Aging and sarcopenia associate with specific interactions between gut microbes, serum biomarkers and host physiology in rats

Gut microbial diversity in aged rats

The gut microbiome is a complex ecosystem that reflects the contribution of multiple environmental (such as diet, drugs and pathogens) and host-related (immunity) factors. We first sought to understand the composition of the microbiota across the ages to determine key bacteria associated with the aging phenotype.

We began by conducting a 16S rRNA-based OTU survey of the ecological state of the fecal microbiomes (S1.xlsx, ). Our observations of the microbial ecology (using NMDS plots of the community structures Beta diversity) suggest that the microbiome is affected by aging. The 18M communities appeared unique to the 8M and 24M communities that could not be distinguished from one another (Figure 1B). The composition (alpha diversity) of the microbiome at the three ages demonstrated unique OTUs at each age, but generally did not distinguish one age-related community from another (Figure 1C).

The OTUs identified in the 16S survey were organised into categories based on the observed abundance across the ages (Figure 1D) (further described in Supplementary Materials). We observed 7 of the possible 18 categories (theoretically possible) that would demonstrate a significant change. In particular, categories 9 and 10 demonstrated the same pattern as seen in the NMDS plots identifying potential OTUs that drive the distinction of the 18M samples.

We also used indicator analysis [21] to investigate the contributions of different OTUs to the differences between the age groups. Thirty-nine unique indicator OTUs with P values < 0.001 and statistically different levels upon comparison of two age groups were identified for ages 8M, 18M and 24M (Supplementary Materials, S1.xlsx–IndicatorAnalysis Sheet). Selected indicator OTUs (indicator values > 65), their average relative abundance, their categorization and the age group they indicate are depicted in Figure 2A. OTUs indicating 24M were the most abundant. The only indicator OTU for 18M, with an indicator value (69.32), was OTU0560 (Thermotalea) which was also significantly upregulated at 18M vs 8M. Focussing on the category 9 and 10 OTUs, OTU 0499 (Lachnospiraceae insertae sedis) and OTU 0809 (Lachnobacterium) were both identified as indicators of the 24M group, consistent with the observed decrease at 18M (Figure 1D). Based on Spearman correlations, these two OTUs were also some of the most connected OTUs based on their positive correlation with other OTUs across the ages (Figure 2B, Supplementary Figure 1). OTU 0914 (Lactonifactor), the only OTU found in category 9, was also well connected but negatively correlated with a number of OTUs including OTUs 0499 and 0809. Based on the composition of the microbes across the ages, these three OTUs appear to be markers and keystones of the 18M microbial ecology.

Association of bacteria genera and rat sarcopenia

We next examined the association between the microbial ecology and the observed physiological changes in the rats. Aged rats gradually lose muscle mass and function (i.e. sarcopenia) through multi-factorial mechanisms involving mitochondrial and neuromuscular dysfunction [20,22]. Having characterized the genetic composition of the microbiome, we investigated whether the age-related changes observed in the ecology of the gut microbiota could account for some of the physiological parameters associated with aging and sarcopenia. We previously demonstrated in the same rats that they demonstrated selective age-related sarcopenia (Table 1, complete data provided in Supplementary Materials S3.xlsx). Using age window comparisons of the physiological measurements we observed that body weight and fat mass (%) increased with age (statistically different in 8M-24M and 18M-24M comparisons) while gastrocnemius muscle mass (statistically different in 8M-18M, 8M-24M, 18M-24M comparisons), lean mass (%) and sciatic response amplitude (statistically different in 8M-24M and 18M-24M comparisons) decreased with age (Figure 3A). To investigate the potential associations between the microbiome and the host, we analysed the correlations (details in materials and methods) between the physiological measurements and the OTUs significantly correlated with age (Figure 3B). We focussed on the indicator OTUs for the 24M samples consistent with the development of the sarcopenia phenotype. In general, these OTUs (24M) correlated with increased fat mass, decreased lean and gastrocnemius muscle mass. Several of these OTUs also correlated with vitamin B12 and folate levels. The group 10 OTU (OTU0499) positively correlated with both folate levels in the plasma and heart mass. Although folate levels did not significantly change across the ages, its role in the context of cardiovascular disease has been studied extensively. Vitamin B12, as well as folate, have been similarly studied in the context of lipid physiology. Considering the association of these bacteria with these vitamins as well as the physiological impact of aging, we next sought to determine the underlying mechanistic links between the microbiota and host physiological responses to aging.

Metagenomic functional content analysis

We next tried to identify potential molecular mechanisms implicated in the aging phenotype by applying PICRUSt [23] to identify the Metagenomic Functional Content (MFC) for different groups using the 16S rRNA data (Figure 4A). The tabulated entries for the entire list of statistically different MFC’s are available in the Supplementary Materials (S2.xlsx).

To help focus the analysis, the MFCs were put into the same 18 categories previously used for the OTUs in Figure 1 (Figure 4B, 4C and Figure 1D). All the statistically different MFC’s could be classified into 9 categories, with the majority (82% or 307 KO IDs, categorised into categories 1, 2, 4 and 7). Most of the physiological responses (Figure 3A) could be categorised into Category 7 or 8 (significantly higher/lower at the 24M vs 8/18M). Since there were no MFCs observed in category 8, we focussed our attention on the Category 7 observations. These can be summarised in three main categories: secretion systems (often associated with pathogenic mechanisms), ABC transporters (often associated with uptake of essential nutrients) and dietary metabolism (protein, carbohydrate and lipid metabolism) suggesting that the host/microbiome interactions contributing to the aging phenotype involves immune and dietary components.

Microbial NMDS analysis and the community analysis summarised in Figures 1 and 2 indicated that the 18M microbiota was distinct from the 8M and 24M communities. While the composition of the 8M and 24M communities could not be distinguished based on 16S rRNA survey, their functional capacity was distinct as demonstrated in Figure 4. One interpretation of this is that there was a transition in the community membership that led to a functional alteration of the microbiota. The 18M community MFC therefore may give an indication of the molecular mechanisms that underlie this transition. Figure 1 identified Category 9 and 10 OTUs as associated with the 18M samples. While no category 10 MFCs were observed, there were an abundance of category 9 MFCs. We further summarized the MFC’s into their corresponding KEGG pathways to hypothesize the molecular mechanisms that may underlie the transition to the aged phenotype (Figure 4D). Category 9 pathways were primarily related to diet including metabolism of carbohydrate, protein, lipids and vitamin biosynthesis (Supplementary Materials, S2.xlsx) suggesting that the microbiome contributes in part to aging through their established role in digestion. We observed a general increase in MFC’s assigned to Amino Acid (25/29) and Carbohydrate (12/15) metabolism with a concurrent general decrease (6/9) in those assigned to Lipid Metabolism suggesting a shift in the potential to digest, process, and synthesize these dietary components. One of the increased MFCs was cholesterol oxidase (steroid biosynthesis) indicating a predicted change towards cholesterol versus fatty acid synthesis.

Furthermore, consistent with the analysis of MFCs related to Metabolism of Cofactors and Vitamins (Figure 4D), statistically reduced levels of serum Vitamin B12 were observed. We also explored the correlations between specific MFC’s and the physiological measurements (Table 1, Supplemental Figure 2). None of the Category 9 MFCs associated with physiological parameters. However, there were several associations of other diet-related MFCs on many physiological factors including gastrocnemius muscle mass and sciatic response amplitude (Supplementary Figure 2, S2.xlsx; MFC_Metadata_Correlation Sheet). Fat mass was positively correlated to Cobalamin biosynthetic protein CobC (K02225), carnitinyl-CoA dehydratase (K08299) and allantoin permease (K10975). Furthermore, there was a strong negative correlation between lean mass and allantoin permease (K10975). Allantoin has previously been demonstrated to increase lifespan when administered to the nematode worm Caenorhabditis elegans [24,25]. The negative association could relate to microbial scavenging of this purine metabolism biomarker of oxidative stress [26].

We considered the following key observations: 1) analysis of the aging microbiome identified a diet-related ecology specifically associated with the transition at 18M to the sarcopenic phenotype, 2) the OTUs characterized at the 18M correlated with observed Vitamin B12 and folate levels, and 3) the MFC analysis identified candidate molecular mechanisms including pathogenesis and dietary metabolic function associated with the sarcopenic phenotype. Taking these observations together, they indicate an important role for microbial-derived dietary metabolic pathways in aging and sarcopenia including carbohydrate, lipid and vitamin metabolism that could alter the metabolic status of the host and contribute to the physiological state in aging.

Serum proteomics of aged rats

As shown by the analysis above, microbial ecology is a complex process that involves the integration of many factors leading to an outcome in the host. We performed proteomic analysis of the serum to identify the potential host response to the changes in the microbiota and their association with the physiological state. Serum proteins for the 3 groups of rats (n=9 at 8M, n=10 at 18M and n=8 at 24M) were analysed using aptamer-based detection [27]. Similar to the OTUs and MFC analysis, we performed pairwise comparisons to identify proteins that were significantly altered between two age groups (8M-18M, 8M-24M and 18M-24M) and put them into the categories previously described. (Figure 5A, Supplementary Materials S4.xlsx).

Since we were looking specifically at the host response, we focussed on categories that demonstrated a significant association with the host physiological measurements shown in Figure 3. In general, fat mass, lean mass and sciatic response amplitude demonstrated a 'Category 7 or 8' pattern. There were only two serum proteins detected demonstrating this pattern; Amnionless (AMN) and Insulin-like growth factor I receptor (IGF1R). IGF1R did not show any significant correlation with the physiological measurements. However, AMN correlated with most of the physiological measurements with the exception of folate (Figure 5B). AMN plays a key role in Cobalamin (Cbl, vitamin B12) transport as well as for lipid metabolism, HDL-mediated lipid transport, lipoprotein metabolism and lipid digestion. We also detected an association of AMN with cholesterol oxidase and cobalamin biosynthetic protein CobC in the MFCs (Supplementary Figure 4).

Subsequently, we used Reactome [28,29] (www.Reactome.org) to identify participation of the differentially observed proteins in different pathways in rats (S4.xlsx). Overall, protein levels which consistently increased with age (Cat 1), i.e. Neutral ceramidase (ASAH2), Nuclear receptor subfamily 1 group D member 1 (NR1D1), Parathyroid hormone (PTH) and X-linked ectodysplasin-A2 receptor (EDA2R) were implicated in glycosphingolipid metabolism, sphingolipid metabolism, GPCR signalling, TNFR2 non-canonical NF-kB signalling and the immune system.

Summarizing the genetic analysis of the microbiome and the serum protein response of the host there was a common theme associated with the microbiome and aging. Alterations in dietary metabolism, possibly via their influence on Vitamin B12 and folate, seemed to be involved in the aging phenotype.

Serum lipidomics of aged rats

Vitamin B12 and folate have defined roles in lipid metabolism. Vitamin B12, adenosylcobalamin, act as a cofactor on methylmalonyl-CoA mutase to convert methylmalonyl-CoA (MM-CoA) to succinyl-CoA. When MM-CoA mutase is blocked in the absence of Vitamin B12, MM-CoA accumulates and inhibits the rate-limiting enzyme of fatty acid oxidation (CPT1 - carnitine palmitoyl transferase) thus causing lipogenesis. Low levels of Vitamin B12 are also associated with shifts toward cholesterol synthesis and hepatic steatosis in mice [30]. Folate deficiency decreases flux through phosphatidylethanolamine N-methyltransferase (PEMT), an enzyme that synthesizes phosphatidylcholine (PC) via the methylation of phosphatidylethanolamine (PE) [31]. The alterations in these lipid-modifying vitamins by the microbiota therefore could underlie the association with aging. We analysed 122 lipid species using direct infusion mass spectrometry as described previously [32] (materials and methods, data tabulated in Supplementary Materials, S5.xlsx). Analysing the measured lipid species across the samples (n=10 at 8M, n=10 at 18M and n=8 at 24M), using NMDS analysis, there was no clear separation between 8M and 18M (orange and green circles in Figure 6A). One to one comparison for statistical differences between two groups further confirmed this (no statistically different levels for lipid species measured between 8M and 18M or between 8M and 24M). However, the 18M and 24M groups clearly separated in the lipid space (green and blue circles, Figure 6A). 5 lipid species were statistically significantly different between the two groups (18M-24M) (Figure 6B). Levels of two Lysophosphatidylcholines (LPC 20:5 and LPC 20:3) decreased in old age, while three lipids, Lysophosphatidylinositol (LPI 16:0), Phosphatidylcholine (PC 37:4) and sterol ester (SE 20:4) increased in old age. The changes in phosphatidylcholine/lysophosphatidylcholine are consistent with the observed alterations in Vitamin B12 and folate status. Subsequently, we looked at the correlations of the lipids with physiological measurements (Figure 6C). Gastrocnemius muscle mass (decreased consistently with age) and sciatic response amplitude (decreased in 8M-24M and 18M-24M) significantly correlated to the statistically different lipids (Figure 6C). Gastrocnemius muscle mass positively correlated with LPC 20:3 (decreased in 18m-24m). Sciatic response amplitude positively correlated with LPC 20:5 and LPC 20:3 (both decreased in 18M-24M) and negatively correlated with PC 37:4 and SE 20:4 (both increased in 18M-24M) (Figure 6C). Lipid metabolism, therefore, may play a particular role in the development of the sarcopenic phenotype observed in these rats.

Given the possible role of microbes in altering lipid and inflammatory status, we looked at the correlations of the lipid species with the microbial members (OTUs – Figure 6D, MFC’s – Supplementary Figure 5) and serum protein levels (Figure 6E). In total there were 14 OTUs that correlated with the age-related lipid species of which OTU 0229 (Lachnospiraceae insertae sedis) was the only category 10 OTU identified. In particular, it negatively correlated with LPC 20:5 (Figure 6D). The other OTUs identified were, similar to the lipids identified, Category 15 and 18 OTUs. Of these, OTUs 0126, 0954, 0979 and 0599 were also identified in the indicator analysis in Figure 2. Focussing on these, OTU 0954 also negatively correlated with Vitamin B12 levels (Figure 3). The category 18 OTU 0269 was also identified and positively correlated with B12 (Figure 3). LPC 20:5, LPC 20:3 and PC 37:4 were among the most correlated lipid species consistent with a potential role for microbial metabolism of Vitamin B12 and folate. In addition, the levels of PC 37:4 and SE 20:4 positively correlated with the MFC Cobalamin Biosynthesis Protein CobW (Supplementary Figure 5). Reduced LPC 20:5 were reported in obese patients as opposed to control subjects [33]. LPC 20:5 and LPC 20:3 were earlier reported to be negatively correlated to BMI in Obesity [34]. However, the association of these lipids with a sarcopenic phenotype suggests an important role for the microbiome in regulating lipid metabolism, potentially contributing to sarcopenia as well as obesity.

We next determined potential host responses to the altered lipid physiology. We identified 9 proteins, which are significantly correlated with the lipid levels. Proteins, which are the most correlated, included IL17RD, ERAP1, TNFSF11 and ATP5B. While the lipids, which were the most correlated, included LPC 20:3, LPC 20:5, PC 37:4 and SE 20:4. Reduction of NOTCH1 was negatively correlated to SE 20:4 and positively correlated to LPC 20:3. Potential anti-inflammatory lipids LPC 20:5 and LPC 20:3 were negatively correlated to increased levels of IL17RD, CBX5, ERAP1 and TNFSF11 and positively correlated to IGF1R. Alterations of the EPH-ephrin signalling (via EPHA10) was positively correlated to LPC 20:5. Proteins altering the immune status (i.e. TNFSF11, ERAP1 and IL17RD) were negatively correlated to decreased levels of LPC 20:5 and LPC 20:3 and positively correlated to increased levels of PC 37:4 and SE 20:4. The negative correlation of P4HB to PC 37:4 levels indicates potential effects in VLDL biosynthesis and chylomicron-mediated lipid transport.

Our results demonstrate the complexity of host microbe interactions in the context of aging and highlight the potential interaction of nutrition, the microbiome, and host metabolism in sarcopenia. Our thorough survey of microbial, proteomic and lipidomic biomarkers that underlie the aging process suggest that regulation of dietary metabolism by the microbiota may underlie some of the consequences of aging and identify new nutritional therapeutic avenues for further exploration in age-related conditions such as sarcopenia.

Animal study protocol

Male Wistar rats aged 8 months to 24 months were obtained from Janvier Labs (Le Genest-Saint-Isle, France). The rats (littermates) labelled by age were grouped by date of birth within one month, and further grouping was then based on the muscle phenotype in hind limbs under the following categories: adult (8 months of age), early-sarcopenic (18 months of age), and sarcopenic (24 months of age). Upon arrival, all animals were housed by two in standard type 4 cages with ad libitum access to food and water on a 12 hour light/dark cycle at a temperature between 20-24°C and a relative humidity between 50-60%. The experimental procedures on animals were in agreement with Swiss and EU ethical guidelines and approved by the local animal experimentation committee of the Canton of Vaud (license number VD2630). Body composition (lean mass and fat mass) was measured non-invasively in awake animals using quantitative NMR (1H NMR relaxometry device, echoMRI). Electromyography (EMG) measurements were performed on rats under isoflurane anaesthesia. Briefly the left limbs were shaved and recording needle electrodes (twisted pairs, wire 150cm, needle 0.4 x 13mm, Neurolite, Switzerland) were sequentially placed into the gastrocnemius, and triceps brachii muscles. Supra-maximal electric stimulation was achieved via stimulating needle electrodes sequentially placed around the sciatic nerve and the radial nerve, and the resulting compound muscle action potential (CMAP) was recorded using the Keypoint software (Neurolite, Switzerland). The animals were sacrificed by exsanguination under isoflurane anaesthesia and blood was collected from the portal vein and resulting serum was used for proteomic analysis. Skeletal muscles and heart were dissected free of fat, weighed and snap frozen in liquid nitrogen.

Fecal pellet collection, DNA extraction

Fecal pellets from each rat were collected prior to sacrifice in sterile 1.5 ml Eppendorf tubes that were then stored at -80C until DNA was extracted from approximately 50 mg using the MoBio PowerMag DNA extraction kit according to the manufacturers specifications. The DNA was stored at -20C until library preparation.

Amplicon library preparation and sequencing

An Amplicon library of partial 16s rRNA genes was produced according the methods outlined in Caparaso et al. [50]. Briefly, we used primers targeting the V4-6 region of the 16S ribosomal genes at positions 515-806 (E. coli numbering) composed of the Illumina adapters, pad, barcode (forward primer only) and gene specific nucleotides. The resulting PCR products after 23 cycles of amplification from 3 reactions per samples were pooled and quantified using a Caliper LabChip GX. The samples were then pooled in equimolar ratio and cleaned using Agencourt AMPure XP. The cleaned amplicons were then pooled with PhiX DNA spike as per manufacturer recommendation and sequenced using a MiSeq using chemistry V2.4.

16S data processing

Sequences were demultiplexed from fastq files according to the specific barcode for each sample introduced at the PCR step. Quality control was performed with MOTHUR [51] using the default parameters in the MiSeq SOP [52] accessed on Feb 2016. The resulting sequences were then clustered into OTUs (with 0.03 cutoff). These were also converted into biom files with closed OTU picking with greengenes taxonomy (greengenes version 13.5) for use in PICRUSt [53]. The resultant OTUs were also used for plotting NMDS with JClass and ThetaYC calculators, additionally an indicator analysis was also performed. Statistical differences in OTUs between age groups were assessed using the Wilcoxon Rank Sum test and after FDR correction, Q-values < 0.05 were considered significant.

Metagenomic functional content analysis

Phylogenetic investigation of communities by reconstruction of unobserved states (PICRUSt) [53] was used to predict the metagenomic functional content (MFC) of the different samples. Statistical differences between age groups were estimated by using the Kolmogorov-Smirnov test on the rank of the z-scores of the relative abundances of the MFC’s and P-values were considered significant using a FDR of 1%.

Serum analysis

Vitamin B12 and folate levels were measured using the competitive binding assay from Beckman (Beckman Coulter, Nyon, Switzerland). For protein measurement in serum, samples were analyzed using DNA-aptamer-based recognition on the SOMAscan platform (SomaLogic, Boulder, CO, USA), as described [27]. Median normalized relative fluorescence units (RFUs) were log₂-transformed before applying principal component analysis and linear models. Statistical analyses were performed in R 3.1.3 (R Foundation for Statistical Computing).

Lipidomics analysis

Lipid extraction and analysis were performed as reported previously [32]. In summary, 100 ul of (diluted serum 1 in 50 ammonium bicarbonate buffer) was mixed with 80 ul of ammonium bicarbonate solution and then 810 ul of methyl tert-butyl ether/methanol (7:2, v/v) solution was added. Internal standard mixture was pre-mixed with the organic solvents mixture. The internal standard mixture contained: PC 17:0/17:0, Chol D6, DAG 17:0/17:0, TAG 17:0/17:0/17:0, Cer 18:1;2/17:0, SM 18:1;2/12:0, LPC 12:0, LPE 17:1, PE 17:0/17:0, SE 20:0, PI 16:0/16:0. Solution was mixed at 700 rpm, 15 min at 4 °C using a ThermoMixer C and then centrifuged at 3000 RCF for 5 min. 100 μl of the organic phase was transferred to a 96-well plate, and dried in a speed vacuum concentrator. Lipid extract was reconstituted in 40 μl of 7.5 mM ammonium acetate in chloroform/methanol/propanol (1:2:4, V/V/V). All liquid handling steps were performed using Hamilton STAR robotic platform.

For MS data acquisition, samples were analyzed by direct infusion in a QExactive mass spectrometer (Thermo Fisher Scientific) equipped with a TriVersa NanoMate ion source (Advion Biosciences). Samples were acquired in both polarity modes in a single acquisition at Rm/z=200 = 140000. All data was analyzed with in-house developed lipid identification software based on LipidXplorer. Data post-processing and normalization were performed on an in-house developed R based package. Statistical significance between age groups was assessed using the Kolmogorov-Smirnov 2 sample test and after FDR correction Q-Values were considered significant if < 0.05.

Correlation analysis

In general (unless otherwise stated), correlations (Spearman’s Rho) were assessed between different measurements and checked for statistical significance. It was considered as statistical significant after FDR correction with Q values < 0.05.

Ethics approval and consent to participate

Animal Study: The experimental procedures on animals were in agreement with Swiss and EU ethical guidelines and approved by the local animal experimentation committee of the Canton of Vaud (license number VD2630).