) was increased in the gut microbiomes of humans living in urban areas,15 potentially as a result of the rapid lifestyle changes (on an evolutionary scale) that have necessitated the adoption of new functions in our gut microbiome at a faster scale than mutational evolution could support. Additionally, we previously found that the gene abundance of various dietary genes varies between different host populations and that the sequence variation of such genes in some cases is correlated with the host population.10 However, the extent to which individual microbial genes are structurally and functionally adapted to the host diet remains poorly understood. In this study, we aimed to investigate differences that arise from mutational evolution of the human gut microbiome by examining variation within species and within orthologous genes shared between bacterial strains from host populations with different lifestyles. Using high-resolution comparative metagenomics on a set of metagenomic assembled genomes (MAGs), we identified a xyloglucanase-encoding gene (gene 380) that was highly conserved among individuals in the Hadza population, suggesting a specialized and potentially essential function of this enzyme. Functional characterization of gene 380 variants revealed that the Hadza-associated variant (380 H) degrades tamarind xyloglucan more slowly than the variant (380 A) that is associated with the non-indigenous adult microbiomes, thus indicating a difference in the catalytic rates of the enzyme variants. Collectively, our results suggest the selective adaptation of bacterial gut species to host lifestyle and diet. Results Xyloglucanase in Prevotella shows Hadza-enriched prevalence and reduced allelic diversity To identify species with important functions in the Hadza gut microbiome, we used 15,146 NC (Near Complete) MAGs (Metagenome-Assembled Genomes) created previously from 546 samples.10 We screened these MAG species present in both indigenous and non-indigenous populations, and with a high correlation between species phylogeny and the host population (Supplementary Results and Discussion). Amongst these, Prevotella sp002251295 was particularly distinct and hence interesting. Within this species, most Hadza-derived MAGs clustered into two distinct groups with high intra-cluster genomic similarity (Figure 1a), suggesting the presence of two conserved strains of this species within the Hadza gut microbiome. Prevotella species have previously been found in higher abundances in non-western gut microbiomes,1,2 and this was also the case for the Prevotella sp002251295, which was both more abundant and prevalent in the Hadza population when compared to the other populations in our cohort (Figure 1b, c). This enrichment may reflect dietary or environmental factors unique to the Hadza lifestyle, such as a high-fiber, plant-rich diet, which has been associated with increased Prevotella abundance.1,2 The consistent presence (77%) of this species across Hadza individuals suggests that it plays a stable and potentially beneficial role in the Hadza gut ecosystem. This phenomenon is correlated with the complex carbohydrate utilization and, hence, the associated carbohydrate-active enzymes encoded by the members of the Prevotella genus (Segatella). Prevalence, abundance, and phylogenetic distribution of Prevotella sp002251295 and gene 380 across host populations. (a) Phylogenetic tree of Prevotella sp002251295 (C26) MAGs, with tips colored by host population: Indigenous Hadza (orange), non-Indigenous Adults (blue), and non-Indigenous Children (red). The tree reveals two distinct Hadza-associated clades. (b) Abundance of Prevotella sp002251295 across host populations, measured as TPM (Transcripts Per Million). (c) Prevalence of Prevotella sp002251295 across the same populations, shown as the percentage of samples in which the species was detected. (d) Phylogenetic tree of gene 380, a xyloglucanase-encoding gene, with tips colored by host population as shown in panel (a). The tree shows strong conservation of the gene among the Hadza samples. (e) Abundance (top) and prevalence (bottom) of two gene 380 variants (Hadza and non-Indigenous adult) across six groups: Almeida, non-Indigenous adults, non-Indigenous children, Tanzanian, Indigenous Hadza, and Ultra-deep Hadza. Gene 380 is more abundant and prevalent in Hadza samples compared to other populations. To explore potential drivers of the high strain conservation observed in selected species, we examined gene-level variation correlated with host population structure (Supplementary Results and Discussion). In Prevotella sp002251295, we identified one gene (gene 380) that exhibited remarkably high sequence similarity across nearly all Hadza-associated MAGs (Figure 1d), indicating strong conservation of this gene within the Hadza population. Even though the Prevotella sp002251295 species tree revealed at least two distinct and conserved strains within the Hadza population (Figure 1a), gene 380, stood out as even more conserved, showing minimal sequence variation across nearly all Hadza-associated MAGs (Figure 1d and Supplementary Figure 7a). This level of conservation is particularly noteworthy because gene 380 exhibited the greatest overall variation (8.9% difference) among all gene variants in our broader analysis (Supplementary Figure 7a). Gene 380 was predicted to encode xyloglucanase, an enzyme involved in the degradation of xyloglucans, which are complex polysaccharides commonly found in the primary plant cell walls of dicots.16 Its conservation may reflect adaptation to Hadza’s fiber-rich diet, suggesting a functional role in the microbial processing of plant-derived carbohydrates and highlighting the potential importance of this gene in host nutrition. The contrast between species-level diversity and gene-level conservation could reflect a strong selective pressure to maintain enzyme function in response to dietary substrate characteristics of the Hadza’s diet. To ensure that our observations were not limited by MAG recovery, we investigated the presence of the xyloglucanase gene across all contigs, not only the ones included in the Prevotella MAGs. Using this approach, we were able to identify the gene in a few more individuals, with the phylogenetic tree of the gene variations being similar to the previous ones (Supplementary Figure 7b). We also investigated the presence of this xyloglucanase gene across all MAGs in our collection, not only Prevotella sp002251295, but we did not find the gene in any other species (>90% similarity). To increase sensitivity, we mapped metagenomic reads from our samples (n = 546) to two representative variants of gene 380, which were selected to reflect the predominant sequences observed in the Hadza (380 H) and non-Indigenous adult (380 A) populations, respectively. This analysis included additional metagenomes from a larger Hadza cohort (n = 469),2 other Tanzanian individuals (n = 318),17 and populations from other regions (n = 179).18 Both gene variants were more prevalent and abundant in Hadza samples compared to other Tanzanian or global samples (Figure 1e), consistent with the higher abundance and prevalence of Prevotella species in the Hadza gut microbiome.1,2 Outside the Hadza population, both gene variants were present at very low abundances, although 380A remained detectable in 38.6% of non-indigenous adults. Together, these findings reinforce the specificity and potential functional importance of the xyloglucanase gene in the Hadza microbiome. Its widespread presence and conservation within this population, contrasted with its limited distribution elsewhere, suggest a strong link between microbial gene content and host lifestyle, highlighting how dietary pressures may shape microbial community composition by selecting for functionally important genes. While our analyses so far have focused on genetic variation, they point to a potentially essential role for this gene in the Hadza gut ecosystem. To further investigate its functional relevance, we next examined the structural variation of the encoded xyloglucanases, aiming to understand how sequence conservation may translate into conserved enzymatic function. Structural differences suggesting specialized xyloglucan degradation in Hadza To assess if the genetic variation had any structural impact with the potential to result in functional differences in the xyloglucanase enzyme, we used AlphaFold219 to predict the protein structure of 380H and 380A (Figure 2a, b). Even though the genetic variance for gene 380 resulted in 45 amino acid substitutions between the protein sequences of 380H and 380A, the predicted protein structures were overall similar in topology, although the domain orientations differed (Supplementary Figure 6a). Both proteins contain two domains: a Glycoside Hydrolase family 5 (GH5) domain and a PFAM 13004 domain (Figure 3a and b). The PF13004 domain has been proposed to function by distancing the catalytic domains from the cell surface, thereby enhancing the protein’s mobility.16,20 Structural differences between Hadza and non-Indigenous adult variants of the xyloglucanase encoded by gene 380. (a) Predicted structure of the xyloglucanase protein from a non-Indigenous adult sample, generated using AlphaFold2. (b) Predicted structure of the xyloglucanase protein from a Hadza sample. Both AlphaFold2 models were aligned to the xyloglucanase structure BoGH5A (PDB id: 3ZMR), and the polysaccharide substrate from this reference structure is shown in orange. (c) Comparison of the binding site architecture between BoGH5A and AlphaFold2 models. The BoGH5A structure is shown in grey, with its substrate in orange; side chains from the non-Indigenous adult and Hadza variants are highlighted in red and blue, respectively. (d) Distribution of amino acid motifs at positions 145–149 across non-Indigenous adult and indigenous Hadza samples, including 98 samples from Carter et al. (2023)2. The motif variation suggests potential biochemical differences between population-associated variants. Genomic context of gene 380. (a) Polysaccharide utilization locus (PUL) architecture surrounding gene 380, annotated as GH5_4B. The schematic shows adjacent genes with functional annotations and color-coded identities, highlighting the genomic arrangement around gene 380. (b) Phylogenetic tree of flanking sequences (3600 bp upstream and 5000 bp downstream of gene 380), showing relationships among samples. The left coloured strip indicates sequence identity to the non-Indigenous adult reference variant of gene 380, and the right strip denotes the host population. The alignment of flanking regions reveals conserved genomic organization across samples. (c) Enzymatic activity time course of gene 380 variants. The line plot shows the release of glucose equivalents (nmol) over time (minutes) during substrate hydrolysis. Each curve represents the mean activity of three repeated experiments for the respective gene variant (Hadza in blue, Adult in red, negative control in green). Error bars indicate the standard deviation across replicates. The data illustrate differences in hydrolytic performance between populations, with the Adult variant exhibiting higher activity throughout the time course. This fits well with our finding of a flexible segment between the two domains, suggesting that the domains can move relative to each other, as was also reported in the original characterization of this protein. This flexibility was also evident in the BACON/PF13004 domain pointing in opposite directions in the top-scoring prediction of the Hadza or non-Indigenous adult protein structure (Figure 2a and b), likely reflecting two known structural orientations of this domain. To identify the catalytic site of the xyloglucanase encoded by gene 380, we aligned the AlphaFold2 structural models with a crystal structure from the Protein Data Bank (PDB id: 3ZMR) as a reference (Figure 2a–c). The reference structure, named BoGH5A,16 is a GH5 protein from Bacteroides ovatus. We selected this structure as a reference based on its high sequence similarity to the two AlphaFold2 models and because it had a bound substrate. The BoGH5A structure was originally described by Larsbrink et al.,16 and these authors found that the protein was anchored to the cell surface through a lipidation site at Cys 1, facilitating its localization. This residue was also found in our sequences, along with a signal peptide (Supplementary Figure 7 d), supporting that this protein is localized at the cell surface. The function of the GH5 encoded xyloglucanase is the initial step in xyloglucan utilization, specifically the degradation of xyloglucans by cleaving them into shorter oligosaccharides.16 When looking into the catalytic domain and binding site of this enzyme, we found that most residues are conserved. However, one motif in the binding site differed between the PDB structure, the 380A structure, and the 380H predicted structure; at positions 145–148, different residues were found between all three sequences (Figure 2c). These residue differences may give rise to functional differences between these proteins. When investigating the presence and variation of the entire gene in a large set of publicly available Hadza MAGs,2 we found the gene to be highly prevalent in MAGs assigned to Prevotella sp002251295 (found in 185/215 MAGs). Even though the genetic variance was not as conserved among the additional Hadza samples (Supplementary Figure 7c) as in our own sample collection, we found a significant difference in binding site motifs (X2 = 6.14, p = 0.01) between the non-Indigenous adult and the indigenous Hadza populations (Figure 2d). In the majority of the non-Indigenous adult samples (78%), we found the residues Gly-Asp-Asn-Lys (GDNK) at positions 145–148, whereas in the majority of the indigenous Hadza (69%), these positions contained the residues Gly-X-X-Thr (with either Asn-Asn, Ser-Asn, or Ser-Ser being in the middle two positions). This finding suggests that even though genetic variance is found throughout the Hadza population, these residues in the binding site of the GH5-protein have been conserved throughout many Hadza variants of gene 380, suggesting that these residues possess functional properties that are selectively advantageous. It has been suggested that the GH5-containing PULs (polysaccharide utilization locus) are specialized for the degradation of xyloglucans derived from various plant sources, potentially reflecting the dietary habits of the host.16 Xyloglucans are found in various fruits and vegetables of the human diet. It is a polysaccharide, consisting of a β(1→4)-glucan main chain and branches of xylosyl units, potentially linking additional monosaccharides. The branching can vary extensively, depending on the origin of the xyloglucan.16 A considerable part of the Hadza diet consists of various seasonal fruits that contain high proportions of xyloglucans,21 and thus, one explanation for the conserved residues could thus be a specialization of the enzyme towards degrading xyloglucans from such fruits that are found in large amounts in the Hadza diet. Gene 380 is located in a polysaccharide utilization locus associated with dietary fiber utilization To further investigate the function of gene 380 and surrounding genes, we selected a representative Hadza sequence for higher resolution examination, utilizing Cazyme annotation from the CUPP database22 (Figure 3a). Based on this, we were able to refine the annotation of the gene to the GH5_4 sub-family and found that it was located in a highly conserved PUL. This PUL has previously been identified in a Segatella copri (previously Prevotella copri) strain,23 which aligns with our discovery of this PUL and gene in a Prevotella species. Recent transcriptional work assessing the PULS response in Segatella copri DSM18205 when grown on different types of carbohydrates has further supported the role of distinct PULs in the utilization of dietary plant glycans, which is consistent with the functional assignment of the conserved locus identified here.24 In the previous study, Hibberd et al. found that certain strains of P. copri contained this PUL and that the prevalence of these strains was associated with weight gain in malnourished children in Bangladesh.23 The findings align with observations reported by Hibberd et al., which suggest that this PUL plays a significant role in microbiome-mediated energy extraction from the diet. Another study has also found the GH5-encoding gene is located in a conserved PUL containing SusD- and SusC-like proteins, which were suggested to be involved in the binding and transport of xyloglucan, respectively.16 The data thus align with our hypothesis that gene 380 contributes to dietary plant fiber breakdown in the Hadza gut microbiome. To further explore the genomic context of gene 380, we examined gene synteny across samples. The surrounding gene content was largely conserved (Mantel statistic = 0.716, q-value = 0.0002), with the only notable variation being the presence of either a SusC-like gene or a T2KPJ3 gene upstream of gene 380 (Figure 3b). Both genes have been implicated in extracellular polysaccharide degradation,25,26 suggesting that they may serve similar roles. Interestingly, this upstream variation led to a separation in the phylogenetic tree of the flanking sequences into two distinct groups (Figure 3b), resembling the species-level phylogeny (Figure 1a) more closely than the gene-level tree (Figure 1d). This pattern highlights the possibility that selective pressure acts specifically on gene 380, rather than on the surrounding genomic regions, and underscores its potential functional importance in the Hadza gut microbiome. Gene synteny analysis also revealed that gene 380 occurs as two adjacent copies in all investigated samples, both in the Hadza and adult populations (Figure 3b). This arrangement likely reflects a duplication event during PUL evolution, suggesting an important role for the xyloglucanase. Duplication could provide functional redundancy, ensuring that at least one active copy is present if the other accumulates deleterious mutations. Alternatively, the two copies may have diverged to perform complementary roles, each specialized for hydrolyzing differently branched xyloglucans. Hadza and adult xyloglucanase variants show distinct time-dependent cleavage of xyloglucan To determine whether the observed sequence and structural differences between 380H and 380A translated into measurable enzymatic divergence, both gene variants were synthesized and expressed in E. coli BL21 DE3 using the pET-28a(+) vector (see Methods). Purified recombinant proteins were assayed for xyloglucanase activity using a commercially available xyloglucan (from tamarind seeds) as a substrate. Both variants exhibited time-dependent cleavage of xyloglucan, while the negative control showed no detectable activity (Figure 3c). The adult enzyme consistently produced higher levels of the product, releasing approximately 25 nmol glucose equivalents after 30 min, compared with 15 nmol for the Hadza enzyme. Product accumulation was linear during the first 10 min and reached saturation after 25 min, suggesting substrate depletion or product inhibition. These results confirm that both alleles encode functional xyloglucanases but differ quantitatively in activity on tamarind xyloglucan, with 380A showing approximately 1.6-fold greater activity under identical assay conditions. The lower activity of the 380H enzyme on tamarind xyloglucan could indicate specialization toward structurally different xyloglucans, such as those found in baobab fruit, wild tubers, or other traditional plant foods consumed by the Hadza. These sources can contain more substituted xyloglucans, requiring modified binding interactions for efficient hydrolysis as observed in cassava.27 Thus, together with the observed motif variation and enzymatic divergence, they may represent modest biochemical consequences and fine-tuning to distinguish dietary carbohydrate repertoires. Additionally, slower degradation of tamarind xyloglucan may confer ecological advantages in high-fiber diets: controlled oligosaccharide release can sustain cross-feeding among gut microbes, prevent rapid substrate depletion, and reduce product inhibition. However, given that this assay was performed with purified recombinant proteins on a commercial substrate, we interpret this difference cautiously. Further work, including e.g. catalytic reaction characterization and attack analyses on a broader panel of xyloglucan substrates as well as more detailed comparisons of the enzyme kinetics, will be needed to fully determine the functional and ecological relevance of these differences. Discussion By combining high-resolution metagenomic analyses, structural modeling, and biochemical validation, we provide a compelling example of microbial gene specialization in response to the host diet. The identification of a conserved xyloglucanase gene in Hadza-associated Prevotella genomes was supported not only by population-level genomic data, but also by structural predictions that revealed motif-level divergence near the catalytic site. Importantly, our enzymatic assays verified that these sequence differences are associated with a measurable, although modest, quantitative difference in reaction rate of the enzymes under identical assay conditions. This integrative approach highlights the power of combining computational and experimental methods to uncover fine-scale microbial adaptations and underscores the ecological relevance of gene-level specialization in the human gut microbiome. Our discovery of genomic and functional differentiation in Prevotella sp002251295 underscores how host lifestyle and diet can shape microbial evolution at fine scales. The strong enrichment and structuring of this species in the Hadza gut microbiome suggest selective pressures favor population-specific lineages. This pattern aligns with previous observations of Prevotella as a long-term host-associated genus28 and its documented pangenome variability linked to dietary habits.11 Notably, studies on P. copri have shown that carbohydrate metabolism is central to its colonization strategy, including reliance on plant-derived polysaccharides in murine models.29 Understanding the genetic and functional adaptations of gut bacteria to different diets can inform nutritional strategies and interventions tailored to specific populations. Beyond documenting enzyme divergence, our findings also show that adaptive processes within gut microbiomes can occur through sequence-level modification of conserved genes rather than through taxonomic turnover.2,10 Such fine-scale molecular adaptation may help explain how core microbial taxa persist across human populations while still maintaining dietary responsiveness. Future work combining kinetic analyses of additional xyloglucan substrates, Prevotella growth studies, and in vivo functional validation will further clarify how microbial enzymes evolve to mirror human dietary ecology. Gene 380 and Prevotella in this study are only a small example that exemplifies how host lifestyle and diet can drive evolutionary refinement of microbial enzymes, linking microbial genomic variation, structural adaptation, and biochemical performance into a unified model of functional specialization within the human gut microbiome. Materials and methods Data Fecal samples from humans and chimpanzees in Tanzania were collected between 2015 and 2017. The sample set includes samples from two different wild chimpanzee populations living in Issa Valley and Mahale (n = 27 and n = 63, respectively) and samples from non-indigenous adult humans living around the Mahale Mountains (n = 116), non-Indigenous school children living in the Moshi area (n = 292) and from an Indigenous human population of Hadza hunter gatherers, living in Sengeli (n = 48). The samples were collected and handled as previously described.10 In addition to our primary dataset, we also included metagenomes from three external resources to compare the abundance and prevalence of gene 380 across a broader variety of populations: other Hadza populations2 (n = 469), other Tanzanian populations17 (n = 318), and a combined group of human populations from Asia (n = 72), North America (n = 15), and Europe (n = 92), hereafter referred to as the “Almeida” dataset.18 Genome binning and quality DNA was extracted and fragmented as described in Otani et al.10 Illumina Nova-Seq S4 PE were used for sequencing of libraries to an average of ~105 million reads per sample. The re-sequenced samples were handled as described under the relevant methods. Adapter removal and quality-trimming of the reads were performed using Bbduk230 (version 36.49). High-quality reads were de novo assembled using metaSpades31 (v. 3.13.0). Scaffolds > 2,000 bp were binned into metagenome assembled genomes (MAGs) using VAMB32 (v3.0.1) as described in Jespersen et al.33 We evaluated the quality of MAGs larger than 1 Mbp with CheckM34 (v1.1.3). The nearly complete (NC) MAGs were defined as >0.9 completeness and <0.05 contamination, and these were used for further analysis. Taxonomy, prevalence, and abundance The taxonomy of all NC MAGs was obtained using GTDB-Tk classify_wf35 (v2.3.0), with the release 214 database. To identify species-level genomes (MAG species), we used dRep compare36 (v2.2.3) with mash and Average Nucleotide Identity (ANI) thresholds of 90% and 95%, respectively. dRep dereplication was used to identify the best representative MAG for each MAG species, as the one with the highest score. Additionally, the dRep score was used to select between two MAGs from duplicate samples when necessary. We identified the prevalence of each MAG species as the number of NC MAGs found within the corresponding dRep cluster, and selected MAG species present in at least two indigenous or chimpanzee samples, as well as at least two non-Indigenous samples, for further analysis. We obtained read count abundances for all MAG species using CoverM37 (v 0.6.1). Read counts were scaled according to gene lengths, and zeros were imputed with the zCompositions38 package in R. Likewise, the zCompositions package was used for closing counts and centered-log ratio (clr) calculations. The prevalence of Prevotella sp002251295 and Gene 380 was identified as the number of samples with an abundance above zero. The abundances of Prevotella sp002251295 and Gene 380 were calculated as Transcripts Per Million (TPM) from mappings of trimmed reads with kma;39 abundances were filtered to an expected coverage of 0.5, as in Jespersen et al.33 Phylogeny Based on the marker gene set from GTDB-Tk, the phylogenetic trees representing all (NC) MAG species belonging to selected genera were created with FastTree40 (v2.1.11). For the ten selected MAG species, we created a phylogenetic tree containing all MAGs belonging to each MAG species. To do so, we predicted protein sequences from the MAGs using Prodigal41 (v2.6.3) and used these predicted proteins as input to SonicParanoid42 (v1.3.4) for orthologous gene identification. Prior to inferring gene phylogenetic trees with IQ-TREE43 (v1.6.8), we employed TrimAl44 (v1.4) for the conversion of alignments to the Phylip format. To mitigate ambiguity, we constructed gene trees exclusively from samples containing a single ortholog of a gene, and additionally, a minimum of three samples was required for tree construction. For inferring the overall species phylogeny, we employed ASTRAL45 (v5.7.4), with all orthologous gene trees used as input. IQ-TREE43 was used to adjust the branch lengths. We utilized the ggtree46 (v2.0.4) package in R for visualization of all the genera, species, and gene trees used for the analysis. To identify additional occurrences of gene 380 in samples where Prevotella sp002251295 may have been present but not recovered as MAGs, we searched all contigs across samples using Flankophile.47 To further investigate gene 380 variation within a broader Hadza cohort, we retrieved all MAGs annotated as Prevotella sp002251295 from an external Hadza metagenomic resource.2 For phylogenetic analysis, including the construction of the Ultra-deep Hadza tree and the contig-based gene 380 tree, we used the gene variants identified via Flankophile47 across the respective datasets. Phylogenetic trees were generated using IQ-TREE43 and visualized with the R package ggtree46 (v2.0.4). Functional annotation of genes We functionally annotated all genes using InterProScan48 (v5.36–75.0) with the Prodigal-predicted protein sequences as input. From these functional annotations, we used the GO.db-package49 (v2.1) in R to group genes according to the biological process GO term annotations. For selected gene sequences, we also searched the NCBI50 and CUPP22 databases to obtain a more detailed functional annotation. These genes were selected based on the correlation between host populations and gene variance. To identify genes with high correlation, we performed a permutational multivariate analysis of variance (PERMANOVA) test on all species and gene trees to obtain a host population R2 value. This R2 value denotes how much of the variation found in the tree can be explained by the host population. Only trees with at least two genes representing two different host populations were tested with PERMANOVA (Supplementary Table 1). Additionally, if a host population was represented by a single gene sequence in a tree, we excluded this node from the tree prior to testing. If any re-sequenced samples were present multiple times in a tree, the best MAG chosen to represent the sample was selected based on dRep scores. The p-values from PERMANOVA testing were multi-test corrected with the Benjamini & Hochberg (BH) algorithm.51 Gene synteny For a deeper investigation of host population adaptations, we selected 25 genes to focus on (Supplementary Table 2). These were selected from the genes with the highest R2 values that were also annotated to “carbohydrate metabolic process” from the GO term annotations. To avoid the R2 values being artificially increased by random differences between a low number of samples, we included only genes found in at least ten samples. For the investigation of the flanking sequences of these genes, we employed Flankophile47 (v0.2.10). The minimum coverage and identity were set to 90% to initially include as many gene variants as possible. The flanking region lengths were set to 3600 bp and 5500 bp for C26 for prior to and after the gene sequence, respectively, and to 5000 bp for both for C27. We used a Mantel test from the Vegan package52 in R, to test for correlation between the flanking sequence distance matrix obtained from Flankophile47 and the gene distance matrix from the IQ-TREE43 created gene trees. To adjust for multiple testing, the P-values from the Mantel tests were corrected (Supplementary Table 1) using the BH algorithm.51 PUL prediction and architecture SusC-like, SusD, and Surface-Glycan-Binding-Proteins (SGBP) annotations were derived from the InterProScan48 results. The modular architecture and classification of CAZymes centered around the SusC- and SusD-like pairs were obtained from BLASTP53 analysis towards known representative sequences from the CAZy Database.54 The PUL architecture boundaries were inferred by visual inspection of syntenic PULs from S. copri species in PULdb.55 Protein structure prediction For protein structure predictions, we selected two representative sequences for each of the 25 selected genes. We selected the gene sequences from the gene trees to represent different host populations, e.g. Hadza/non-indigenous or chimp/human. From the genes in the tree belonging to the same host population, one representative gene was selected from the largest “cluster” of a certain host population. If the representative gene sequence was incomplete, the closest, complete gene sequence was chosen, when possible (selected samples can be found in Supplementary Table 2). Prior to generating structural models, we used SignalP56 to predict and remove signal peptides, ensuring that only the mature protein structure was modeled. The structures of all the selected protein sequences were then predicted using AlphaFold219 on the Danish National Supercomputer, Computerome. Default settings were utilized, except for the exclusion of the relaxation step. Among the five output structures generated, the lowest-ranking structure with the highest predicted local distance difference test (pLDDT) score was chosen as the final structural model. The pLDDT scores of the lowest-ranking structure can be found in Supplementary Table 2. Visualization of the resulting AlphaFold2 models was performed using PyMOL57 (v2.3.5). To visualize the most likely location of the active site on the AlphaFold2 structural model, we utilized a crystal structure containing a bound substrate from the Protein Data Bank.58 The identification of the best reference structure with a substrate involved a search for PDB structures with high sequence similarity to the selected protein. Upon finding a suitable substrate structure, an alignment to the AlphaFold2 model was performed, displaying only the substrate and the AlphaFold2 model. A list of the PDB IDs for the selected crystal structures with bound substrates for each chosen protein can be found in Supplementary Table 2. Template modeling (TM) scores were calculated using TMscore.59 To compare structural differences between structural models, we used the color_by_mutation function in PyMOL57 (Supplementary Figure 6). This function aligns the sequences and superimposes the structures. Aligned residues that differ between the two proteins are highlighted and color-coded according to their substitution scores in the BLOSUM90 matrix, indicating the degree of similarity. Specifically, residues in the reference structure are colored blue, while residues in the other structure are colored along a gradient from blue to red, where blue indicates high similarity, and red indicates greater dissimilarity. Residues without differences are shown in gray. Furthermore, we also visualized the pLDDT scores from AlphaFold2, coloring each residue based on its pLDDT score (Supplementary Figure 6). Gene synthesis and cloning into expression vector The coding sequence of the xyloglucanase-encoding gene 380 (≈1400–1500 bp) was custom-synthesized (GenScript, USA) and subcloned into the pET-28a+ vector using the NcoI and XhoI restriction sites to maintain the C-terminal His tag and remove the N-terminal His tag and thrombin cleavage site. The insert was sequence-verified using Sanger sequencing (Eurofins, Germany). The gene was designed with flanking NcoI (5′ end) and XhoI (3′ end) restriction sites; internal sites for these enzymes were avoided by synonymous codon substitution where possible. The provider delivered the plasmid clone (pET-28a+ :gene_of_interest) at >80% supercoiled form. For expression, E. coli BL21 DE3 competent cells were transformed with the plasmid and selected on LB+ kanamycin (50 µg/mL) agar plates. A single colony was used to inoculate 30 mL LB medium preculture supplemented with kanamycin (50 µg/mL), and the preculture was incubated at 37 °C overnight at 180 rpm. The preculture was used to inoculate 1 L LB medium supplemented with kanamycin (50 µg/mL) to a starting OD600 of 0.05, which was incubated at 37 °C until the OD600 of 0.8, induced with 0.25 mM IPTG (Isopropyl β-D-1-thiogalactopyranoside), and incubated at 23 °C for 20 h. at 180 rpm. The cells were harvested by centrifugation at 4400 rpm at 4 °C for 15 min and resuspended in 20 mL binding buffer containing 20 mM Tris, 500 mM NaCl, and 50 mM imidazole at pH 7.4. The cells were lysed by 3 min of active sonication on ice with 10 s pulse and 40 s pause between pulses for a total of 18 rounds. The cell debris was removed by centrifugation at 1100 rpm at 4 °C for 30 min. The supernatant was loaded onto a 2 mL Ni2+ gravity flow column, and the target protein was eluted with 20 mM Tris, 500 mM NaCl, and 250 mM imidazole at pH 7.4. The imidazole was removed from the protein by desalting with a PD-10 (Cytiva, US) column using buffer containing 20 mM Tris, 100 mM NaCl at pH 7.4. Protein verification and concentration were conducted by SDS–PAGE and a NanoDropLite (Thermo Fischer Scientific, US) (data not shown). Enzymatic activity on tamarind xyloglucan substrate An enzymatic time course on tamarind seed xyloglucan (Megazyme by Neogen, Bray Co. Wicklow, Ireland) was conducted in reactions with 0.01 µM enzyme, 2.5 mg/mL xyloglucan in 50 mM UB4 buffer, pH 7.0, at 37 °C for 30 min at 800 rpm. Time points were measured by extracting 60 µL of the reaction, adding to a 1:1 volume BCA reagent (24 g/L Na2CO3, 12 g/L NaHCO3, 1 g/L disodium 2,2’-bicinchoninic acid, 0.625 g/L CuSO4, and 0.6 g/L L-serine), followed by heating at 80 °C for 15 min. After cooling to room temperature, the absorbance was measured at 562 nm and converted to glucose equivalents by a glucose standard curve. For each variant, the experiments were repeated three times, and the resulting activity is the mean of these three experiments. Supplementary Material Jespersen_et_al_Supplementary_Tables_S1_S2.docx Jespersen_et_al_Supplementary_Tables_S1_S2.docx Jespersen_et_al_Supplementary material.docx Jespersen_et_al_Supplementary material.docx Acknowledgments The authors would like to thank the anonymous reviewers for their helpful comments. The authors also acknowledge that this was part of the MLJ- published PhD thesis entitled “Intra-species variations in global, host, and familial microbiomes”, Doctoral thesis, Technical University of Denmark, Denmark 2024. MLJ, FMA, and SO conceived the idea and conceptualized the study. KKM performed the protein structural analyses. SF, BP, and ASM performed the enzyme purification and assays. MLJ analyzed all the metagenomic data, wrote the first draft of the manuscript, and made the figures. MLJ, FMA, and SO interpreted the results and wrote the final version of the manuscript. All authors reviewed and commented on the draft and approved the final version. Funding Statement This project was supported by the Danish International Development Agency (DANIDA) through SeqTanzania project-DFC No. 20-12-TAN and from the Global Health EDCTP3 Joint Undertaking (Global Health EDCTP3) program under grant agreement No. 101103059 (GREATLIFE). MLJ was supported by the Novo Nordisk Foundation (grant NNF14CC0001). Disclosure of potential conflicts of interest No potential conflicts of interest were disclosed. Data availability statement No new sequencing data (FASTQ) were generated in this study. All data used in this study have already been deposited in the European Nucleotide Archive and can be accessed without restrictions under Project ID PRJEB58441 and Study ID: ERP143500 (https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJEB58441; https://www.ncbi.nlm.nih.gov/sra/?term=ERP143500). Ethical statement Not applicable; no human or animal samples were used in the study. Inclusion and diversity We support inclusive, diverse, and equitable conduct of research. Supplemental material Supplemental data for this article can be accessed at https://doi.org/10.1080/29933935.2026.2673265. References Associated Data Supplementary Materials Jespersen_et_al_Supplementary_Tables_S1_S2.docx Jespersen_et_al_Supplementary_Tables_S1_S2.docx Jespersen_et_al_Supplementary material.docx Jespersen_et_al_Supplementary material.docx Data Availability Statement No new sequencing data (FASTQ) were generated in this study. All data used in this study have already been deposited in the European Nucleotide Archive and can be accessed without restrictions under Project ID PRJEB58441 and Study ID: ERP143500 (https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJEB58441; https://www.ncbi.nlm.nih.gov/sra/?term=ERP143500).