Age-dependent remodeling of gut microbiome and host serum metabolome in mice

The interplay between microbiota and host metabolism plays an important role in health. Here, we examined the relationship between age, gut microbiome and host serum metabolites in male C57BL/6J mice. Fecal microbiome analysis of 3, 6, 18, and 28 months (M) old mice showed that the Firmicutes/Bacteroidetes ratio was highest in the 6M group; the decrease of Firmicutes in the older age groups suggests a reduced capacity of gut microflora to harvest energy from food. We found age-dependent increase in Proteobacteria, which may lead to altered mucus structure more susceptible to bacteria penetration and ultimately increased intestinal inflammation. Metabolomic profiling of polar serum metabolites at fed state in 3, 12, 18 and 28M mice revealed age-associated changes in metabolic cascades involved in tryptophan, purine, amino acids, and nicotinamide metabolism. Correlation analyses showed that nicotinamide decreased with age, while allantoin and guanosine, metabolites in purine metabolism, increased with age. Notably, tryptophan and its microbially derived compounds indole and indole-3-lactic acid significantly decreased with age, while kynurenine increased with age. Together, these results suggest a significant interplay between bacterial and host metabolism, and gut dysbiosis and altered microbial metabolism contribute to aging.


INTRODUCTION
Aging is a complex trait that is influenced by individual genetics, diet, lifestyle, and environmental factors. The increase in life expectancy is often accompanied with additional years of susceptibility to chronic disorders such as obesity, insulin resistance, and cognitive impairment [1][2][3][4]. Furthermore, aging is increasingly recognized as being associated with a pro-inflammatory state that promotes the development of chronic diseases, hence the term "inflammaging" [5]. During the past decade, accumulating evidence suggests AGING that the gut microbiota impacts body weight, energy homeostasis, innate immune system and aging [6][7][8].
For example, epidemiological studies show that loss of diversity in the core microbiota groups is associated with increased frailty [9,10]. It is important to note that gut microbiota do not age per se, but people growing older may experience comorbidities associated with the gut and with gut bacteria [11]. How the microbiota in the gut affect the aging process, or whether gut microbiota simply change as a function of age, remain unclear.
Recently, advances in analytical platforms have accelerated metabolomics studies, which aim to systematically identify and quantify a global set of metabolites or low-molecular-weight intermediates (molecular weight < 1.5 kDa) in a given biological system in response to pathophysiological stimuli, genetic modification, or environmental factors [12]. Several metabolomics studies reported age-related changes in metabolite levels and found strong correlation between metabolomics profiles and chronological age as well as longevity [13][14][15][16][17][18]. Signature metabolites that are associated with aging include amino acids, lipids, carbohydrate, TCA cycle, and redox metabolism; however, how these diverse metabolites regulate the aging process and its complex networks remains unclear. Furthermore, most of the serum metabolome studies assess the fasting state, using blood samples collected after overnight fasting [13][14][15][16][17]. Interestingly, a previous study comparing plasma metabolomes of conventional and germ-free mice at the fed state demonstrates a significant effect of the gut microbiome on mammalian blood metabolites [19]. For example, the bacterialmediated production of bioactive indole-containing metabolites derived from tryptophan is significantly impacted, and colonization of bacterium Clostridium sporogenes restores the production of indolemetabolites. Hence, these results suggest a significant interplay between bacterial and host metabolism, when plasma profile was assessed at the fed state.
In this study using mice at different ages, we first defined the core microbiota changes associated with aging, followed by serum metabolome profiling at the fed state. We found age-dependent changes in microbial taxa associated with gut dysbiosis and pro-inflammatory signaling, and age-associated decrease in circulating levels of the gut microbe-dependent metabolites indole and indole-3-lactic acid. Signature pathways associated with aging include tryptophan metabolism, purine metabolism, amino acids, and nicotinate and nicotinamide metabolism. Together, these data suggest that gut dysbiosis and altered gut microbial metabolism may contribute to the metabolic dysfunction and inflammation in aging.

Age-associated changes in microbial biodiversity and composition
To examine the impact of age on mouse gut microbiota, the variable region 4 (V4) of bacterial 16S rRNA genes in fecal samples from 3M, 6M, 18M and 28M C57BL/6J male mice were amplified by PCR and sequenced using the Bio Scientific NEXTFlex TM platform. These mice were born and raised in house, and fecal samples were collected on the same day and stored at -80° C until processing. To investigate how aging affects the microbiota phylogenetic richness and diversity in each fecal sample, we analyzed the α-diversity ( Figure 1A). The Observed index is a richness-based measure which calculates the actual number of unique taxa observed in each sample, while Chao1 analysis predicts the number of taxa in a sample by extrapolating the number of rare organisms that may have been missed due to under sampling [20]. Shannon diversity index is a hybrid measure of the richness of a sample and the evenness of taxa in the sample. Interestingly, the evenness index of Shannon and the richness measures of Observed index and Chao1 index all showed significant age-dependent increases from 3M to 28M.
To compare the group differences in bacterial communities, the β-diversity of microbial composition was calculated by using principal coordinate analysis (PCoA). A clear age-dependent separation between the communities was observed at the first principal coordinate (PC1 axis), which explained 25.4% of the variance ( Figure 1B). Moreover, there is an agedependent shift in the 4 age groups, from left to right, on the PC1 axis; the 18M and 28M groups overlapped on the PC1 axis. Interestingly, 6-months old mice showed a greater individual variance as suggested by the larger green eclipse in Figure 1B, and larger box-whisker in Figure 1A. Analysis of microbial composition at the phylum level indicated that the fecal microbiota was dominated by 5 major phyla: Bacteroidetes, Firmicutes, Proteobacteria, candidatus Saccharibacteria and Deferribacteres ( Figure 1C). Correlation analysis showed that there were age-associated increases in Verrucomicrobia and Proteobacteria, and age-associated decreases in Bacteroidetes and Deferribacteres ( Figure 1D). Interestingly, the abundance of Firmicutes and Candidatus peaked at 6M, then age-dependently decreased in 18M and 28M groups.
Next, we compared microbial differences at the genus level ( Figure 2A). To identify specific bacterial taxa or genus-level phylotype that contributes to age, we applied Linear discriminant analysis Effect Size (LEfSE); 17 significant genera were identified as shown in Figure 2B. AGING There were age-dependent increases in Proteobacteria, Parabacteroides, Allistipes and Akkermansia. On the other hand, there were age-dependent decreases in Porphyromonadaceae, Mucispirillum, and Prevotellaceae.

Age-associated changes in serum metabolome
To identify metabolites whose abundances may be associated with aging, we profiled serum metabolites at the fed state from 3M, 12M, 18M and 28M male mice, using untargeted LC-MS metabolomics. We obtained abundance measurements for 9482 metabolite features, 1820 of which were annotated with Compound Discoverer (Supplementary Table 1).
We first analyzed the data using Principal Component Analysis (PCA), which is an unsupervised method to interpret the variance in a dataset (X) without referring to class labels (Y). PCA showed that 12 and 18M age groups clustered together, while there was clear difference between the 3M, 12-18M and 28M age groups in the first two principal components, PC1 (24.3%) and PC2 (16.6%). This suggested that there were age-associated changes in metabolites in serum ( Figure 3A). Next, for each of these metabolites, we computed the mean fold change (in log 10 space) and the statistical significance between the 4 age groups using one-way ANOVA followed by Tukey's post-hoc tests. This resulted in 510 metabolites with a log(p-value) above 1.8 (FDR-adjusted p value < 0.05, Supplementary Table 2). To explore the potential pathways disturbed in aging, the metabolites and p-values were analyzed using MetaboAnalyst 4.0 to identify differentially altered pathways. The MS Peaks to Pathways analysis predicts biological activity directly from peak list data, thereby bypassing metabolite identification. This analysis identified 6 pathways that are impacted by aging (Gamma values < 0.05, Figure 3B and Table 1). The top ranked pathways include tryptophan metabolism, purine metabolism, and amino acids metabolism (Table 1).
We further assessed the data for correlations of metabolites to age, using the pattern-hunter algorithm in MetaboAnalyst. This pattern search identified 372 significant features with positive or negative correlation coefficients between 1 and 0.6. We extracted the relative abundance of metabolites in tryptophan, purine, arginine and proline metabolism identified in the pathway analysis to show the direction of change. Of note, tryptophan and indole were positively identified by matching the spectra from pure standards and chromatographic retention time, while the other metabolites are putatively annotated by matching to either the spectral library (mzCloud) or accurate mass (ChemSpider). We found that guanosine and allantoin, two metabolites in the purine metabolism, showed significant age-associated increase in abundance ( Figure 4A, 4B). Hydroxyproline and The color and size of each circle corresponds to its p-value and enrichment factor, respectively. The enrichment factor of a pathway is calculated as the ratio between the number of significant pathway hits and the expected number of compound hits within the pathway [69]. The corresponding pathways are listed in Table 1.  AGING N-acetylornithine, metabolites in arginine and prolife metabolism, also showed age-associated increase in abundance ( Figure 4C, 4D). Erythrose phosphate, metabolite in the pentose phosphate pathway (P7 in Table 1), showed significant age-associated decrease in abundance ( Figure 4E). On the other hand, nicotinamide, a metabolite in the nicotinamide pathway (P20 in Table 1) and a significant feature identified by the correlation pattern hunter algorithm, showed significant age-associated decrease in abundance ( Figure 4F).

AGING
Regarding tryptophan metabolism, we found ageassociated decrease in serum levels of tryptophan, while L-kynurenine, a metabolite in the tryptophan catabolism pathway, showed age-associated increase in abundance ( Figure 5A, 5B). Notably, indole and indole-3-lactic acid, metabolites in the bacterial tryptophan catabolism pathway, showed significant age-associated decreases in serum abundance levels ( Figure 5C, 5D).

DISCUSSION
Here we explored, under controlled environmental conditions, changes in mouse gut microbiome and serum metabolome with age. We found that aging was associated with increased alpha diversity, a measure of the richness of species present. This is consistent with previous studies reporting greater alpha diversity with age in mouse gut microbiome [21][22][23][24]. On the other hand, human microbiome has higher complexity, due to the myriad of external factors influencing microbial populations between aging individuals [25,26]. Jeffery et. al [10] applied an iterative bi-clustering algorithm to microbiota composition data from the ELDERMET cohort to identify 4 subpopulations within the microbiota that are associated with biological phenotypes; moreover, loss of diversity in the core microbiota groups is associated with increased frailty, but not significantly associated with chronological aging [11]. Notably, the Clostridiales subpopulation is significantly associated with increased frailty, to a much greater extent than the loss of gut microbiota diversity. Our data showed ageassociated increase in Clostridiales from 3 to 18M, then decreased in the 28M group. A previous study has shown that centenarians have a microbiota that differs from those of older adults [9]. Our data also showed that Alloprevotella peaked at 6M, then age-dependently decreased in the 18M and 28M group. Interestingly, Alloprevotella has been associated with a decreased lifetime cardiovascular disease risk in the participants of the Bogalusa Heart Study [27].
At the phylum level, we observed that the Firmicutes/ Bacteroidetes (F/B) ratio is highest in the 6M group, and age-dependently decreased in the 18M and 28M groups. Previous study showed that the F/B ratio evolves during different life stages: the ratios were 0.4, 10.9 and 0.6 for infants, adults and elderly individuals, respectively [28]. Increased F/B ratio has been associated with excess energy harvest from food in obese patients [29,30]. The phylum Firmicutes includes Clostridium cluster XIVa, which contains many butyrate producing species [30]. We speculate that the decrease in Firmicutes in the 28M group may reflect a decreased capacity of the gut microflora at the extreme old age to harvest energy from food.
The age-dependent increase in Proteobacteria is consistent with a previous study showing significantly higher levels of Proteobacteria in the centenarian population [9]. Interestingly, Proteobacteria have been associated with altered mucus structure that is more penetrable by bacteria, leading to increased intestinal inflammation [31]. Furthermore, transferring aged microbiota to young germ-free (GF) mice lead to inflammation in the small intestine in the GF mice, increased leakage of inflammatory bacterial components into the circulation, and promoted increased T cell activation in the systemic compartment; these effects are associated with higher abundance of Proteobacteria in the aged microbiota after transfer [32].
In contrast to previous reports showing decreased abundance of Akkermansia in aging [23], here we found age-associated increase. Akkermansia are typically linked to health beneficial and antiinflammatory effects [33], such as improving glucose homeostasis in diet-induced obesity [34], and slowing progression of dextran sulfate sodium-induced colitis [35], although the mucin degradation properties of the members of this genus may also exacerbate certain types of infection [36]. Future studies are needed to further explore these differing findings. Taken together, our results indicate that older mice demonstrated several bacterial markers of gut dysbiosis and inflammation.
We found age-dependent decrease in serum nicotinamide, consistent with previous studies showing age-related decline in nicotinamide adenine dinucleotide (NAD + ) levels [37]. Emerging evidence suggests that NAD + is a critical signaling molecule and essential substrate for sirtuins, a class of enzymes that mediate several of the beneficial effects of calorie restriction in model organisms, including the maintenance of cardiovascular function [38]. Thus, the age-related decline in the cellular bioavailability of NAD + and related metabolites in animals and in humans may contribute to physiological aging by reducing sirtuin activity [37,[39][40][41]. NAD + can be synthesized de novo from the amino acid tryptophan, or from a salvage pathway for regenerating NAD + from other intracellular intermediates, which are primarily made available through dietary sources [42]. Importantly, supplementation of nicotinamide intermediates has been shown to improve cardiac health in older humans [43] and in aged mice [44]. Furthermore, a study of 12 adult men, ranging from 80-90 years of age, showed an anti-inflammatory effect of nicotinamide riboside supplement [45].
One of the significant pathways associated with aging identified in our metabolome data is purine metabolism. We found significant increases in allantoin and guanosine levels in older mice. Purines are endogenous organic molecules that are essential for all cells; they are structural constituents of nucleic acids, and acts as second messengers in intracellular signaling pathways [46]. Purines consist of the two-ring nitrogenous bases adenine, guanine, and their derivatives such as adenosine and guanosine. Interestingly, in a study profiling the metabolome of mouse brain at different stages of the life cycle (12, 18 and 24 months) and across different anatomical regions (hippocampus, frontal cortex and caudoputamen) revealed metabolic imbalance in the aging brain, characterized by NAD + decline, increased AMP/ATP, and purine/pyrimidine accumulation [47]. Thus, while the physiological significance of elevated levels of allantoin and guanosine in circulation is currently not clear, our metabolome data may reflect a metabolic imbalance in the aging mice.
Pathway analysis of our metabolome data showed tryptophan metabolism to be a major pathway altered in aging. Correlation analyses revealed that, while there is age-associated increase in kynurenine levels, there are significant age-associated decreases in tryptophan, indole and indole-3-lactic acid levels. The regulation of tryptophan concentration is critical for the maintenance of systemic homeostasis, since it is involved in several pathways including nutrient sensing, metabolic stress response, and immunity [48,49]. The majority (>95%) of dietary tryptophan is fed into the host kynurenine pathway, giving rise to a number of downstream AGING metabolites, while approximately 4-6% of tryptophan is available to the microbiota for biotransformation [48,50,51]. In addition to tryptophan being important as a precursor for the synthesis of the neurotransmitter serotonin, several catabolites along the kynurenine axis are neuroactive [52]. Furthermore, aging affects tryptophan metabolism, giving rise to higher kynurenine and lower tryptophan concentrations that are associated with neuropsychiatric symptoms [52]. On the other hand, indole is produced by a variety of both grampositive and gram-negative bacteria with tryptophanase activity, a bacteria-specific enzyme that catabolizes tryptophan to indole, pyruvate, and ammonia [50,51]. Indoles have been reported to regulate virulence in pathogenic bacteria [53,54], protect hosts from infection, and limit colitis induced by pathogens or chemical stressors [55][56][57]. The effects of indole and metabolites are thought to proceed largely via the aryl hydrocarbon receptor, and IL-22 signaling [58,59]. Interestingly, indoles have been shown to extend healthspan in worms, flies and mice via the aryl hydrocarbon receptor, without significantly altering lifespan [60]. In addition, indole enhances barrier functions of gut epithelium by inducing the expression of genes involved in tight junction, adherens junctions, actin cytoskeleton and mucin production [57,61]. Taken together, our metabolome data showing age-associated decrease in indole and its metabolite levels may reflect changes in gut microbial metabolism, and compromised gut barrier function in aging.
Pathway analysis showed several amino acid metabolism pathways to be significantly impacted by aging; however, correlation analyses identified limited metabolites in these pathways. Interestingly, we found age-associated increases in hydroxyproline. Hydroxyproline is one of the most abundant amino acids found in collagen, and increased levels of hydroxyproline in blood has been associated with connective tissue degradation [62]. At cellular level, hydroxyproline may also scavenge oxidants and regulate the redox state of cells [63]. Hence, metabolome analyses at tissue levels are required to complement and confirm the findings from serum metabolome. Furthermore, our serum metabolome data, at fed state, reflects systemic metabolites from both host and microbiome, and causal links between gut microbiome and serum metabolome cannot be inferred from our current data. Future interventional studies, such as fecal transplantation from young or old mice into microbiota-depleted old or young mice, are required to decipher the causal relationship between gut microbiome and metabolome changes in aging. For future follow-up study, we plan to transplant fecal pellets collected from 18 months-old donor mice, into 3 months-old recipient mice that have been treated with a cocktail of broad spectrum antibiotics as described [64], and assess fecal microbiome and metabolome before and after fecal transplantation.
In summary, here we showed that gut microbiome composition is altered by primary aging in mice, including changes in microbial taxa associated with gut dysbiosis and pro-inflammatory signaling. We showed that aging is associated with a decrease in circulating levels of the gut microbiota-dependent metabolites indole and indole-3-lactic acid, which may reflect compromised gut barrier function and pro-inflammatory state. Collectively, these data suggest that gut dysbiosis and altered gut microbial metabolism may contribute to the metabolic imbalance in the aging mice. Our findings support the development of gut microbiome-targeted agents and interventions to promote healthy aging. For feces collection, mice were singly housed for 4 days, transferred to fresh cages just before the beginning of dark phase, and fecal pellets produced overnight were collected the next morning before 10:00 AM. Fecal pellets were flash frozen and stored at −80° C. At end of collection the mice were returned to group housing with previous littermates.

MATERIALS AND METHODS
Serum samples were collected under fed condition in the morning between 8-10:00 AM. Animals were anaesthetized with isoflurane, blood collected via retroorbital bleeding, then mice were euthanized by cervical dislocation. Blood samples were centrifuged at 3000 g for 15 min at 4° C for serum collection, which were stored at -80° C until metabolome analysis.

16S rRNA sequencing analysis
Fecal microbiome analysis was performed as we previous described [65]. Briefly, fecal pellets were homogenized, and microbial DNA was extracted from the homogenate using the PowerSoil DNA extraction kit (MO BIO Laboratories, Carlsbad, CA). The V4 region of 16S rRNA was sequenced on a MiSeq platform (Illumina, AGING San Diego, CA) with a minimum of 800 and an average of 7500 sequences generated per sample. Sequence data were processed as previous described [66]. Taxonomic assignment was performed with RDP as the classifier, and HitDB and SILVA as the selected databases. Bacterial operation taxonomic units (OTUs) were counted for each sample to express the richness of bacterial species with an identity cutoff of 97%. Alpha diversity and beta diversity were calculated using the web-based tool MicrobiomeAnalyst [67]. Comparisons between groups were made at various taxonomic levels.