Metapangenomics reveals an increased proportion of an Escherichia coli-dominated enterotype in older Chinese people


 Background:Gut microbial communities are likely remodeled along with accumulated physiological decline during aging. Here, we performed a metagenomics-based enterotype analysis in a geographically homogeneous cohort of 367 enrolled Chinese individuals between the ages of 60 and 94, with the goal of characterizing the gut microbiome of the older Chinese individuals and identifying factors linked to enterotype variations.Results:In addition to two adult-like enterotypes were dominated by Bacteroides spp. (ET-Bacteroides spp.) and Prevotella copri (ET-P. copri), we further identified a novel enterotype dominated by Escherichia coli (ET-E. coli), whose prevalence increased in advanced age. ET-E. coli presented some characteristics previously observed in the microbiomes of older people, including low species diversity and diminished abundance of the butyrate producer Faecalibacterium prausnitzii (F. prausnitzii), and exhibited distinct functions as well as a complex, highly stable microbial co-occurrence network, as evidenced by cohesion and natural connectivity analysis. Furthermore, a series of correlation analyses and co-abundance network analyses were carried out and showed that several factors were likely linked to the overabundance of Escherichia members, including advanced age, different vegetable, and fruit intake. Notably, there may be a niche pre-emption of E. coli by P. copri in ET-P. copri via several functional pathways involved in queuosine, braided glycoside, UMP, L-lysine and aromatic amino acid biosynthesis. Overall, our data demonstrated that age explained more of the variance in the gut microbiome than previously identified factors such as type 2 diabetes mellitus (T2DM) and diet. Overgrowth of Escherichia members might occur in a senior's gut where the dominance of core functional microbiota was impaired by long-term changes in dietary habits, gut physiology, and gastrointestinal motility during aging.Conclusion:This study demonstrates an enterotype variation featured by E. coli enrichment in some older Chinese individuals, indicating a potential dissimilation of the gut microbial community with altered host conditions eventually leading to enterotype transitions in advanced age. These findings provide new insights into the changes in the gut microbiome in older age and clues for mechanistic investigations on the driving forces.


Background
As people live longer with the help of modern medicine, it has become more urgent to promote healthy aging. Human aging and disease are deeply in uenced by the gut microbiome [1][2][3][4]. However, the development of dietary recommendations or interventions targeting the gut microbiome of older adults for clinical relevance is challenging because of limited understanding of the large individual variation in the gut microbiome. The advantage of enterotype classi cation is that the large number of individual microbiome pro les can be represented by several typical patterns, which can then be linked to clinical traits to evaluate disease risk. However, few comprehensive analyses in terms of enterotypes have been performed within the elderly population to date. Although the transition of infant enterotypes to adult-like enterotypes has been observed at an early age [5][6][7], a transition from the adult-like enterotypes to an aging-associated enterotype in older individuals has not been con rmed.
Although several common age-related trajectories of the human gut microbiome have been observed in a Han Chinese population-based cohort of 2,338 adults (26-76 years), as well as in Israeli and Dutch adults, some of these characteristics, such as diversity, are inconsistent with the results reported in previous studies on the gut microbiota of the elderly [2,8,9]. This discrepancy is probably because dynamic remodeling is highly modi able during aging, which causes the microbial communities in older individuals to be more complex and variable than those in younger adults [4,10]. Moreover, previous studies have shown that the large inter-and intrasubject variation across gut microbiome studies performed across different regions and in populations with wide age ranges complicates the signatures of the aging-associated microbiome [11].
Therefore, we performed metagenomic-based enterotype analysis to characterize the aging-associated microbiome in a geographically homogeneous cohort of older individuals to minimize the variance in diet, lifestyle, and genetics. Comprehensive analysis of gut microbiomes was performed to investigate enterotype-speci c taxonomic and functional differences. Moreover, several ecological theories and tools [12][13][14] were employed in this study to delineate competing and cooperating microorganisms, compare stabilities between co-occurrence networks, and identify key factors that may be linked to these variations. In our cohort of Chinese people older than 60 years of age, the effect of age on the human gut microbiome was more potent than that of previously reported factors, such as chronic noncommunicable diseases such as type 2 diabetes mellitus (T2DM) [15] and long-term dietary habits [16]. Our study expands earlier observations on enterotypes based on the PG cohort reported by Zhang et al. to include a population older than 76 years [8]. In addition to the two adult-like enterotypes (ET-P. copri and ET-Bacteroides spp.) reported previously, we rst identi ed and characterized an Escherichia coli-enriched enterotype in the elderly population, which exhibits several aging signature as previously reported, such as low diversity [1] and low abundance of F. prausnitzii [17,18]. Our results provide new insights into the aging-associated microbiome and enhance our understanding of how age and other external in uencing factors contribute to enterotype variations.

Study cohorts
In this study, we used metagenomic shotgun sequencing to investigate the microbial community structures of fecal samples from 367 older Chinese individuals (154 males and 213 females) aged 60 to 94 years. Time-matched clinical phenotypic data were collected to investigate external in uencing factors that are associated with enterotype variations in advanced age. Advanced organic disease (end-stage cancer or renal or liver disease) and cognitive impairment were exclusion criteria. Participants could not have taken antibiotics within 6 months prior to enrollment, proton pump inhibitors, H2 receptor antagonists, tricyclic antidepressants, narcotics, anticholinergic medications, laxatives, or antidiarrhea medications within 4 weeks of enrollment, or antacids within 2 weeks prior to enrollment. Each volunteer completed questionnaires to provide clinical information, including demographic, lifestyle, and diet frequency data and sleep information. T2DM, dyslipidemia, hypertension, frailty, and sarcopenia were diagnosed by trained doctors. The Chinese volunteers enrolled were living in three adjacent residential communities in Shaoxing City, Zhejiang Province (in the Yangtze River Delta metropolitan area of China). These communities were formed during the process of rural urbanization, with most residents initially making a living by agriculture and gradually shifting to industrial work since the 1980s while maintaining traditional large family settlement patterns with modern urban lifestyles.

Sample collection
Fecal samples of all 367 participants were collected after defecation at the visit and transported to the laboratory in liquid nitrogen. 223 participants were willing to donate 5 ml blood sample for our research.
Fasting blood samples were taken between 7:00 and 9:00. Blood and fecal samples were stored at −80 °C for long-term biobanking until total DNA was extracted for sequencing.

Bioinformatic processing
We used metagenomic shotgun sequencing to analyze the fecal samples. DNA from the fecal samples (~200 mg) was extracted from the supernatant using a QIAamp DNA Stool Mini Kit (Qiagen) according to the manufacturer's instructions. DNA library construction based on DNA nanospheres and combinatorial probe-anchor synthesis-based shotgun metagenomic sequencing at a depth of 5 Gb per sample for all 367 samples were performed at MGI (Shenzhen, China) using their routine procedures. The overall accuracy (≥0.8) control strategy was used to perform quality control on the raw sequencing reads to remove low-quality reads. Sequencing libraries were metagenomically classi ed using MetaPhlAn2 [19] to obtain gut microbial pro les that included bacteria, archaea, eukaryotes, and viruses. Taxon-speci c community functional pro les were generated using Human Microbiome Project (HMP) Uni ed Metabolic Analysis Network 2 (HUMAnN2) [20]. The National Center for Biotechnology Information (NCBI) database (2014 edition) and HUMAnN2 were used to annotate the nonredundant gene sets. The functional pro les were annotated according to Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathways [21] and MetaCyc-reactions functional categories as described in previous studies [22]. The relative abundances of each taxonomic unit were generated by summing the relative abundances of their annotated genes per individual [15].
For gut microbial diversity analyses, diversity indexes (Shannon and Simpson) were calculated based on the abundance pro les at the species level using the 'diversity' function in the R package 'vegan' (v2.5-6).
To investigate the difference in metabolic functional pro le among enterotypes, Wilcoxon rank-sum tests were performed to compare the relative abundance of functional pathways grouped into different metabolic categories. Jensen-Shannon divergence (JSD) distance and the partitioning around medoid (PAM) clustering algorithm were applied to cluster samples based on the microbial functional pro le, and sparse inverse covariance estimation for ecological association inference (SPIEC-EASI) using the glasso model was applied to reduce the pathway dimension to the module [23]. Spearman's rank correlations were used to determine the relationship between external variables and functional pathways and to determine the co-abundance that exhibits the relationship between species and functional pathways [24,25]. All the undirected network graphs were visualized using Gephi (https://gephi.org/).

Enterotype clustering
Enterotypes were classi ed in the cohort (n = 367) according to methods previously described by Arumugam et al[26]. Low-abundance and low-prevalence ltering was applied to exclude taxonomic features whose relative abundance across all samples was lower than 0.1% and whose prevalence was lower than 10%. PAM using the JSD of the normalized species counts was performed to cluster the samples. The analysis was performed using the 'ade4', 'cluster', and 'clusterSim' packages in R. The Calinski-Harabasz (CH) index indicated that the optimal number of clusters for this dataset was 3 (Supplementary Figure S1A). The dominant species with both high relative abundance in samples and high mean decrease accuracy were considered the main contributors to each enterotype. The signi cance of the model and cross-validated R 2 were evaluated with 5,000 permutations of the response variable (ecosystem multifunctionality) using the 'A3' R package.

Co-occurrence network and diversity analyses
For the co-occurrence network analysis, the bacterial correlations in each enterotype were computed based on the abundance of each species using the SparCC function of the 'SpiecEasi' package with 100 bootstraps to estimate the P value constructed using the 'SpiecEasi' package in R. Prior to this analysis, species with prevalences lower than 20% were removed to decrease noise. Absolute correlation values higher than 0.4 with a P value <0.05 were retained. The closeness of species in the main microbial community groups was calculated to measure node centralities in each network. To identify external in uencing factors of microbial co-occurrence networks, 28 variables were considered as potential covariates, including demographic factors, chronic noncommunicable diseases, sleep time per day, habitual long-term diet information, and information about participation in sports or physical work. The correlation coe cients between potential covariates and topological features were calculated using Spearman's rank correlation test. Multiple regression on distance matrices (MRM) in the ecodist package was used to estimate the importance of potential covariates. The robustness of a network was revealed by the natural connectivity of a complex network [13]. Positive and negative cohesion values were calculated as described in previous studies [14]. We calculated the NetShift (NESH) score to quantify community changes in gut microbial co-occurrence networks between enterotypes (https://web.rniapps.net/netshift) [12]. Species with NESH score values > 2 were identi ed as important microbial taxa that serve as 'drivers' facilitating prominent changes in networks.

Identi cation of external variable associations with the three enterotypes
Potential covariates, including clinical characteristics, dietary habits, and participation in sports or physical work, were used as explanatory variables in redundancy analysis (RDA) computed based on Bray-Curtis dissimilarity matrices from the relative abundances of species using the vegan package. The associations between potential covariates and the overall bacterial composition were determined by PERMANOVA at the species level. Spearman's rank correlation test was used to calculate the correlations between potential covariates and the relative abundances of selected species as the major contributor to each enterotype identi cation. We also applied the multiple regression model with variation decomposition analysis to estimate the importance of the potential covariates in explaining the dissimilarities in microbial communities and differences in the relative abundance of selected species and microbial metabolic pathways. This analysis was performed using the 'psych', 'reshape2', 'relaimpo', and 'packfor' packages in R [27].

Piecewise linear regression analysis of the relationships between frailty risk markers and age in each enterotype
We assessed peripheral concentrations of fasting plasma glucose (FPG), hemoglobin A1c (HbA1c), creatinine, high-sensitivity C-reactive protein (hs-CRP), P, Ca, folic acid, and albumin levels using standard in-hospital methods. The relationships between age and peripheral levels of various risk markers were visualized with loess regression models. The breakpoint (age threshold) was determined using a piecewise linear regression analysis with the R package SiZer v.0.1.5 [28]. For the analysis of FPG, the data from patients with T2DM were excluded due to their abnormal blood glucose levels.

Statistical analysis
All statistical analyses were conducted and visualized in R within RStudio. The statistical signi cance of differences in the relative abundance of each taxon among enterotypes was determined by the Kruskal-Wallis H-test with Benjamini-Hochberg false discovery rate (FDR) adjustment. The statistical signi cance of the differences between enterotypes in the top 25 most dominant species with average relative abundance >1% in the cohort was also determined by the Wilcoxon rank-sum test. Correction for multiple testing was based on the Benjamini-Hochberg FDR, with the corrected P value cutoff set at 0.05.

Results
3.1. Three enterotypes were identi ed in the gut microbiome of elderly individuals.
In this study, we performed a metagenomics-based enterotype analysis in a cohort of 367 enrolled Chinese seniors between the ages of 60 and 94 years. After bioinformatic analyses, taxonomic assignment showed that 98.995%, 0.736%, 0.101%, and 0.001% of the sequencing reads corresponded to the domains Bacteria, Viruses, Archaea, and Eukaryota, respectively. Using JSD distance and the PAM clustering algorithm, we identi ed three enterotypes in this cohort (Supplementary Figure S1A). The predominant species identi ed by vefold cross-validation analysis using a random forest model were Prevotella copri and Prevotella stercorea in enterotype 1 (designated ET-P. copri), Bacteroides uniformis and Bacteroides coprocola in enterotype 2 (designated ET-Bacteroides spp.) and E. coli in enterotype 3 (designated ET-E. coli) ( Figure 1A and 1B). The gut microbial communities of the E. coli-dominated enterotype exhibited a distinctive taxonomic pro le compared with those of ET-P. copri and ET-Bacteroides spp. (Figure 1C-1F). For β-diversity analysis, principal coordinate analysis (PCoA) plots demonstrated that sample clustering at the order level did not result in a notable separation between ET-P. copri and ET-Bacteroides spp., but most ET-E. coli samples were well separated from these other enterotypes (Supplementary Figure S1B), the separation of all three enterotypes became clearer with clustering at the species level (Supplementary Figure S1C). The 25 most dominant taxa showed signi cantly different abundance patterns across enterotypes (Supplementary Figure S2). Notably, F. prausnitzii, a key anti-in ammatory bacterium that is depleted in early frailty [17], was found to be signi cantly enriched in ET-P. copri but depleted in ET-E. coli, accounting for 8.31% of ET-P. copri, 6.88% of ET-Bacteroides spp. and 3.20% of ET-E. coli (Supplementary Figure S2B). The taxonomic differences across enterotypes are further summarized in Additional le 1. Overall, these results demonstrated enterotype variation in some older persons, characterized by a remarkable enrichment of E. coli in their gut microbiomes.

Enterotype variation in the older Chinese population is associated with older age.
To better understand the associations between enterotypes and health status in the elderly population, we examined 28 variables as potentially signi cant microbiota covariates, including demographic factors (2), chronic noncommunicable diseases (10), sleep time per day, habitual long-term diet information (11), and participation in physical work, daily indoor exercising, or outdoor sports (4). The RDA showed that age, frailty, sarcopenia, T2DM, and fruit consumption had a higher impact on ET-E. coli samples, while vegetable consumption and other potential covariates had opposite effects on gut community composition ( Figure 2A). Among the variables tested in this study, the results of PERMANOVA clearly suggested that age was the strongest factor associated with the overall microbiota composition, with 3.61% explained variance (Adonis analysis calculated based on Bray-Curtis similarity, P < 0.001, Figure  2B). Additional variables achieving signi cance, in order, were constipation, egg consumption, bone loss, T2DM, vegetable consumption, gastrointestinal disturbances, soy consumption, and stroke. Our results indicated that the enterotype prevalence in the elderly population is related to age. Speci cally, ET-P. copri was the most prevalent enterotype among people younger than 75 years of age, whereas people over 75 years of age tended to harbor ET-E. coli or ET-Bacteroides spp. The prevalence of the E. coli-dominant enterotype tended to increase dramatically with age. ET-Bacteroides spp. was present at similar rates across all ages studied ( Figure 2C). Deployment of piecewise linear regression analysis enabled us to visualize a clear difference in the age-related trajectories of representative blood in ammatory and nutritional marker levels among enterotypes. Many of them showed generally different trends of associations with age in the older adults with ET-Bacteroides spp. or ET-E. coli, characterized by different slopes and age breakpoints of the relationships ( Figure 2D), suggesting the inconsistent pace of aging in the individuals with different enterotypes.
We next integrated the Spearman correlation coe cients with other statistical analyses to quantify the associations between key distinguishing bacterial species for the three enterotypes with each variable ( Figure 2E). The data showed that potential covariates explained 14.53% of the variation in abundance for P. copri across individuals but only 4.51% for E. coli, indicating that the abundance of P. copri in the elderly population is more susceptible to external in uences. In addition to age, the relative abundance of unclassi ed Escherichia was positively correlated with fruit intake but negatively correlated with vegetable intake and constipation. These results indicated that the altered gut microbiome in older age might be a result of long-term adaptation to accumulated changes in dietary habits, gut physiology, and gastrointestinal motility.
3.3. Distinctive pattern of the microbial co-occurrence network in ET-E. coli.
Most microorganisms living in the gut are auxotrophs whose growth relies both on external nutrients obtained from the diet or metabolism of the host and the exchange of electron donors, amino acids, vitamins, and other cofactors with other microbes [29]. Thus, a robust steady-state bacterial consortium is formed in the co-occurrence community based on cooperative interactions for low metabolic costs and e ciency within member species. In this study, co-occurrence network analysis was performed to predict biotic interactions such as resource competition and metabolic cross-feeding within the residing microbial community, as revealed by negative edges and positive edges, respectively. We investigated the difference in co-occurrence communities of the three enterotypes to look for potential in uencing factors linked to transitions between the enterotypes (Figure 3). Approximately 82-145 species (nodes) and 117-502 connections (edges) were retained at a correlation cutoff of 0.4 in the co-occurrence network for an enterotype, and most nodes and edges were speci c to each enterotype (Supplementary Figure S3). The maximum number of unique nodes and edges was found in the co-occurrence network of ET-E. coli, which exhibited a highly complex co-occurrence network ( Figure 3A).
There was a signi cant difference among the network-level topological features of the co-occurrence networks of the three enterotypes, including degree assortativity, average path length, and betweenness centralization (Kolmogorov-Smirnov test, P < 0.001, Additional le 2). Fourteen nodes with NESH score values > 2 were considered the key drivers facilitating the main changes between co-occurrence networks, and most of them showed signi cantly different enrichment or depletion across enterotypes, except Lactobacillus salivarius and Streptococcus salivarius (Supplementary Figure S4). To identify key factors affecting the diversity and stability of the human microbiome, external in uencing factors were assessed based on their correlations with these fourteen key drivers and topological features. For the microbial network of ET-Bacteroides spp., constipation was positively correlated with average path length (Spearman's r = 0.335, P <0.01) and negatively correlated with degree assortativity (Spearman's r =-0.32, P <0.01). Interestingly, opposite associations with degree assortativity were also observed for vegetable and fruit consumption (Spearman's r=-0.334, P <0.01, for vegetable consumption and Spearman's r= 0.304, P <0.01, for fruit consumption). Similarly, the relative abundance of S. salivarius was positively correlated with age and fruit consumption in seniors and negatively correlated with constipation and vegetable consumption (Spearman's r = 0.31, P <0.0001, for age, Spearman's r = -0.37, P <0.0001, for constipation, Spearman's r = -0.34, P <0.0001, for vegetable consumption, and Spearman's r= 0.33, P <0.0001, for fruit consumption). These results suggest that constipation symptoms and preference between fruits or vegetables may contribute to variations in microbial co-occurrence networks in different enterotypes by changing the interactions among several speci c microbes, such as S. salivarius (Supplementary Figure  S5).
To investigate the bacterial consortia in each co-occurrence community, we analyzed the ve main microbial community groups (group I to group V) of co-occurring species that appeared across enterotypes, whose total members accounted for 36% of the nodes in ET-P. copri, 44% in ET-Bacteroides spp., and 45% in ET-E. coli (Supplementary Figure S6A). We found that these main microbial community groups consisted mainly of taxa from the same or closely related species (Supplementary Figure S6B). Speci cally, 13 species appeared in group I across all three enterotypes, mainly from the genera Streptococcus, Veillonella, Granulicatella, Lactobacillus, and Rothia, 6 species mainly from the genus Alistipes in group II, and 7 species mainly from the genera Clostridium, Blautia, Coprobacillus, and Flavonifractor in group III. The main microbial community groups in ET-Bacteroides spp. and ET-P. copri consisted of similar components, suggesting high community homogeneity in their co-occurrence network when compared with that of ET-E. coli. In line with that hypothesis, the nodes of ET-E. coli showed signi cantly higher NESH scores than those of ET-Bacteroides spp. or ET-P. copri (Wilcoxon test, P < 0.001, Additional le 3 and Supplementary Figure S7). Notably, there were more oral microbes in cooccurrence group I of ET-E. coli co-occurrence network.
The stability of the microbial co-occurrence network was quanti ed by the robustness of the microbial networks via a natural connectivity analysis. The data showed that the gut microbial network of ET-E. coli had the highest robustness and the strongest cohesion ( Figure 3B-3E). In addition, ET-E. coli exhibited the lowest α-diversity indexes among the three enterotypes and the exclusive growth of one single cooccurrence group in an older individual, implying that many more out-group species were suppressed by a strong colonization resistance in ET-E. coli ( Figure 3F-3H and Supplementary Figure S8). However, within ET-E. coli samples, Spearman's test showed a signi cantly positive correlation between the positive cohesion value and Shannon index, suggesting that the more microbes that bene ted from cooperative interactions within a group, the lower the potential risk of extinction due to environmental degradation ( Figure 3I).

Gut microbial function differences among enterotypes.
For this analysis, genes were grouped based on functional pathways. We found that PAM clustering of ET-E. coli samples was also supported based on microbial functional pro les. Most ET-E. coli samples grouped into a cluster discrete from the cluster consisting of the adult-like enterotype ET-P. copri and ET-Bacteroides spp. samples ( Figure 4A). The distinct functionality of the gut microbiome dominated by E. coli is shown in the heatmap in Supplementary Figure S9. Obviously, the functional pro le of ET-E. coli varied markedly from those of ET-Bacteroides spp. and ET-P. copri. Our data showed that 319 of 431 microbial metabolic pathways presented statistically signi cant differences among the three enterotypes, as identi ed by the Wilcoxon rank-sum test ( Figure 4B, Additional le 4). Of the 158 pathways with 100% prevalence in the cohort, 128 pathways presented signi cant differences in their abundance among enterotypes ( Figure 4C). Overall, the core microbial functionality that facilitates bacterial survival is performed by generally different functional pathways in ET-E. coli and the adult-like enterotypes, with involvement of all metabolic categories (Supplementary Figure S10). We next investigated the patterns of age-associated changes in the functional pro le in each enterotype. The trend of the decreased relative abundance of metabolic pathways, such as glycolysis, glycerol degradation, stachyose degradation, and purine ribonucleoside degradation, with age was similar in ET-Bacteroides spp. and ET-P. copri. The relative abundance of several amino acid biosynthesis pathways involved in L-threonine, L-lysine, L-arginine, and L-ornithine biosynthesis diminished with age in ET-P. copri but gradually increased with age in ET-Bacteroides spp. and ET-E. coli ( Figure 4D).
The SPIEC-EASI method was utilized to identify functional modules that were constructed by coabundant functional pathways ( Figure 4E). Module I consisted of the highly abundant and highly prevalent functional pathways that were signi cantly enriched in the two adult-like enterotypes (ET-P. copri and ET-Bacteroides spp.), suggesting their particular importance for the core functionality of the human gut microbiome, whereas module II was composed mainly of the functional pathways signi cantly enriched in ET-E. coli, featuring many functional pathways involving cofactor, carrier, and vitamin biosynthesis (Supplementary Figure S9). Interestingly, vast negative correlations existed between most ET-P. coprienriched pathways and ve ET-Bacteroides spp.-enriched pathways in module I with some ET-E. colienriched pathways in module II, which indicated the potential participation of these microbial functions in the competition between the predominant bacteria of the enterotypes. The 336 negative correlations (edges) are listed in Additional le 5.

Factors associated with enterotype variations in the older population.
To investigate the factors associated with enterotype variations, especially for the overgrowth of E. coli in older individuals, we further analyzed differences in species-pathway and species-host associations among the three enterotypes. Application of Spearman correlation coe cient analysis showed that members of co-occurrence group I (mainly from Streptococcus and Veillonella) generally might have a niche overlap with Enterobacteriaceae members (e.g., Klebsiella pneumoniae, Enterobacter cloacae, E. coli, and unclassi ed Escherichia), who presented growth optima involving various metabolic pathways ( Figure 5). Their close relationship existed in all enterotypes. In contrast, only in ET-P. copri, it was found that nine metabolic biosynthesis pathways showed a positive correlation with P. copri but a negative correlation with E. coli, suggesting their potential roles in the competition between microbes. These nine pathways were including queuosine, UMP, L-lysine (L-lysine biosynthesis III and L-lysine biosynthesis VI), and aromatic amino acid biosynthesis. We also observed similar relationships of E. coli with F. prausnitzii linked by genes involved in L-isoleucine biosynthesis I and Megamonas unclassi ed linked by genes involved in Calvin-Benson-Bassham cycle, pentose phosphate pathway (non oxidative branch), and stachyose degradation. These results indicated the growth of E. coli might be suppressed in ET-P. copri via a niche pre-emption generated by P. copri, F. prausnitzii, and Megamonas unclassi ed.
We further analyzed the correlation between the relative abundances of metabolic pathways and age and other external in uencing factors ( Figure 6). Consistent with the apparent age-associated increase in the relative abundance of several amino acid biosynthesis pathways involved in L-threonine, L-lysine, Larginine, and L-ornithine biosynthesis in the enterotypes dominated by Bacteroides spp. and E. coli, signi cant positive correlations of these pathways with age were con rmed in the whole cohort using Spearman's rank correlation test. Interestingly, inverse associations were observed in two polyamine biosynthesis pathways and two functional pathways involving the biosynthesis of proin ammatory components of bacterial wall components, such as O-antigen and UDP-N-acetyl-D-glucosamine, and their correlations with age and vegetable consumption were also inverted. In addition to two functional pathways involved in cell structure biosynthesis, most of the functional pathways involved in amino acid biosynthesis (except L-lysine biosynthesis) were positively correlated with age and fruit consumption but negatively correlated with vegetable consumption. Notably, the relative abundance of two functional pathways involved in L-lysine biosynthesis (L-lysine biosynthesis III and L-lysine biosynthesis VI) was negatively associated with age and frailty and positively associated with vegetable, milk, and soy food consumption.

Discussion
Although a previous study showed that some older people harbor higher levels of Enterobacteriaceae (the family including E. coli) than younger adults [2,30,31], the dominance of E. coli in the gut microbiota might not have been observed in previous studies due to the poor accuracy of 16S rRNA sequencing for Enterobacteriaceae and Escherichia/Shigella [32]. This work included a metagenomic-based enterotype analysis and rst reported the association of an increased prevalence of ET-E. coli with advanced age in a cohort of older Chinese individuals with geographic homogeneity. Compared with the previously reported effects of single factors on gut microbiomes [16,33], such as chronic noncommunicable diseases and long-term dietary habits, we show that the effects of age are much stronger, suggesting that the enterotype variations in older people may be a result of altered environmental factors superimposed on pathological changes and physiological aging. Moreover, we found that different enterotypes showed inconsistent age-related trajectories of representative blood in ammatory and nutritional markers, such as creatinine, whose increasing serum level indicates a higher risk of sarcopenia and low renal function in older individuals [34]. These ndings in our study partly support the hypothesis that interindividual differences in the gut microbiota may indicate different degrees of aging for older individuals, just as differences in the intestinal bacterial community of infants in terms of enterotype represent different stages of maturation in early life[6].
We found that the relative abundance of unclassi ed Escherichia was inversely correlated with preference for fruits or vegetables, although both are foods rich in dietary ber and carbohydrates. One interpretation is that more live bacteria may be consumed when eating fruit than when eating heat-treated vegetables and that the swallowed bacteria cannot be eliminated effectively in older people due to a decline in the gastric acid barrier and impaired immune function. Although the reasons for the evident overgrowth of E. coli in the gut are not well understood, some available evidence implies that it may result from a long-term adaptation of the gut microbiome to an aging host with a selective advantage in the presence of immune dysfunction. For example, it has been observed in some infants that the domination of E. coli was abolished as their immune function developed [5]. Animal studies have shown that nonpathogenic E. coli could thrive in IL-10 KO mice and complement factor D-de cient mice, appearing to the bacteria from the Firmicutes phylum rather than Bacteroidetes [35,36]. Consistent with this nding, we found that E. coli grew particularly abundantly in elderly people (>75 years) with signi cantly reduced relative abundances of the anti-in ammatory F. prausnitzii and Ruminococcus members. This observation suggests an association between E. coli overgrowth and immune senescence and in ammatory aging. On the other side, overgrowth of Escherichia members might occur in a senior's gut where the dominance of core functional microbes, such as P. copri, F. prausnitzii, and Megamonas unclassi ed, was lost. Inhibition of E. coli growth by resident microbial communities has been observed in vitro using a gut microcosm approach [37]. Our results support the mechanisms by which such suppression of invaders may be generated by resource competition. We found notable negative relationships between some signature pathways of the adult-like enterotypes and those of ET-E. coli. It is likely that the dominant growth of P. copri may deplete these substrates, such as L-aspartate for the biosynthesis of UMP and L-lysine, in the microenvironment and exclude other competitors overlapping the ecological niche. However, the metabolic intermediates and end-products in microbial resource competition have not been con rmed. It would be helpful to investigate the metabolic mechanisms involved in the ecological niche competition between E. coli and P. copri or other species by well-designed cell culture analyses in the future.
The redundancy of members in the co-occurrence network could mitigate the threat of extinction of gut microbes. In this cohort of older individuals, we found the highest robustness in the co-occurrence network of ET-E. coli, suggesting a greater capacity to resist uctuations driven by the loss of members and a better adaptation to a senescent host [38]. Conversely, the weak robustness of ET-P. copri indicates the vulnerability of this ecosystem, which might explain why this enterotype disappears substantially with age in the population [29]. On the other side, negative cohesion in microbial cooccurrence networks may represent competition among co-occurrence groups of species with overlapping ecological niches, which may cause overactivation of host immune function or depletion of nutrient substrates to inhibit other organisms from growing nearby, nally resulting in disruption of intestinal homeostasis [39,40]. The highest absolute value of negative cohesion in ET-E. coli suggested the strongest colonization resistance between the species in different co-occurrence groups. Accordingly, the predominance of E. coli or a single co-occurrence group of species was observed in most older individuals with ET-E. coli. In contrast, harmonious coexistence of different co-occurrence groups of species was present in the microbial community dominated by Bacteroides spp. And α-diversity indexes revealed that ET-Bacteroides spp. samples showed the highest diversity, whereas ET-E. coli samples showed the lowest diversity. Thus, it is likely that in the gut microbial community dominated by E. coli, the "friend or foe" style survival strategy adopted by a few groups might result in a hostile and deteriorating environment for out-group members and reduce species diversity. That may explain the inconsistent reports of the gut microbial diversity in older individuals across studies, which may be caused by different proportions of enterotypes like ET-E. coli and ET-Bacteroides spp. in their observed cohorts [1,30].
The complex microbial interactions of ET-E. coli are also associated with the increased ectopic colonization of multiple oral bacteria in the gut, which has been considered as an important gut microbiota signature of aging and cancer [9,41]. We found that the lactic acid producers Bi dobacterium spp. disappeared in co-occurrence group I of the ET-E. coli co-occurrence network, whereas multiple members of the oral microbiota are present. Some of them are opportunistic human pathogens, such as Atopobium parvulum. A. parvulum has been reported to control the central hub of H 2 S-producing bacterial communities, such as Streptococcus and Veillonella, in colorectal cancer [42,43].
Our study would contribute to the development of enterotype-speci c nutritional management for healthy aging. The gut bacterial communities dominated by P. copri and Bacteroides spp. presented quite similar functional pro les. In contrast, loss of core module I and gain of module II make the E. colidominated gut microbiome functionally different from the two adult-like enterotypes. Speci cally, several functional pathways involved in the amino acid biosynthesis of L-ornithine, L-arginine, and L-lysine showed low abundance in ET-P. copri. Additionally, the relative abundance of the functional pathways involved in polyamine biosynthesis is negatively correlated with age and shows an inverse association with functional pathways involving the biosynthesis of the O-antigen building block (E. coli), a component of E. coli lipopolysaccharide. As the anti-aging properties of polyamines have been determined in animal and clinical studies [44], older individuals may bene t from dietary supplementation with polyamines. In addition, live biotherapeutic agents could be used to replenish the lost core gut bacteria in ET-E. coli during aging, such as F. prausnitzii.
A limitation of this study is that there are insu cient well-controlled longitudinal human studies to support the hypothesis that the enterotype transition occurs in an individual during aging or that the mortality rates of older persons vary by enterotype. In addition, patterns of enterotype transition during aging and the underlying mechanisms remain unclear due to a lack of high-resolution temporal data based on longitudinal cohorts with large sample sizes to clarify age-dependent changes. Moreover, it is di cult to extrapolate fundamental aspects of nonpathogenic E. coli ecology and their effects on host health status from studies of pathogenic and/or antibiotic-resistant strains from an unhealthy person, whereas most nonpathogenic E. coli strains living in the human gut may be common residents [45]. Therefore, caution should be used when assuming that ET-E. coli is detrimental, although overgrowth of E. coli may present a threat of fatal endogenous infections to older patients with immunosenescence and diminished gut barrier function because of its abundant antibiotic resistance genesis and high tolerance to the in ammatory environment [46]. Future comprehensive examinations in humanized gnotobiotic mice and human fecal transplantation may provide a better understanding of the relationship of ET-E. coli and healthy aging with regarding to its impacts on nutritional metabolites, host energy hemostasis, and immunosenescence.

Conclusion
Our data revealed an enterotype dominated by E. coli strongly associated with advanced age, indicating that enterotype transitions may occur not only in infants at an early age but also in some individuals of advanced age. Given that the microorganisms living in the intestine play an important role in maintaining host health, the aging-associated microbiome characterized by an overabundance of E. coli might profoundly affect age-related changes in metabolism, diet nutrient absorption, intestinal functions, and the immune system. Our integrated analysis of the gut microbial pro le in older individuals will help researchers better understand the biological changes in the gut microbiome that occur with aging. This has important implications for the prediction of aging and health loss, as well as investigation of the occurrence of mechanisms that can result in an aging-associated gut microbiome.    Dramatic variation in functional pro le in older persons with the Escherichia coli-enriched enterotype.
A: Principal coordinate analysis (PCoA) for visualization of functional clusters based on partitioning around medoids (PAM) of Jensen-Shannon distance (JSD). The point color represents the enterotype (red, ET-Bacteroides spp., gold, ET-E. coli, blue, ET-P. copri), and the shape represents the functional cluster (triangle, cluster I, square, cluster II). B: Functional differences identi ed by Wilcoxon rank-sum test are based on the metagenome sequence data. Samples are colored to match their enterotype assignments. C: One hundred twenty-nine microbial metabolic pathways were signi cantly differentially abundant among the three enterotypes. The scatterplot shows only the pathways with 100% prevalence in our cohort. The ten most signi cantly differentiated metabolic pathways of the three enterotypes are labeled. D: Enterotype-and age-speci c patterns in the metabolic pathways. The heatmap shows the top 44 most signi cantly differentiated metabolic pathways of the three enterotypes, with relative abundance varying with age. E: The descriptions of the three modules are as follows. Module I: Core functional pathways that are signi cantly enriched in ET-P. copri or ET-Bacteroides spp. Module II: ET-E. coli-associated functional pathways. Module III: Scarce functional pathways that are present only in the minority with low relative abundance. The size of each node is proportional to the mean relative abundance.