Gut bacterial nutrient preferences quantified in vivo

SUMMARY

Great progress has been made in understanding gut microbiome’s products and their effects on health and disease. Less attention, however, has been given to the inputs that gut bacteria consume. Here we quantitatively examine inputs and outputs of the mouse gut microbiome, using isotope tracing. The main input to microbial carbohydrate fermentation is dietary fiber, and to branched-chain fatty acids and aromatic metabolites is dietary protein. In addition, circulating host lactate, 3-hydroxybutyrate and urea (but not glucose or amino acids) feed the gut microbiome. To determine nutrient preferences across bacteria, we traced into genus-specific bacterial protein sequences. We find systematic differences in nutrient use: Most genera in the phylum Firmicutes prefer dietary protein, Bacteroides dietary fiber, and Akkermansia circulating host lactate. Such preferences correlate with microbiome composition changes in response to dietary modifications. Thus, diet shapes the microbiome by promoting the growth of bacteria that preferentially use the ingested nutrients.

Graphical Absctract

In Brief:

Isotope tracing into bacterial-specific protein sequences allows for a determination of nutrient preferences across gut microbes in vivo, and reveals how diet alters microbiome composition.

INTRODUCTION

The gut microbiome possesses an enormous diversity of enzymes, exceeding the number in mammals’ genomes by more than 100-fold (Qin et al., 2010). This enzymatic capacity enables the processing of incoming dietary nutrients into a broad spectrum of microbial metabolites. Some of these reach the host circulation at substantial concentrations (Lai et al., 2021; Quinn et al., 2020). Microbial metabolites can play important roles in host pathophysiology. For example, short-chain fatty acids (SCFAs; acetate, propionate, butyrate) (Dalile et al., 2019; Koh et al., 2016), trimethylamine N-oxide (Tang et al., 2013), secondary bile acids (Arab et al., 2017; Funabashi et al., 2020), indole-3-propionate (Wikoff et al., 2009), and imidazole propionate (Koh et al., 2018) affect immune maturation (Campbell et al., 2020; Hang et al., 2019), insulin sensitivity (Koh et al., 2018), cancer growth (Garrett, 2015; Yoshimoto et al., 2013), and cardiovascular disease (Nemet et al., 2020; Wang et al., 2011).

Both to replicate themselves and to release metabolic products, gut bacteria require nutrient inputs. These come in forms including ingested food, host-synthesized gut mucus (Desai et al., 2016; Sicard et al., 2017), and host circulating metabolites (Scheiman et al., 2019). The availability of dietary nutrients to gut microbiota depends on the extent of host absorption: nutrients that are absorbed in the small intestine, like starch, are not available to the colonic microbiome. In contrast, nutrients that are poorly digested in the upper gastrointestinal tract, like fiber, can be key microbiome feedstocks (Lund et al., 2021; Wong and Jenkins, 2007).

Isotope tracing enables quantitative measurement of the inputs to metabolites and biomass. Studies employing radioactive tracers defined the basics of mammalian metabolism (Wolfe, 1984). Recent work has increasingly relied on stable isotope tracers coupled to mass spectrometry detection, which enables the measurement of labeling in specific downstream products (Fernández-García et al., 2020; McCabe and Previs, 2004). This approach has revealed fundamental features of host metabolism, such as circulating lactate being a major TCA fuel (Faubert et al., 2017; Hui et al., 2017). In addition, it has provided important insights into host-microbiome metabolic interplay. For example, it revealed that dietary fructose is processed by the microbiome into acetate, which fuels hepatic lipogenesis (Jang et al., 2018; Zhao et al., 2020).

In principle, stable isotope tracing coupled to mass spectrometry can also be applied to determine the metabolic inputs to specific microbes, based on measuring labeling in bacteria-specific peptide sequences (Berry et al., 2015; Holmes et al., 2017; Oberbach et al., 2017; Reese et al., 2018; Zhang et al., 2016a, 2016b). By infusing nitrogen-labeled threonine to label host mucus, investigators were able to compare the contribution of dietary versus mucus protein to the gut microbiome and observed a shift towards more mucus contribution in mice fed a low-protein diet (Holmes et al., 2017).

Here, we perform large-scale, quantitative assessment of the metabolic inputs to the gut microbiome and its products. We examine the contributions from dietary starch, fiber, and protein, and from host mucus. We also examine most major circulating host nutrients, finding that lactate, 3-hydroxybutyrate, and urea stand out for passing from the host to the gut microbiome. Based on the measurement of bacteria-specific peptide sequences, we assess the nutrient preferences of different bacterial genera and show that these preferences align with microbiome composition changes in response to altered diet.

RESULTS

Microbiome consumes less digestible dietary components

A major mechanism by which the microbiome may impact host physiology is via secreted metabolic products. We measured, in the portal, systemic circulation and the cecal contents, the absolute concentrations of more than 50 metabolites characterized in the literature as microbiome-derived (Campbell et al., 2020; De Vadder et al., 2014; Han et al., 2021; Hang et al., 2019; Koh et al., 2018; Mager et al., 2020; Ridlon et al., 2014; Wikoff et al., 2009) (Figure S1A-B, Table 1, S1). Most were elevated in the portal circulation relative to systemic blood, and all but two (inosine and N-acetyl-tryptophan, which are apparently mainly host derived) were depleted by antibiotics treatment.

Table 1. Absolute concentrations and sources of microbiota-associated metabolites. See also^{Table S1,Figure S1,S4}.

Data are from ad lib fed state (ZT0); for ad lib fasted state (ZT12), see Supplementary Table S1. Absolute concentration is mean, N = 5 mice. Portal/systemic = fold-change in concentration between the portal vein and tail vein (median, N = 5 mice). Abx/Conv refers to fold-change in portal blood concentration between mice treated with antibiotics cocktail versus not (median, N = 5 mice/group). Source bar indicates the relative contribution to the indicated metabolite from dietary inulin, algal protein and circulating lactate (based on isotope tracing). Percentages indicate quantitative relative contributions from those nutrients (median, N = 4). Numbers typically add up to less than 100%, as other sources (e.g., mucins) contribute.

The dominant excreted products on a molar basis (0.4 – 2 mM in the portal blood) are SCFAs. Other relatively abundant microbiome products (10 – 30 uM) are aromatic amino acid fermentation products (phenol, indoxyl sulfate, and 3-phenylpropionate) and branched-chain fatty acids (valerate, isovalerate, 4-methylvalerate, isobutyrate, 2-methylbutyrate). Primary bile acids, while present in the portal circulation at up to ~ 10 uM concentration, are produced by the host and accordingly were not included in Table 1. Secondary bile acids, which are produced from primary bile acids by the microbiome, were lower in absolute concentration, the most abundant being tauroursodeoxycholic acid (3 uM in portal vein).

To probe the dietary inputs to gut microbial products, we began by feeding mice via oral gavage, starch (readily digestible glucose polymer) and inulin (slowly digestible fructose polymer, i.e., soluble fiber) (Figure S1C). Following ¹³C-starch gavage, labeled glucose, lactate, and alanine quickly appeared in the portal circulation and accounted for most starch carbons (~75%) (Figure S1D–F, S1G) (Jang et al., 2018). In contrast, after ¹³C-inulin gavage, substantial labeled fructose, glucose, lactate, and alanine were not observed, and instead labeled portal metabolites slowly appeared in the form of SCFAs, with ~ 40% of inulin carbons becoming SCFAs and the remainder being undigested and excreted in the feces (Figure S1E–F, S1H–I). Moreover, dietary inulin, but not starch, extensively labeled glycolytic and TCA intermediates and amino acids in the cecal content (Figure S1G).

We next carried out similar experiments, comparing the gavage of a free amino acid mixture to algal protein, both uniformly ¹³C-labeled (Figure S1C). The free amino acids resulted in the rapid appearance of labeled amino acids in portal circulation (Figure S1J–L), while the algal protein substantially labeled amino acids within the cecal contents (Figure S1M). Moreover, the algal protein copiously labeled microbiome-derived portal vein metabolites: SCFAs, branched-chain fatty acids, and aromatics (indole, indole-3-propionate, 3-phenylpropionate) (Figure S1K–L). Thus, poorly digestible carbohydrates and protein feed the microbiome directly, and the host indirectly via microbiome-derived products.

Few circulating metabolites reach the microbiome

Next, we examined the possibility that nutrients in host circulation feed the gut microbiota. We infused deuterated water and eighteen major circulating nutrients (¹³C-labeled) into the systemic circulation of pre-catheterized mice (Figure 1A). The infusion rates were selected to achieve modest but readily measurable labeling without substantially perturbing circulating concentrations. Circulating labeling reached a steady state by 2.5 h, at which time we collected serum and feces to quantitate the carbon contributions of each circulating nutrient to the corresponding fecal metabolites. Upon intravenous infusion of ¹³C-lactate, fecal lactate labeled rapidly (Figure 1B). Most infused circulating nutrients, however, did not penetrate the feces (Figure 1C–D). Indeed, while water fully exchanged with the feces, among abundant circulating carbon carriers, only lactate and 3-hydroxybutyrate penetrated. Glucose, amino acids, TCA intermediates and fatty acids did not. Both lactate and 3-hydroxybutyrate are substrates of monocarboxylate transporters (MCTs), which are highly expressed in the colonic epithelium (Halestrap and Price, 1999, p. 1). Pharmacological MCT inhibition prevented lactate from penetrating the feces (Figure 1E). Thus, in contrast to most host circulating metabolites, which do not reach the colonic microbiome, monocarboxylic transporters render circulating lactate and 3-hydroxybutyrate accessible to gut microbes.

Figure 1. Circulating lactate and 3-hydroxybutyrate feed the gut microbiome. See also^{Figure S2}.

(A) Schematic of intravenous infusion of isotope-labeled nutrients to identify circulating metabolites that feed gut microbiome.

(B) Circulating lactate rapidly enters the feces. Mice were infused with ¹³C-lactate and serum and fresh feces enrichment were compared. Mean±s.e., N = 3.

(C) Circulating citrate does not enter the feces. As in (B), for ¹³C-citrate. Mean±s.e., N = 3.

(D) Passage of circulating ¹³C-labeled nutrients into the feces. Mice were infused with labeled nutrients for 2.5 h, and labeling fraction in feces was normalized to labeling fraction in serum. Mean±s.e., N = 3 except for lactate (N = 8) and 3-hydroxybutyrate (N = 7).

(E) Pharmacological inhibition of MCT1 transporter decreases the passage of circulating lactate to feces. Mice were injected i.p. with saline or 100 mg/kg AZD3965, and fresh feces lactate enrichment measured. Mean±s.e. N = 6 for saline and N = 5 for AZD3965. ***P<0.001 by two-sided Student’s t-test.

(F) Passage of circulating ¹⁵N-labeled nutrients into the feces. As in (D), for ¹⁵N-lableing. Mean±s.e. N = 3 except for urea (N = 4) and ammonia (N = 5).

Circulating urea is a microbiome nitrogen source

In addition to carbon, nitrogen is a fundamental constituent of all living cells. To assess nitrogen sources of the gut microbiome, we infused twelve abundant circulating nutrients in ¹⁵N-labeled form. Nitrogen from circulating urea and ammonia, but not amino acids, penetrates the feces and contributes to microbiome amino acids and ammonia (Figure 1F, S2A–B). Urea usage by the microbiome involves its re-conversion to ammonia via the enzyme urease, which is expressed by a subset of gut microbes (Mora and Arioli, 2014; Ni et al., 2017), and gnotobiotic mice colonized intentionally with only urease-negative bacteria showed no labeling from urea (Figure S2C-D).

Urea, which is made from ammonia in the liver, was a quantitatively greater source of microbiome nitrogen than ammonia. Moreover, urea but not ammonia was more abundant in the host circulation than cecal lumen, consistent with only urea being able to passively flow into the gut lumen (Figure S2E–F). We hypothesized that circulating host ammonia might be feeding the microbiome mainly indirectly, after being converted by the host liver into circulating urea (Figure S2G) (Bartman et al., 2021). This indirect contribution was calculated by multiplying circulating urea’s contribution to fecal amino acids (L_AAs←urea) by the circulating urea fraction that comes from circulating ammonia ($L_{\mathit{\text{urea}}\leftarrow N{H}_3}=33\%$). It fully explained the observed microbiome labeling from circulating ammonia (Figure S2H). Further supporting the indirect pathway, antibiotics treatment blocked both circulating urea and ammonia from becoming cecal ammonia (Figure S2I–J), which makes sense if flux of ammonia into the cecal contents goes through host urea and microbial urease (Figure S2K).

Microbiota synthesize amino acids from fiber and urea

To determine quantitatively the sources of microbiome metabolites, we measured their labeling after ad libitum feeding of isotopically enriched food. To this end, we fed mice standard chow with a portion of the fiber, fat, or protein ¹³C-labeled, with cecal labeling reaching steady-state within 12 h (Figure S3A). To account for circulating nutrient inputs, we also infused ¹³C-lactate or 3-hydroxybutyrate (Figure 2A). These studies identified a majority of the carbon feeding into most microbiome central metabolites, with glycolytic and pentose phosphate metabolites labeling almost exclusively coming from dietary fiber (inulin), while pyruvate and TCA metabolites are also labeled from dietary protein and circulating lactate (Figure 2B and S4A).

Figure 2. Quantitative analysis of dietary and circulating nutrient contributions to gut microbiome. See also^{Figure S3,S4}.

(A) Experimental design. Mice were fed chow containing ¹³C-protein, ¹³C-inulin, ¹³C-fatty acids, or ¹⁵N-protein for 24 h. Alternatively, mice were intravenously infused with ¹³C-lactate, ¹³C-3-hydroxybutyrate or ¹⁵N-urea for 24 h. The labeling of cecal content metabolites was analyzed by LC-MS.

(B) Contribution of dietary and circulating nutrients to carbohydrate fermentation pathways in gut microbiome. Mean ± s.e. N = 4.

(C) Contribution of dietary and circulating nutrients to cecal amino acid carbon. The names of essential amino acids (EAA) are written in blue and non-essential amino acids (NEAA) in black. Mean ± s.e. N = 4.

(D) Contribution of dietary and circulating nutrients to cecal amino acid nitrogen. As in (C), for nitrogen.

(E) Positive correlation, across amino acids in the cecal contents, of carbon contribution from dietary inulin and nitrogen contribution from circulating urea. Mean ± s.e. N = 4.

We next examined inputs to microbiome free amino acids, tracing also with ¹⁵N-labeled dietary protein and infused urea. Unlike mammals, most gut bacteria have the biosynthetic capacity to make all 20 proteogenic amino acids. Nevertheless, we observed that “essential amino acids,” which cannot be made by mammals and require the expression of extensive biosynthetic pathways in bacteria, are derived mainly from dietary proteins (Figure 2C). In contrast, “non-essential amino acids” are primarily synthesized within the gut microbiome, using dietary inulin and circulating lactate as carbon sources. Microbiota depletion with antibiotics or in germ-free mice favored cecal accumulation of those amino acids coming (based on our isotope tracing studies) largely from dietary protein, and depletion of those being synthesized by the microbes (Figure S3B–G).

Dietary protein was the main nitrogen source for both essential and non-essential amino acids, with host urea also contributing substantially to the non-essential amino acids (Figure 2D). Dietary protein provides nitrogen to cecal amino acids mainly directly, not through circulating urea (Figure S3H–I). Consistent with the gut microbiome synthesizing amino acids from fiber carbon and urea nitrogen, across amino acids, urea’s nitrogen contribution correlated with inulin’s carbon contribution (Figure 2E).

Amino acid labeling from inulin was typically partial (i.e., one or a few of the amino acid’s carbons atoms were labeled), reflecting inulin’s carbons being scrambled with other inputs into central metabolism (Figure S3J). In contrast, labeling from ¹³C, ¹⁵N-proteins were typically complete (or complete except for the nitrogen label, Figure S3K), indicating direct usage of intact amino acids after proteolysis (sometimes after a cycle of deamination and re-amination). Consistent with such re-amination, the combination of ¹⁵N-urea infusion and ¹³C-protein feeding produced some double-labeled (¹³C,¹⁵N-labeled) amino acids (Figure S3L).

Lastly, the amino acids synthesized by the microbiome, stay in the microbiome: We do not observe discernible labeling of these amino acids in the host (Figure S3M). Taken together, we found that: (i) essential amino acids, although capable of being synthesized by the microbiome, come mainly from the diet and do not go through any carbon rearrangements, (ii) the most closely TCA-linked non-essential amino acids are substantially synthesized by the microbiome using carbon from fiber scrambled with other carbon via central metabolic reactions, and (iii) transamination reactions partially mix nitrogen from diet-derived amino acids with nitrogen from host urea.

Diverse microbiome products come from dietary protein

We next examined the carbon inputs to the other major microbiome products, especially the ones excreted into the portal circulation (Table 1). As expected, SCFAs, the most abundant microbial metabolites, come mainly from dietary fiber. Many less abundant ones, however, are mainly derived from dietary protein (Table 1).

In addition to classical microbiome products, we also observed metabolites that are made in a collaborative manner, with the host carrying out the final synthesis using microbiome-derived inputs. For example, a wide range of microbiome-derived carboxylic acids are conjugated to glycine in the liver and the kidneys to make different acyl-glycines (Figure S4B–E) (Wikoff et al., 2009).

We also examined the host clearance mechanisms of microbiome metabolites, based on arterial-venous gradients across the liver and kidney and levels in the urine. SCFAs and branched-chain fatty acids were avidly consumed by the liver. Most microbiome-derived metabolites were excreted by the kidney into the urine, with the notable exception of SCFAs, which are actively reabsorbed (Table S1A) (Jang et al., 2019; Ullrich et al., 1982). Thus, we establish dietary protein as a major precursor to many microbiome metabolites and identify host-microbiome interplay in the metabolism of SCFAs, including their renal reabsorption and use by liver and kidney for the synthesis of acyl-glycines.

Circulating levels of microbiota metabolites are controlled by protein digestibility

We found that many microbiome-derived metabolites are derived from unabsorbed dietary protein that reaches the colon. We hypothesized that the circulating levels of such metabolites would depend on the extent of dietary protein reaching the colon microbiome. To manipulate this, we fed mice diets in which a portion of the protein (casein, which in part reaches the colonic microbiome) was replaced with free amino acids (which are essentially fully absorbed in the small intestine) (Figure 3A). After 2 weeks, we performed metabolomics on the systemic blood. As expected, diets with less intact protein and more free amino acids tended to increase circulating amino acid levels (Figure 3B). Importantly, protein-derived circulating microbial metabolites (phenols, indoles, and acyl-glycines) fell in tandem (Figure 3C–I). Thus, knowledge of the nutrient sources of microbiome metabolites can be applied to manipulate their systemic levels.

Figure 3. Circulating levels of microbiota metabolites depend on protein reaching the microbiome.

(A) Compositions of diets used in the figure. “Protein” is casein. “Amino acids” are composition-matched free amino acids.

(B) Concentration of circulating amino acids in systemic circulation after two weeks test diet relative to free amino acids diet. Serum was taken at ad lib fed state. Each metabolite is a line. Mean, N = 4 mice.

(C) As in (B), for phenols. Mean, N = 4 mice.

(D) As in (B), for indoles. Mean, N = 4 mice.

(E) As in (B), for acylglycines. Mean, N = 4 mice.

(F) As in (B) for benzoic acid. Mean ± s.e., N = 4 mice.

(G) As in (F), for serotonin. Mean ± s.e., N = 4 mice.

(H) As in (F), for valerylglycine. Mean ± s.e., N = 4 mice.

(I) Correlation between dietary protein (as opposed to free amino acid) fraction in diet and metabolite abundances (relative to amino acid diet). The volcano plot shows Pearson coefficient and P value of correlation between metabolite levels to casein abundance in diet.

Gut bacterial growth is synchronized with host feeding

Thus far, we have reported inputs and outputs of the gut microbiome as a whole. We now shift to examining the growth and metabolism of specific bacterial genera. To this end, we deployed proteomics to measure gut microbial peptides and their labeling, focusing on peptide sequences specific to a single bacterial genus (Figure 4A).

Figure 4. Growth rate of different gut bacterial genera quantified by isotope tracing. See also^{Figure S5}.

(A) Experimental approach for isotope tracing into specific gut bacteria. Only peptides that are specific to a particular bacterial genus were examined.

(B) Growth rate quantification using D₂O. Mice received D₂O by i.p. injection followed by D₂O drinking water and cecal content labeling was measured over time by proteomics and metabolomics. Mice were fed ad lib; tissues were harvested at 9am.

(C) Calculation of newly synthesized peptide fraction (θ). The experimentally observed peptide mass isotope distribution was fit to a linear combination of unlabeled peptide (“old,” heavy forms from natural isotope abundance) and newly synthesized peptide (“new,” heavy forms from isotope labeling pattern of free cecal amino acids and from natural isotope abundance).

(D) Different cellular compartments from the same bacterial genus show similar labeling rate. Mean, N = 5 mice for each time point.

(E) Genus-specific growth rates were determined by a single exponential fitting, as a function of time, of θ (mean across both different peptides measured from that genus and replicate mice). Mean±s.e., N = 5 mice for each time point.

(F) Bacterial replication half time of different gut bacteria. Data are exponential fits ±s.e.

(G) The gut bacteria synthesize protein in sync with the physiological feeding patterns of the host. The figure shows the average newly synthesized peptide fraction (θ) for different gut bacterial genera after D₂O labeling during daytime vs nighttime. Each line connects the daytime and nighttime measurements for one genus. Mean, N = 5 mice for daytime and for nighttime.

To quantify protein synthesis in different gut microbial genera, we used deuterated water (D₂O) tracing (Holmes et al., 2015; O’Brien et al., 2020). To achieve steady-state labeling of body water, we gave mice D₂O by bolus injection followed by mixing it into drinking water. Peptide labeling in the cecal contents was then measured by proteomics (Figure 4B).

A key technical challenge in using proteomics to read out metabolic activity is the complexity, arising from natural isotope abundances, of peptide mass spectra. We used liquid chromatography-high resolution mass spectrometry to obtain the full scan (MS1) mass isotope distribution for each peptide of interest, with MS/MS analysis of the unlabeled form used to determine the peptide’s identity. We then calculated, based on the mass isotope distribution, the fraction of peptide that was newly synthesized (θ). To this end, first, we calculated the mass isotope distribution of unlabeled peptides based on natural isotope abundances (“old”). Second, we calculated the expected mass isotope distribution of a newly-synthesized peptide generated from cecal free amino acids, whose labeling we experimentally measured by metabolomics. Then, we determined the fraction of newly synthesized (θ) by linear interpolation between the “old” and “newly synthesized” spectra (Figure 4C). To verify this approach in vitro, we cultured Clostridium sporogenes and Bacteroides dorei in media enriched with D₂O and measured growth rate as is typically done (based on OD₆₀₀) and as above (using media in place of cecal amino acid labeling), finding good agreement (Figure S5A–C).

We then measured the newly synthesized fraction (θ) for a minimum of 5 peptides for each bacterial genus in vivo, with abundant gut bacteria yielding θ for over 100 characteristic peptides. Irrespective of their intracellular location, different peptides from the same bacterial genus tended to label at a similar rate (Figure 4D, S5D–E). Labeling rate varied across bacterial genera, with a half doubling time ranging from 2.5 h for Akkermansia to 8 h for Lactobacillus, which still markedly exceeded the labeling rate of host intestinal proteins (> 24 h half doubling time) (Figure 4E–F, Figure S5F).

Our prior analyses revealed that the microbiome is fed substantially by dietary components. Accordingly, we hypothesized that microbial growth synchronizes with physiological feeding, which in mice occurs mainly during the nighttime. To assess the diurnal rhythm of gut bacterial protein synthesis, mice were given D₂O for 6 h intervals throughout the diurnal cycle, followed by proteomic analysis of their cecal contents. Every measured bacterial genus showed greater protein synthesis during nighttime than daytime (Figure 4G). Thus, gut bacteria grow in sync with the physiological feeding patterns of the host.

Preferred carbon sources differ across gut bacteria

Next, we quantitated the carbon feedstocks of different microbes, by combining ¹³C-nutrient labeling and proteomics. Each ¹³C-labeled nutrient (dietary inulin, dietary algal protein, or circulating lactate) was provided for 24 hours, which is sufficient to achieve steady-state labeling in the gut bacteria. Our analysis strategy involved two steps: first, we calculated, based on each genus-specific peptide’s observed mass isotope distribution, its relative ¹³C-enrichment (γ) compared to that of cecal free amino acids (Figure 5A). Mathematically, this calculation is identical to the calculation of θ in the D₂O case, except here, the tracer is a particular ¹³C-labeled nutrient, which unlike D₂O is used preferentially by certain bacterial genera. The observed peptide’s relative ¹³C-enrichment multiplied by the average contribution of that ¹³C-tracer to the gut microbial amino acids pool (L_{AA_avg←nutrient}) gives a quantitative measure of the tracer’s contribution to the observed genus-specific peptide. Averaging across such peptides gives a fractional contribution of the ¹³C-labeled nutrient to protein synthesis in a bacterial genus.

Figure 5. Preferred carbon sources differ across gut bacteria. See also^{Figure S6}.

(A) Calculation of peptide relative ¹³C-enrichment (γ) and carbon contribution from the tracer to a bacterial genus (f_{genus←nutrient}). First, the experimentally observed peptide mass isotope distribution was fit to a linear combination of unlabeled peptide (heavy forms from natural isotope abundance) and a peptide made from free cecal amino acids (heavy forms from isotope labeling pattern of free cecal amino acids and from natural isotope abundance), yielding. Then, f_{genus←nutrient} was determined by correcting for the fractional contribution of that tracer to the cecal free amino acid pools.

(B) Carbon contribution of dietary inulin across bacterial genera. Mean±s.e., N = 4 mice.

(C) Carbon contribution of dietary algal protein across bacterial genera. Mean±s.e., N = 6 mice.

(D) Carbon contribution of circulating lactate across bacterial genera. Mean±s.e., N = 7 mice.

(E) Experimental scheme of high-inulin diet feeding followed by 16S rRNA gene amplicon sequencing.

(F) Genus-level microbiota composition changes after high-inulin diet. The genera increased after high-inulin diet prefer inulin in (B). Mean±s.e., N=3 mice. *P<0.05 and **P<0.01 by two-sided Student’s t-test.

(G) Correlation between genera abundance changes and carbon-source preference.

(H- J) As in (E - G), for algal protein-supplemented diet.

Using this method, we measured feedstocks of the bacterial genera that were detected in every proteomics experiment. We were also able to make species-specific measurements in some cases (Figure S6A–F). We observed marked differences in nutrient preferences across microbiota. For example, Bacteroides and Clostridium use over four-fold more inulin than Akkermansia, Muribaculum, or Alistipes (Figure 5B, S6A). Overall, bacteria from the phylum Firmicutes, used more dietary protein than did Bacteroidetes (Firmicutes 0.237 ± 0.052; Bacteroidetes 0.175 ± 0.031, p = 0.02). Akkermansia, which is generally considered a health-promoting gut microbe, used among the least dietary inulin and protein (Figure 5B–C, S6A–B). In contrast, it used by far the most circulating lactate from the host (Figure 5D, S6C).

We were curious whether these bacterial nutrient preferences predict microbiome composition changes upon dietary changes. To explore this possibility, we fed mice an inulin-enriched or algal protein-enriched diet for two days and measured microbiome composition by 16S rRNA gene amplicon sequencing. Bacteroides, the top consumer of ¹³C-inulin, increased by 4-fold after high inulin diet (Figure 5E–G). Clostridium, another high inulin consumer, also increased by 2-fold. Other genera that use less inulin carbon were either unchanged or slightly decreased. Similar consistency between microbes’ nutrient preference and abundance changes was observed in mice fed the algal protein-enriched diet (Figure 5H–J). Carbon-source preference measured by proteomics (f_{genus←nutrient}) correlates with abundance change following a diet shift measured by 16S rRNA gene amplicon sequencing, for both the inulin and algal protein conditions (Figure 5G, J). Thus, the nutrient preferences of different gut bacteria help explain microbiome compositional changes following dietary manipulations (David et al., 2014).

Firmicutes consume dietary protein while Bacteroidetes consume secreted host protein

Lastly, we turned to the nitrogen source preferences of different gut bacteria, comparing ¹⁵N-labeled dietary protein feeding to ¹⁵N-urea infusion. The analytical approach was identical to that employed above for carbon source preferences. Bacterial genera that highly use carbon from dietary protein also highly use nitrogen from dietary protein, consistent with amino acids from dietary protein being assimilated intact in bacterial proteomes (Figure 6A, S6D, S6G).

Figure 6. Firmicutes favor dietary protein while Bacteroidetes prefer secreted host protein. See also^{Figure S6,S7}.

(A) Nitrogen contribution of dietary algal protein across bacterial genera. Mean±s.e., N = 6 mice.

(B) Nitrogen contribution of circulating urea across bacterial genera. Mean±s.e., N = 6 mice.

(C) Experimental scheme of 72 hr urea infusion followed by 16S rRNA gene amplicon sequencing.

(D) Urea infusions increased urea concentration in systemic circulation. N = 5 mice. ***P<0.001 by two-sided Student’s t-test.

(E) Genus-level microbiota composition changes after urea infusion. The genera increased after urea infusion prefer urea in (B). Mean±s.e., N=5 mice. *P<0.05 by two-sided Student’s t-test.

(F) Correlation between genera abundance changes and nitrogen-source preference.

(G) Experimental schematic of long-term ¹⁵N-lysine and ¹⁵N-arginine infusion to probe the contribution of secreted host proteins to different bacterial genera.

(H) Nitrogen contribution of secreted host proteins across bacterial genera. Mean±s.e., N = 5 mice.

(I) Negative correlation between f_{genus←dietary proteins N} and f_{genus←secreted proteins N}.

(J) Summary of carbon and nitrogen inputs to different gut bacteria. Firmicutes prefer dietary carbon sources (fiber and protein) and nitrogen from host circulating urea. Bacteroidetes heavily use dietary fiber, while using on host secreted proteins for nitrogen. Verrucomicrobia prefers host secreted nutrients, both protein and circulating small molecules (lactate, urea).

Conversely, among members of the phylum Firmicutes, genera preferring urea nitrogen tended to be avid inulin users, i.e., to synthesize their own amino acids using inulin and urea (Figure 6B, S6E, S6H). This includes some urease-negative genera, which presumably acquire urea nitrogen via cross-feeding. Moreover, again among Firmicutes, we also saw the expected trade-off where some genera prefer nitrogen from dietary protein, and others from circulating urea (Figure S6I). Following intravenous urea infusions to raise circulating urea concentrations, abundance of those Firmicutes preferring urea, along with Akkermansia, increased substantially (Figure 6C–F).

Compared to Firmicutes, the lower use of both dietary protein and circulating urea nitrogen by Bacteroidetes raised a key question: How do Bacteroidetes get nitrogen? Some members of gut microbiome (e.g. Bacteroides and Akkermansia) are capable of digesting host secreted proteins such as mucins (Berry et al., 2013; Reese et al., 2018). We hypothesized that host secreted proteins are a key source of Bacteroidetes nitrogen. To probe this possibility, we performed long-term ¹⁵N-labeled lysine and arginine infusions (12, 18, 36 h) to label host proteins in the colon (Figure 6G and Figure S7A–E). Despite not directly feeding the microbiome (Figure 1F, S7E), lysine and arginine did contribute after 36-h infusion, consistent with the labeling occurring via host proteins. Such labeling occurred preferentially in Bacteroidetes and Akkermansia (Figure 6H, S6F). The nitrogen contributions from dietary and secreted host proteins were anti-correlated, consistent with some gut bacteria preferentially consuming dietary protein, and others host protein (Figure 6I). Bacterial genera with a greater preference for dietary protein, whose availability depends on host feeding, grow more differently between daytime and nighttime (Figure S6J–K). Thus, dietary proteins and circulating urea are the major nitrogen feedstock of Firmicutes, while secreted host proteins provide nitrogen to Bacteroidetes.

DISCUSSION

As for most microbial communities, the composition of the gut microbiome is shaped by nutrient availability. Here we developed quantitative isotope tracing approaches to measure the nutrient preferences of gut bacteria. In addition to dietary fiber and secreted host proteins, we establish dietary protein and circulating host lactate, 3-hydroxybutyrate, and urea as important nutrients feeding gut bacteria. Importantly, we rule out direct contributions from other circulating host nutrients, like glucose and amino acids, to the colonic microbiome.

A key technical achievement is enabling tracing from different carbon and nitrogen sources into bacteria-specific peptides, thereby revealing the nutrient preferences of different bacteria within the complex and competitive gut lumen environment. We find that Firmicutes and Bacteroidetes differ systematically in their utilization of host secreted protein versus dietary protein: Firmicutes tend to acquire amino acids from dietary protein, while Bacteroidetes rely more on secreted host protein (Figure 6J). This may relate to different localization of bacteria within the colon, either in terms of central versus peripheral (closer to host mucus) or distal versus proximal (closer to incoming food remnants) (Albenberg et al., 2014; Li et al., 2015; Yasuda et al., 2015).

Within these two major families of gut bacteria, we found marked disparities in the use of dietary fiber as a carbon source. The most abundant Bacteroidetes’ genus is Bacteroides, and it was the most avid assimilator of fiber (inulin). In contrast, other types of bacteria in the same phylum hardly consumed inulin. Likewise, some Firmicutes like Clostridium avidly used fiber, while others did not. Strikingly, feeding a fiber-enriched diet led to an increased abundance of Bacteroides and Clostridium, the precise genera that most actively assimilate fiber based on isotope tracing.

A similar trend was observed in the case of dietary supplementation with algal protein: Firmicutes, which actively use such protein, tended to increase in abundance. Algal protein (the only type commercially available in bulk in ¹³C-labeled form) may be particularly hard for mammals to digest. This is reflected in the limited appearance of ¹³C-labeled amino acids from algal protein in the portal circulation, and instead extensive passage from the intestine into the colon. This influx of dietary protein to the microbiome was a major contributor to secreted microbiome metabolites: As shown by replacing intact dietary protein with more absorbable (and thus less microbiome-accessible) free amino acids, the production and hence systemic concentration of these products depends on dietary protein reaching the colonic microbiome. An important future question is whether the nature of dietary protein (e.g. plant or animal-based) impacts passage through the small intestine to the colonic microbiome and thereby shapes microbiome composition or metabolite secretion (Madsen et al., 2017; Wali et al., 2021).

Host circulating metabolite levels may also impact microbiome nutrient access and ultimately composition. Here we show such effects are likely limited to the few host metabolites that meaningfully penetrate the microbiome: urea, 3-hydroxybutyrate, and lactate. Among them, lactate was recently shown to feed the gut microbiome in human marathon runners (Scheiman et al., 2019). Among gut bacteria, Akkermansia most avidly use circulating lactate. Akkermansia are mucin degraders, and their proximity to the gut epithelial wall may augment their access to lactate from the host circulation. Akkermansia are more abundant in athletes, and exercise increases their levels in mice and human (Liu et al., 2017; Munukka et al., 2018). A possible mechanism involves increased circulating lactate levels following exercise directly feeding Akkermansia. Whether lactate induced Akkermansia growth in part mediates beneficial effects of exercise is an important open question. Consistent with their urea preferences measured by isotope tracing, Akkermansia and certain genera within Firmicutes (e.g., Roseburia, Butyricoccus, Ruminococcus) also increase in abundance upon experimental elevation of circulating host urea.

Ultimately, manipulating the microbiome requires understanding which nutrients different bacteria consume, and how such consumption impacts microbiome composition and product secretion. Through isotope tracing, including proteomic measurements that offer bacterial genus specificity, we provide foundational knowledge about which nutrients feed the gut microbiome, and which bacteria prefer which nutrients. The methodologies developed here are poised for broader application, to eventually contribute to the holistic and quantitative understanding of the diet-microbiome-health connection.

LIMITATIONS OF THE STUDY

Our investigation focuses solely on healthy mice fed standard chow (in some cases with specific fiber or protein supplements). Measurements of microbiome feedstocks are limited to isotope tracing and mass spectrometry. Feedstocks of different bacteria are determined based on the isotopic signatures of bacteria-specific peptides. Peptide identification involves a 2% false discovery rate. Taxonomic assignment is based on bacterial proteome sequences available on Uniprot (Gurdeep Singh et al., 2019). Orthogonal approaches, which could provide measurement validation or complementary information, such as fluorescent-activated cell sorting of bacteria, were not explored (Batani et al., 2019). In most cases, taxonomic assignment was limited to the genus level, due to lack of sufficient specificity of the detected peptide sequences. In the future, improved sensitivity may enable species- or strain-specific peptide sequence measurements.

STAR*METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Professor Joshua D. Rabinowitz (joshr@princeton.edu).

Materials Availability

This study did not generate new unique regents or new mouse lines.

Data and Code Availability

The proteomics datasets generated during this study are deposited in PRIDE: PXD031015. The isotope tracing data are included in Table S2. The taxonomic assignment of the detected tryptic peptides in the study are included in Table S3. Composition of the diet used in the study are included in Table S4. The 16S rRNA gene amplicon sequencing datasets generated during this study are available in Table S5. The code for peptide enrichment calculations generated during this study is available at GitHub: (https://github.com/xxing9703/pepMID_simul). Any additional information required to re-analyze the data reported in this work is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mouse studies.

Mouse studies followed protocols approved by the Princeton University Animal Care and Use Committee. Unless otherwise indicated, 7–9-week-old male C57BL/6NCrl mice (strain 027; Charles River Laboratories) were group-housed on a normal light-dark cycle (8:00–20:00) with free access to water and chow.

Bacterial culture studies.

B. dorei, C. sporogenes, E. coli, S. aureus and L. reuteri glycerol stocks were brought into an anerobic chamber (70% N₂, 25% CO₂, 5% H₂) and grown in liquid media: L. reuteri was grown on MRS (MRS Broth, Sigma); E. coli was grown on LB (Luria Broth, Sigma); S.aureus was grown in TSB (Tryptic Soy Broth, Bacto) and C. sporogenes and B. dorei were grown in GAM (GAM Broth Modified, HyServe).

METHOD DETAILS

Mouse gavage and nutrient feeding.

For the ¹³C-nutrient gavage experiments, mice were fasted at 9 am and received a 1:2:4 mixture of inulin, protein/amino acids, and starch (0.5 g kg⁻¹ inulin, 1 g kg⁻¹ protein/amino acids 2g kg⁻¹ starch dissolved in water) at 3 pm via oral gavage with a plastic feeding tube (Instech Laboratories). Food was given back at 8 pm.

For the mouse experiments involving labeled nutrient feeding, the labeled diet was prepared by adding ¹³C/¹⁵N-nutrients to a diet mixture premix (modified from normal diet with reduced protein, inulin, and starch content, Research diets Inc, D20030303). The final enrichment for each labeled dietary nutrient was 10% - 25% (with observed labeling corrected by dividing by the fraction dietary nutrient labeled). The contribution of each dietary nutrient to metabolites is calculated by the metabolite labeling enrichment normalized to the final enrichment of each labeled dietary nutrient. All diets shared the same final macronutrient composition (40% starch, 20% protein or amino acids, 7.5% inulin and 2.5% cellulose). Mice were first adapted to a non-labeled diet (of identical composition to the subsequent labeled diet) for 10 days, and then fed labeled diet for 24 h prior to sacrifice.

For the deuterium water drinking experiment, mice were administered a bolus intraperitoneal injected of D₂O (1.26 % w/w relative to body weight), followed by having ad lib access to 3% D₂O drinking water.

For the protein and amino acids diet feeding experiment, mice were fed on casein or compositional matched amino acids diet (20% casein, 13% casein + 7% amino acids, 7% casein +13% amino acids, and 20% amino acids as protein/amino acids sources, Table S4) for 2 weeks. Serum was sampled by tail-bleed at 9 am ad lib.

Intravenous infusions.

To quantify contribution of circulating nutrients to microbiota metabolism, 9–11-week-old C57BL/6 mice were catheterized in house in the right jugular vein. The mice were infused with carbon or nitrogen-labeled tracer starting at 3:30 pm without any fasting. Infusion rate was 0.1 ul/min/g. Infusion solutions are described in Table S2A. Overnight (24 h) infusions both started and finished around 9 am. The contribution of circulating nutrient to each metabolite is calculated by the metabolite labeling enrichment normalized to the average tracer serum enrichment throughout 24 hr.

Antibiotics treatment.

To deplete the mouse resident microbiome, an antibiotic drinking water protocol was used. In brief, mice were treated with a cocktail of antibiotics (1 g/L ampicillin, 1 g/L neomycin, 1 g/L metronidazole, and 1 g/L vancomycin) in both their drinking water 14 days. To make the drinking water more palatable, 5% aspartame was added. The effectiveness of antibiotics treatments was verified by observing much lower SCFAs in the feces by LC-MS.

Sample collection.

Systemic blood samples (~6 μl) were collected by tail bleeding. For sampling from tissue-specific draining veins, a mouse was put under anesthesia and different tissue veins were exposed, and blood samples were pulled with an insulin syringe (BD insulin syringes, # SY8290328291) insertion into the vein. Successful isolation of portal vein was confirmed by much higher (> 10x) concentrations of SCFAs and secondary bile acids (deoxycholic acid and lithocholic acid) than systemic vein; hepatic vein was confirmed by much lower secondary bile acids, SCFAs and higher glucose, 3-hydroxybutyrate levels compared to portal vein. Mouse urine was collected from the urinary bladder using a syringe. All serum samples were placed on ice without anticoagulant for 15 min, and centrifuged at 16,000 × g for 15 min at 4 C.

Tissues were harvested by quick dissection and snap freezing (<5 sec) in liquid nitrogen with a pre-cooled Wollenberger clamp; intestinal contents were removed before clamping. For cecal content sampling, the mouse cecum was first removed and cut on the surface, then the cecal content was squeezed out using a tweezer followed by freeze clamping. Whole liver, intestine, and intestinal contents were collected and grounded to homogenous powder. To sample fresh feces, the mouse belly was gently massaged to induce defecation and fresh feces were freeze clamped. For long-term feces collection, a mouse was transferred to a new cage and mouse fecal pellets on the bedding were collected every 1~2 h and freeze clamped. Serum, tissue, and feces samples were kept at −80 °C until further analysis.

16S rRNA gene amplicon sequencing and analysis.

Extraction of Bacterial DNA from cecal or fecal samples was performed using the Power Soil DNA Isolation kit (QIAGEN). A section of the 16S rRNA gene (~250 bp, V4 region) was amplified, and Illumina sequencing libraries were prepared from these amplicons according to a previously published protocol and primers (Caporaso et al., 2012). Libraries were further pooled together at equal molar ratios and sequenced on an Illumina HiSeq 2500 Rapid Flowcell or MiSeq as paired-end reads. These reads were 2×150 bp with an average depth of ~20,000 reads. Also included were 8 bp index reads, following the manufacturer’s protocol (Illumina, USA). Pass-Filter reads were generated from raw sequencing reads using Illumina HiSeq Control Software. Samples were de-multiplexed using the index reads. The DADA2 plugin within QIIME2 version 2018.6 was used to inferred Amplicon sequencing variants (ASVs) from the unmerged paired-end sequences (Bolyen et al., 2019; Callahan et al., 2016). The forward reads were trimmed at 150 bp and the reverse reads trimmed at 140 bp, with all other DADA2 as default. Taxonomy was assigned to the resulting ASVs with a naïve Bayes classifier trained on the Greengenes database version 13.8, with only the target region of the 16S rRNA gene used to train the classifier (Bokulich et al., 2018; McDonald et al., 2012). Downstream analyses were performed MATLAB (Hunter, 2007; McKinney, 2010).

Bacterial culture studies.

For the D₂O experiment, 250 – 1000 μl D₂O was added into the media (5–10 mL, to reach a final enrichment of 5–10%) with either B. dorei or C. sporogenes, and OD₆₀₀ was recorded at the addition. After every 25–30 min, OD₆₀₀ was recorded and 100–200 μl bacterial solution was taken for metabolomics and proteomics analysis. The newly synthesized fraction of bacteria was calculated by (OD₆₀₀ – OD_{600, 0min})/OD₆₀₀.

Bacterial colonization in mice.

Mice were treated with antibiotics in drinking water for 10 days. On day 11, no antibiotics were administered, and mice were gavaged with 250 μl of bacterial consortia consisting of urease-negative bacteria (B. dorei, C. sporogenes and E. coli) or a combination of urease-negative and urease-positive bacteria (B. dorei, C. sporogenes, E. coli, S. aureus and L. reuteri).

Metabolite extraction.

For serum samples, 3 ul serum was added to 90 ul methanol and incubated on ice for 10 min, followed by centrifugation at 17,000 × g for 10 min at 4°C. The supernatant was transferred to an MS vial until further analysis. For tissues and feces samples, frozen samples were first ground at liquid nitrogen temperature with a cryomill (Restch, Newtown, PA). The resulting tissue powder was extracted with 40:40:20 methanol: acetonitrile: water (40 ul extraction solvent per 1 mg tissue) for 10 min on ice, followed by centrifugation at 17,000 × g for 10 min, the supernatant was transferred to a MS vial until further analysis.

Measurements of metabolites, protein, and polysaccharides.

To measure metabolites in serum, tissue and feces samples, a quadrupole orbitrap mass spectrometer (Q Exactive; Thermo Fisher Scientific) was coupled to a Vanquish UHPLC system (Thermo Fisher Scientific) with electrospray ionization and scan range m/z from 60 to 1000 at 1 Hz, with a 140,000 resolution. LC separation was performed on an XBridge BEH Amide column (2.1×150 mm, 2.5 μm particle size, 130 Å pore size; Waters Corporation) using a gradient of solvent A (95:5 water: acetonitrile with 20 mM of ammonium acetate and 20 mM of ammonium hydroxide, pH 9.45) and solvent B (acetonitrile). Flow rate was 150 μl/min. The LC gradient was: 0 min, 85% B; 2 min, 85% B; 3 min, 80% B; 5 min, 80% B; 6 min, 75% B; 7 min, 75% B; 8 min, 70% B; 9 min, 70% B; 10 min, 50% B; 12 min, 50% B; 13 min, 25% B; 16 min, 25% B; 18 min, 0% B; 23 min, 0% B; 24 min, 85% B; and 30 min, 85% B. Injection volume was 5–10 μl and autosampler temperature was set at 4°C. For cysteine measurement, samples were derivatized before measurement as follows: Serum, cecal content or feces samples were extracted and centrifuged. To the supernatant, 2 mM N-ethylmaleimide was added and incubated at room temperature for 20 min. The resulting mixture was transferred to a MS vial. Derivatized cysteine has a m/z at 245.06015 in negative mode.

To quantify the metabolite concentration in serum and tissue samples, either isotope spike-in or standard spike-in was performed. For isotope spike-in, known concentrations of isotope-labeled standard were added to the serum or tissues extraction solution, then the concentration was calculated by the ratio of labeled and unlabeled metabolites. When isotope standard is not available, a serially diluted non-labeled standard was added, and a linear fitting between measured total ion count and added concentration of standard was generated. Then, the concentration of endogenous metabolite was determined by the x intercept of the fitting line.

Starch and inulin were measured by acid hydrolysis and LC-MS. In brief, 5–10 mg sample was mixed with 10 μl 2 M hydrochloric acid, and samples were incubated at 80°C for 2 h. After cooling down, the resulting mixture was neutralized with 12 μl saturated sodium bicarbonate, followed with 88 μl 1:1 acetonitrile: methanol solution. After centrifugation at 17,000 × g for 10 min at 4°C, the supernatant was transferred to a MS vial. Inulin and starch concentration in samples was inferred from total ion count of fructose and glucose, respectively.

SCFAs and BCFAs were derivatized and measured by LC-MS. Serum (5 μl) or tissue samples (~10 mg) were added to 100 μl derivatizing reagents containing 12 mM 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide, 21 mM 3-Nitrophenylhydrazine hydrochloride acid and pyridine (2.4% v/v) in methanol. The reaction was incubated at 4°C for 1 h. Then, the reaction mixture was centrifuged at 17,000 g for 10 min. 20 μl supernatant was quenched with 200 μl 0.5 mM beta-mercaptoethanol in 0.1% formic acid water. After centrifugation at 17,000 g for 10 min, the supernatant was transferred to MS vials until further analysis. The measurement of SCFAs and BCFAs are performed using the same Q Exactive PLUS hybrid quadrupole-orbitrap mass spectrometer with different column and LC setup. LC separation was on Acquity UPLC BEH C18 column (2.1 mm × 100 mm, 1.7 5 μm particle size, 130 Å pore size, Waters, Milford, MA) using a gradient of solvent A (water) and solvent B (methanol). Flow rate was 200 μL/min. The LC gradient was : 0 min, 10% B; 1 min, 10% B; 5 min, 30% B; 11 min 100% B; 14 min, 100% B; 14.5 min 10% B; 22 min 10 % B. Autosampler temperature was 5 °C, column temperature was 60 °C and injection volume was 10 μl. Ion masses for derivatized acetate, propionate, butyrate, iso-butyrate, valeric acid, isovaleric acid, 2-methylbutyrate, 4-methylvaleric acid were 194.0571, 208.0728, 222.0884, 222.0884, 236.1041, 236.1041, 236.1041, 250.1197 in negative mode, respectively.

The ammonia derivatization method was modified from the previous reported Berthelot reaction assay (Spinelli et al., 2017b, 2017a). In brief, 20 mg tissue or 10 μl serum was extracted by using 200 μl 80% methanol. 100 μl of the metabolite extract was mixed with 100 μl. Solution #1 (100 mM Phenol, 50 mg/L sodium nitroprusside) and 100 μl. Solution #2 (0.38 M dibasic sodium phosphate, 125 mM NaOH, 1% sodium hypochlorite, available chlorine 10–15%). The mixture was incubated at 40°C for 30 min. Then, 100 μl reaction solution was mixed with 200 μl methanol to oversaturate the inorganic salt to quench the reaction. The final solution was centrifuged for 30 min. Then the supernatant was loaded to LC-MS for analysis. Ion mass for derivatized ammonia is 198.05605 in negative ion mode.

Protein amino acid composition was measured by acid hydrolysis. Approximately 10 mg of protein was extracted with 400 μl methanol, 200 μl chloroform and 300 μl water, followed by centrifugation at 20,000 × g for 10 min at 4 °C. The upper layer was removed. The resulting mixture was further extracted with 600 μl methanol twice and supernatant was discarded. The resulting precipitate was dried under nitrogen gas and then hydrolyzed with 250 μl 6 M hydrochloric acid incubated overnight at 115°C. After incubation, the samples were dried under nitrogen gas and reconstituted in 1 mL methanol, and the supernatant was transferred to a MS vial for analysis. Amino acid composition of the proteins used in the study are shown in Figure S4F, and such differences in protein amino acid composition do not correlate with the quantified dietary protein contribution to cecal amino acids (Figure S4G).

Proteomics sample preparation.

Proteomics samples were prepared mostly as previously described (Gupta et al., 2018; Wühr et al., 2014). Mouse cecal samples (10 mg each) were dissolved in 400 μl lysis buffer (6M guanidium chloride, 2% cetrimonium bromide, 5 mM dithiothreitol, 50 mM (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid) (HEPES), pH 7.2). Then the sample mixture was put on ice and sonicated for 10 cycles (30 s on and 30 s off cycle, amplitude 50%) by a sonicator (Qsonica), followed by centrifugation at 20,000 × g for 20 min at 4 °C. The supernatant was taken and alkylated with 20 mM N-ethylmaleimide for 20 min at room temperature, 5 mM dithiothreitol was added to quench the excessive alkylating reagents. Proteins were purified by methanol-chloroform precipitation. The dried protein pellet was resuspended in 10 mM EPPS (N-(2-Hydroxyethyl) piperazine-N’-(3-propanesulfonic acid)) at pH 8.5 with 6 M guanidine hydrochloride. Samples were heated at 60°C for 15 min and protein concentration was determined by BCA assay (Pierce BCA Protein Assay Kit, Thermo Scientific). The protein mixture (30~50 μg) was diluted with 10 mM EPPS pH 8.5 to 2 M GuaCl and digested with 10 ng/μL LysC (Wako) at room temperature overnight. Samples were further diluted to 0.5 M GuaCl with 10 M EPPS pH 8.5 and digested with an additional 10 ng/μL LysC and 20 ng/μL sequencing grade Trypsin (Promega) at 37°C for 16 h. Samples were desalted using a SepPak cartridges (Waters) and then vacuum-dried and resuspended in 1% formic acid before mass spectrometry analysis.

Proteomics peptide measurement.

Samples were analyzed on an EASY-nLC 1200 (Thermo Fisher Scientific) HPLC coupled to an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) with Tune version 3.3. Peptides were separated on an Aurora Series emitter column (25 cm × 75 μm ID, 1.6 μm C18) (Ionopticks, Australia) and held at 60°C during separation using an in-house built column oven over 180 min, applying nonlinear acetonitrile gradients at a constant flow rate of 350 nL/min. The Fusion Lumos was operated in data dependent mode. The survey scan was performed at a resolution setting of 120k in orbitrap, followed by MS2 duty cycle of 1.5 s. The normalized collision energy for CID MS2 experiments was set to 30%.

Solvent A consisted of 2% DMSO (LC-MS-grade, Life Technologies), 0.125% formic acid (98%+, TCI America) in water (LC-MS-grade, OmniSolv, VWR), solvent B of 80% acetonitrile (LC-MS-grade, OmniSolv, Millipore Sigma), 2% DMSO and 0.125% formic acid in water. The following 120 min-gradient with percentage of solvent B were applied at a constant flow rate of 350 nL/min after thorough equilibration of the column to 0% B: 0% – 6% in 5 min; 6 – 25% in 160 min; 25% –100% in 10 min; 100% for 5 min. For electrospray ionization, 2.6 kV were applied between minutes 1 and 113 (or minutes 1 and 83 for fractionated samples) of the gradient through the column. To avoid carry-over of peptides, 2,2,2-trifluoroethanol (>99% Reagent plus, Millipore Sigma) was injected in a 30 min wash between each sample.

Proteomics data analysis.

The data was analyzed using GFY software licensed from Harvard (Nusinow et al., 2020). Thermo Fisher Scientific. raw files were converted to mzXML using ReAdW.exe. MS2 spectra assignment was performed using the SEQUEST algorithm v.28 (rev. 12) by searching the data against the combined reference proteomes for Mus Musculus, Bos Taurus, and all the abundant bacterial families detected in 16S rRNA gene amplicon sequencing (Bacteroidaceae, Porphyromonadaceae, Prevotellaceae, Rikenellaceae, Muribaculaceae, Lachnospiraceae, Ruminococcaceae, Erysipelotrichaceae, Oscillospiraceae, Clostridiaceae, Eubacteriaceae, Lactobacillaceae and Verrucomicrobiaceae) acquired from Uniprot on Jan 2021 (SwissProt + Trembl) along with common contaminants such as human keratins and trypsin. The target-decoy strategy was used to construct a second database of reverse sequences that were used to estimate the peptide false discovery rate (Elias and Gygi, 2007). A 20-ppm precursor ion tolerance with the requirement that both N- and C- terminal peptide ends are consistent with the protease specificities of LysC and Trypsin was used for SEQUEST searches, two missed cleavage was allowed. NEM was set as a static modification of cysteine residues (+125.047679 Da). An MS2 spectral assignment false discovery rate of 0.5% was achieved by applying the target decoy database search strategy. Linear Discriminant analysis was used for filtering with the following features: SEQUEST parameters XCorr and unique ΔXCorr, absolute peptide ion mass accuracy, peptide length and charge state. Forward peptides within three standard deviations of the theoretical m/z of the precursor were used as positive training set. All reverse peptides were used as negative training set. Linear Discriminant scores were used to sort peptides with at least seven residues and to filter with the desired cutoff. Furthermore, we performed a filtering step on the protein level by the “picked” protein FDR approach (Savitski et al., 2015). Protein redundancy was removed by assigning peptides to the minimal number of proteins which can explain all observed peptide, with above-described filtering criteria.

To quantify the intensities of all the isotopic peaks of the peptides, we used raw intensity. Missed cleavage peptides (more than one K or R in the peptide) and low signal to FT-noise peptides (M₀ S/N < 20) were removed. Peptide phylogenetic assignment was performed using Unipept 4.0 (Gurdeep Singh et al., 2019), ‘Equate I and L’ and ‘Advanced missed cleavage handling’ were not selected. Only peptides that are specific at a genus level were used for further analysis.

Quantification of newly-synthesized fraction of peptide.

To determine the newly synthesized fraction of a bacterial peptide in D₂O drinking water experiment, we first measured the cecal content free amino acids deuterium labeling pattern using metabolomics. Then, for each peptide, we simulated the expected isotope envelope pattern if the peptide were old, i.e., unlabeled with deuterium (I_old), versus if it were newly synthesized by taking up free amino acids from the cecal content (I_new). I_old was calculated based on the peptide’s molecular formula and ¹³C, ¹⁵N, ²H, ¹⁷O, ¹⁸O, ³²S, ³³S and ³⁶S natural abundance. I_new was calculated based on the peptide’s sequence and experimentally observed labeling of the corresponding cecal free amino acids (after natural isotope correction), and the natural isotope abundance of the unlabeled atoms in the peptide’s formula. The simulation of expected peptide isotope distribution and fitting was performed using a MATLAB code: https://github.com/xxing9703/pepMID_simul. Exact mass isotopic peaks with appreciable abundances were bundled by nominal mass into fraction M+0, M+1, …M+n, constituting the final simulated spectrum. A least square fit was used to find the scalar θ that best fit the measured peptide isotopic distribution (I_measured) to a linear combination of I_old and I_new:

$$I_{\mathit{\text{measured}}}=I_{\mathit{\text{old}}}\times (1-\theta )+I_{\mathit{\text{new}}}\times \theta$$

The root mean square error was determined for each peptide fitting, and any fitting with a root mean square error > 1% was removed. For genus-level turnover quantification, only genera with more than two measurements were kept in the analysis, with the median value across peptides reported.

Quantification of contribution of labeled nutrient to peptide.

To determine the contribution of a ¹³C- or ¹⁵N-labeled nutrient to a bacterial peptide, similar to the above approach, we first measured the cecal content free amino acids ¹³C- or ¹⁵N-labeling using metabolomics. Then, for each peptide, we simulated the expected isotope envelope pattern if the peptide were unlabeled (I_unlabeled) versus if it were synthesized from free cecal amino acids (I_free). A scalar γ (analogous to θ above) can then be determined by fitting the measured peptide isotope distribution (I_measured) to a linear combination of I_unlabeled and I_free. Note that γ will exceed 1 when a bacterial genus uses a particular nutrient in excess of that nutrient contribution’s to cecal free amino acids. Because the ¹³C- and ¹⁵N-labeling patterns are simpler than the D₂O labeling patterns, in lieu of carrying out this fitting, we instead determined γ (with the same conceptual and mathematical meaning) using simple algebraic equations.

Specifically, we measured γ for each peptide as follows:

$$\gamma =\frac{{\phi }_{\mathit{\text{measured}}}-{\phi }_{\mathit{\text{unlabeled}}}}{{\phi }_{\mathit{\text{free }}}-{\phi }_{\mathit{\text{unlabeled}}}}$$

where (with the exception of ¹³C-protein feeding data, discussed immediately below) φ is the average number of extra neutrons in a given peptide (or simulated peptide), relative to the M+0 form. This was calculated based on the experimentally observed (or simulated, as above) fraction of M+0, M+1, M+2, and M+3, which account for > 90% of the isotopes for each peptide (with more heavily labeled forms too low abundance and noisy to contribute productively to the measurements):

$$\phi =\frac{{\sum }_{i=0}^{3}i\cdot M_i}{{\sum }_{i=0}^3{M}_i}$$

For the ¹³C-protein feeding experiments, the most readily detected labeled forms involve incorporation of a single midsized U-¹³C-amino acid, which manifests as M+5 or M+6 peptide labeling. Other isotopic forms were sufficiently noisier, as to render their inclusion unhelpful. Accordingly, we calculated γ based on φ′:

$${\phi }^{\prime }=\frac{M_5+M_6}{M_0+M_5+M_6}$$

The above equations give nearly identical values for γ as fitting (as done to determine θ).

For genus-level measurements of feedstock contributions, only genera with more than 3 peptides measured per mouse was kept in the analysis, with the median value across peptides reported as γ_genus. Only genera that were consistently detected in proteomics, and the family of that genera detected (> 0.5%) in 16S rRNA gene amplicon sequencing were analyzed. The product of γ_genus and the contribution of each nutrient to cecal free amino acids (L_{AA_avg←nutrient}) was used to determine the contribution of each nutrient to bacterial genus (f_{genus←nutrient}):

$$f_{\mathit{\text{genus}}\leftarrow \mathit{\text{nutrient }}}={\gamma }_{\mathit{\text{genus}}}\times L_{AA\_\mathit{\text{avg}}\leftarrow \mathit{\text{nutrient}}}$$

where the contribution of each nutrient to bacterial protein pool (L_{AA_avg←Nutrient}) was calculated as the average labeling across amino acids, weighted based on their abundance in that genus’ protein and corrected for fraction of the nutrient interest labeled (T):

$$L_{AA\_\mathit{\text{avg}}\leftarrow nutrient}=\raisebox{1ex}{$\sum f_{\mathit{\text{Cecal}}\_AA\leftarrow \mathit{\text{nutrient}}}\times w{\%}_{AA,\mathit{\text{bacteria}}}$}\!\left/ \!\raisebox{-1ex}{$T$}\right.$$

with w%_AA,bacteria taken from literature (Purser and Buechler, 1966).

Quantification and statistical analysis.

A two-tailed, unpaired student’s t-test was used to calculate P values, with P<0.05 used to determine statistical significance.

Supplementary Material

1

Figure S1. Microbiome consumes dietary fiber and protein. Related to Table 1.

(A) Composition of the measured portal microbial metabolome. The pie charts show the relative molar abundance of different gut microbiota-associated metabolites in mice (N = 6 mice).

(B) As in (A), for cecal content microbial metabolome. (N = 6 mice)

(C) Experimental scheme. Mice received an oral gavage of 4:2:1 starch: protein (or free amino acids): inulin by weight. In each dietary condition, one component was ¹³C-labeled. After gavage of the labeled meal, tissue and serum metabolite labeling were measured over time by LC-MS.

(D) Dietary starch feeds the host, while dietary inulin feeds the microbiome. The data shows concentrations of labeled carbons in hexose in portal circulation (mean ± s.e., N = 3 mice).

(E) As in (D), for acetate (mean ± s.e., N = 3 mice).

(F) Heatmap showing the percentage of labeled carbon atoms in the indicated metabolites in portal circulation, following gavage of 4:2:1 starch: protein: inulin, with the indicated nutrient labeled. Each data point is median of N = 3 mice.

(G) Heatmap showing the molarity of total labeled carbon atoms in the indicated metabolites in cecal content, following gavage of 4:2:1 starch: protein: inulin, with the indicated nutrient labeled. Each data point is median of N = 3 mice.

(H) Metabolic fate of inulin and starch. Stacked bars show the fraction of gavaged inulin and starch that is converted into each of the indicated metabolic products, with the undigested fraction being excreted in the feces (mean ± s.e., N = 3 mice).

(I) Undigested carbohydrate in the feces. Graph shows the fraction of labeled hexose after cecal content hydrolysis, 12 h following gavage as above (mean ± s.e., N = 3 mice).

(J) Dietary amino acids feed the host while dietary algal protein feeds the microbiome. The data shows concentrations of labeled carbons in valine in portal circulation (mean ± s.e., N = 3 mice).

(K) As in (J), for acetate (mean ± s.e., N = 3 mice).

(L) Heatmap showing the percentage of labeled carbon atoms in the indicated amino acids in portal circulation, following gavage of 4:2:1 starch: protein (or free amino acids): inulin, with the indicated nutrient labeled. Each data point is median of N = 3 mice.

(M) Heatmap showing the molarity of total labeled carbon atoms in the indicated amino acids in cecal content, following gavage of 4:2:1 starch: protein (or free amino acids): inulin, with the indicated nutrient labeled. Each data point is median of N = 3 mice.

2

Figure S2. Circulating ammonia contributes to microbiota metabolism via circulating urea. Related to Figure 1.

(A) Normalized labeling of serum urea, ammonia and fecal amino acids after ¹⁵N-urea infusion (mean ± s.e., N = 4 mice). Feces labeling fraction is normalized to serum infused tracer (urea) labeling fraction.

(B) As in (A), for ¹⁵N-ammonia infusion (mean ± s.e., N = 5 mice).

(C) Experimental design. Gnotobiotic mice colonized with urease-negative bacteria (C.sporogenes, B.dorei and E.coli) or a combination of urease-negative and positive bacteria (C.sporogenes, B.dorei, E.coli, L.reuteri and S.aureus) were intravenously infused with ¹⁵N-urea for 24 hr. Cecal content amino acids labeling were measured by LC-MS.

(D) Normalized labeling of cecal ammonia and amino acids after ¹⁵N-urea infusion in urease-negative bacteria colonized gnotobiotic mice and urease-positive and negative bacteria colonized gnotobiotic mice (mean ± s.e., N = 3 mice). *P<0.05, **P<0.01 and ***P<0.001 by two-sided Student’s t-test.

(E) Ammonia concentration in systemic circulation and cecal content (mean ± s.e., N = 5 mice).

(F) As in (E), for urea (mean ± s.e., N = 5 mice).

(G) Contribution calculation from circulating ammonia to circulating urea, then to fecal amino acids.

(H) Contribution of circulating ammonia to fecal amino acids are quantitatively explained by circulating urea (mean ± s.e., N = 4 mice).

(I) Normalized labeling of fecal ammonia and amino acids after ¹⁵N-urea infusion in control and antibiotics treated mice (mean ± s.e., N = 3 mice). *P<0.05, **P<0.01 and ***P<0.001 by two-sided Student’s t-test.

(J) Normalized labeling of fecal ammonia after ¹⁵N-ammonia infusion in control and antibiotics treated mice (mean ± s.e., N = 5 mice for control mice, N = 4 for antibiotics-treated mice). **P<0.01 by two-sided Student’s t-test.

(K) Schematic of the pathway from circulating ammonia to luminal ammonia.

3

Figure S3. Quantitative analysis of microbiome amino acid metabolism. Related to Figure 2.

(A) Labeling fraction of cecal amino acids from 12 h or 24 h ¹³C-algal protein feeding. Mean ± s.e. N = 4 mice.

(B) Contribution of dietary and circulating nutrients to cecal amino acid carbon in antibiotics treated mice. Mean ± s.e. N = 3 mice.

(C) As in (B), for nitrogen.

(D) Cecal amino acids concentration fold-change in antibiotics-treated mice relative to mock treated mice. Mean ± s.e. N = 3. TCA-related amino acids are shown in blue bars and the others are shown in grey bars.

(E) Cecal amino acids coming directly from diet tend to go up with antibiotics treatment. Plot shows the correlation, across amino acids (each point is an individual amino acid), between concentration fold change in antibiotics-treated mice and carbon contribution from dietary proteins based on isotope tracing (without any antibiotic treatment).

(F) As in (D), for germ-free mice relative to conventional mice. Mean ± s.e. N = 3. TCA-related amino acids are shown in blue bars and the others are shown in grey bars.

(G) As in (E), for germ-free mice.

(H) Schematic of the possibility for dietary protein nitrogen to contribute to cecal amino acid nitrogen via circulating urea.

(I) Contribution from dietary proteins to cecal amino acid nitrogen are mostly independent of circulating urea. Mean ± s.e. N = 3.

(J) Cecal amino acid isotopic labeling forms from ¹³C,¹⁵N-Algal protein feeding. Mean ± s.e. N = 3.

(K) Cecal amino acid isotopic labeling forms from ¹³C-Inulin feeding. Mean ± s.e. N = 3.

(L) Total ion chromatogram of ¹³C₅, ¹⁵N₁-glutamate and ¹³C₄, ¹⁵N₁-aspartate from simultaneous ¹⁵N-urea infusion and ¹³C-protein feeding (relative to the ¹⁵N-urea infusion or ¹³C-protein feeding separately).

(M) Amino acids synthesized in the gut microbiome, stay in the microbiome, as urea contributes to microbiome amino acids but not host circulating amino acids. Mean ± s.e. N = 4 mice.

4

Figure S4. Sources of cecal metabolites and acyl-glycines. Related to Figure 2, Table 1.

(A) Heat maps showing the contribution of dietary or circulating nutrients to cecal metabolites. For experimental design, see Figure 3. N = 4 mice.

(B) Experimental design. Mice were intravenously infused with [U-¹³C] glycine for 2.5 h and tissue and serum glycine and acyl-glycine labeling were measured.

(C) Circulating acyl-glycines are made from circulating glycine. Mean ± s.e. N = 4 mice.

(D) Tissue phenylpropionyl-glycine labeling (normalized to circulating glycine labeling). Mean ± s.e. N = 4 mice.

(E) Tissue butyryl-glycine labeling (normalized to circulating glycine labeling). Mean ± s.e. N = 4 mice.

(F) Amino acid composition of algal protein and casein measured by acid hydrolysis. Mean ± s.e. N = 3 for algal protein and N = 1 for casein.

(G) Contribution of dietary algal protein versus casein to cecal amino acid carbon correlates poorly with relative amino acids abundances in algal protein versus casein.

5

Figure S5. Single exponential fit of newly synthesized fraction of microbial peptides over time. Related to Figure 4.

(A) Growth rate quantification using D₂O in vitro. Clostridium sporogenes or Bacteroides dorei were cultured in media with 5% D₂O, and bacteria growth over time was measured by either OD₆₀₀, or proteomics.

(B) Proteomics measured growth rate mirrors the growth rate measured by OD₆₀₀. Mean±s.e. N > 200 characteristic peptides for each time point.

(C) As in (B), for Bacteroides dorei.

(D) Different cellular compartments from the genus Bacteroides show similar labeling rate.

(E) As in (D), for genus Ruminococcus.

(F) Single exponential fit was applied to determine genus-level microbial turnover. Data are mean ± s.e. N = 5 mice.

6

Figure S6. Species level and correlation analysis of nutrient preferences across different gut bacterial genera. Related to Figure 5, 6.

(A) Carbon contribution of dietary inulin across bacterial species. Mean±s.e. N=4 mice.

(B) Carbon contribution of dietary algal protein across bacterial species. Mean±s.e. N=6 mice.

(C) Carbon contribution of circulating lactate across bacterial species. Mean±s.e. N=7 mice.

(D) Nitrogen contribution of circulatory urea across bacterial species. Mean±s.e. N=6 mice.

(E) Nitrogen contribution of dietary algal protein across bacterial species. Mean±s.e. N=6 mice.

(F) Carbon contribution of secreted host proteins across bacterial species. Mean±s.e. N=5 mice.

(G) Positive correlation between f_{genus←dietary proteins C} and f_{genus←dietary proteins N}.

(H) Positive correlation between f_{genus←inulin C} and f_{genus←urea N} in Firmicutes.

(I) Negative correlation between f_{genus←urea N} and f_{genus←dietary proteins N} in Firmicutes.

(J) Negative correlation between $\raisebox{1ex}{${\theta }_{night}$}\!\left/ \!\raisebox{-1ex}{${\theta }_{day}$}\right.$ and f_{genus←dietary proteins N}.

(K) Positive correlation between $\raisebox{1ex}{${\theta }_{night}$}\!\left/ \!\raisebox{-1ex}{${\theta }_{day}$}\right.$ and f_{genus←secreted proteins N}.

7

Figure S7. Host mucins are synthesized using circulating host amino acids and eventually contribute to fecal amino acids. Related to Figure 6.

(A) Experimental design. Mice were infused with ¹⁵N-lysine and ¹⁵N-arginine and mouse colon proteins were analyzed by proteomics.

(B) MUC2 labeling (multiple different MUC2 peptides). N = 2 mice for 12 h, 18 h, N = 3 mice for 36 h.

(C) As in (B), for MUC3.

(D) As in (B), for MUC13.

(E) Passage of circulating amino acids from serum into host colonic protein, and from there into host proteins and finally fecal free amino acids. Figure shows the enrichment of serum AA, colonic MUC2 proteins, host proteins in intestinal lumen and fecal AAs normalized to the serum AA enrichment at 36 h following ¹⁵N-lysine and ¹⁵N-arginine infusions. Mean±s.e., N = 2 mice for 12 h, 18 h, N = 3 mice for 36 h for serum and fecal AAs. N = 14 characteristic MUC2 peptides for 12 h, 18 h, N = 28 characteristic MUC2 peptides for 36 h. N > 50 characteristic peptides for host proteins in intestinal lumen for 12 h, 18 h and 36 h.

8

Table S1A. Microbiota metabolites concentration in portal and systemic circulation. Related to Table 1.

Mean, 95% confidence intervals, N = 5 mice. Renal, portal, hepatic refer to concentrations in those veins. Systemic = tail vein.

Table S1B. Microbiota-associated metabolite concentrations in tissues (ad lib fed state, ZT1). Related to Table 1.

Mean, 95% confidence interval, N = 5 mice. Red indicates metabolite is enriched in cecal content and colon, blue indicates the metabolite is enriched in kidney and yellow indicates enriched in liver.

9

Table S2A. Infusion parameters. Related to STAR Methods

Table S2B. Labeling data generated in the study. Related to Figure 1–2, STAR Methods

Table S2C. Proteomics data information. Related to Figure 4–6, STAR Methods

Table S2D. Cecal content amino acids labeling data. Related to Figure 4–6, STAR Methods

10

Table S3. Taxonomic assignment of the detected peptides in the study. Related to STAR Methods

11

Table S4. Diet composition. Related to STAR Methods

12

Table S5. Tabular results of the 16S rRNA gene amplicon sequencing of diet manipulation experiment. Related to Figure 5, 6, STAR Methods

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, Peptides, and Recombinant Proteins
DEUTERIUM OXIDE (D, 99.9%)	Cambridge Isotope Laboratories	Cat# DLM-4-PK
D-GLUCOSE (U-13C6, 99%)	Cambridge Isotope Laboratories	Cat# CLM-1396-PK
SODIUM L-LACTATE (13C3, 98%) 20% W/W in H2O	Cambridge Isotope Laboratories	Cat# CLM-1579-PK
L-GLUTAMINE (13C5, 99%)	Cambridge Isotope Laboratories	Cat# CLM-1822-H-PK
L-GLUTAMINE (ALPHA-15N, 98%)	Cambridge Isotope Laboratories	Cat# NLM-1016-PK
L-GLUTAMINE (AMIDE-15N, 98%+)	Cambridge Isotope Laboratories	Cat# NLM-557-PK
SODIUM D-3-HYDROXYBUTYRATE (13C4, 99%) 97% CHEMICAL PURITY	Cambridge Isotope Laboratories	Cat# CLM-3853-PK
LINOLEIC ACID, POTASSIUM SALT (U-13C18, 98%)	Cambridge Isotope Laboratories	Cat# CLM-8835-PK
OLEIC ACID, SODIUM SALT (U-13C18, 98%)	Cambridge Isotope Laboratories	Cat# CLM-8763-PK
SODIUM PALMITATE (U-13C16, 98%+)	Cambridge Isotope Laboratories	Cat# CLM-6059-PK
SODIUM ACETATE (1,2–13C2, 99%)	Cambridge Isotope Laboratories	Cat# CLM-440-PK
CITRIC ACID (13C6, 99%)	Cambridge Isotope Laboratories	Cat# CLM-9021-PK
SUCCINIC ACID (13C4, 99%)	Cambridge Isotope Laboratories	Cat# CLM-1571-PK
L-MALIC ACID (13C4, 99%)	Cambridge Isotope Laboratories	Cat# CLM-8065-PK
ALPHA-KETOGLUTARIC ACID, DISODIUM SALT (1,2,3,4–13C4, 99%)	Cambridge Isotope Laboratories	Cat# CLM-4442-PK
L-ALANINE (13C3, 99%)	Cambridge Isotope Laboratories	Cat# CLM-2184-H-PK
L-ALANINE (15N, 98%)	Cambridge Isotope Laboratories	Cat# NLM-454-PK
L-VALINE (13C5, 99%)	Cambridge Isotope Laboratories	Cat# CLM-2249-H-PK
L-VALINE (15N, 98%)	Cambridge Isotope Laboratories	Cat# NLM-316-PK
L-LEUCINE (13C6, 99%)	Cambridge Isotope Laboratories	Cat# CLM-2262-H-PK
L-LEUCINE (15N, 98%)	Cambridge Isotope Laboratories	Cat# NLM-142-PK
L-ISOLEUCINE (13C6, 99%)	Cambridge Isotope Laboratories	Cat# CLM-2248-H-PK
L-ISOLEUCINE (15N, 98%)	Cambridge Isotope Laboratories	Cat# NLM-292-PK
L-SERINE (13C3, 99%; 15N, 99%)	Cambridge Isotope Laboratories	Cat# CNLM-474-H-PK
GLYCINE (13C2, 97–99%)	Cambridge Isotope Laboratories	Cat# CLM-1017-PK
GLYCINE (13C2, 99%; 15N, 99%)	Cambridge Isotope Laboratories	Cat# CNLM-1673-H-PK
L-TYROSINE (13C9, 99%)	Cambridge Isotope Laboratories	Cat# CLM-2263-H-PK
UREA (15N2, 98%+)	Cambridge Isotope Laboratories	Cat# NLM-233-PK
AMMONIUM CHLORIDE (15N, 99%)	Cambridge Isotope Laboratories	Cat# NLM-467-PK
L-LYSINE:2HCL (ALPHA-15N, 98%)	Cambridge Isotope Laboratories	Cat# NLM-143-PK
L-ARGININE:HCL (ALPHA-15N, 98%+)	Cambridge Isotope Laboratories	Cat# NLM-1267-PK
ALGAL STARCH (U-13C, 98%+)	Cambridge Isotope Laboratories	Cat# CLM-1699-PK
INULIN (FROM CHICORY) (U-13C, 97%+) 97% CHEMICAL PURITY	Cambridge Isotope Laboratories	Cat# CLM-9181-PK
Algal crude protein extract-13C	Sigma-Aldrich	Cat# 642878
Algal crude protein extract-15N	Sigma-Aldrich	Cat# 586773
Algal crude protein extract-13C,15N	Sigma-Aldrich	Cat# 608254
Algal amino acid mixture-13C	Sigma-Aldrich	Cat# 426199
Algal fatty acid mixture-13C	Sigma-Aldrich	Cat# 487937
AZD 3965	AstraZeneca	Cat# AZD3965
Ampicillin	Sigma-Aldrich	Cat# A0166
Neomycin trisulfate salt hydrate	Sigma-Aldrich	Cat# N6386
Metronidazole	Sigma-Aldrich	Cat# M1547
Vancomycin hydrochloride from Streptomyces orientalis	Sigma-Aldrich	Cat# V1130
Trypsin	Promega	Cat# V5113
Lysyl endopeptidase R	Wako Chemicals USA	Cat# 12902541
Aspartame	Sigma-Aldrich	Cat# 47135
GAM Broth Modified	HyServe	Cat# 5433
LB Broth (Miller, Luria Broth)	Sigma	Cat# L3522
MRS Broth	Sigma	Cat# 69966
TSB (Tryptic Soy Broth)	Bacto	Cat# 211825
Experimental Models: Organisms/Strains
Mouse: C57BL/6	Charles River Laboratories	Cat #027
Strain: Bacteroides dorei (CL02T00C15)	BEI	#HM-717
Strain: Clostridium sporogenes (ATCC 15579)	ATCC	15579
Strain: Escherichia coli (ATCC 25922)	ATCC	25922
Strain: Lactobacillus reuteri (CF-48–34A)	BEI	#HM-102
Strain: Staphylococcus aureus subsp. Aureus Rosenbach	ATCC	29213
Software and Algorithms
El-MAVEN software	Elucidata	https://resources.elucidata.io/elmaven
AccuCor	GitHub	https://github.com/XiaoyangSu/AccuCor
PepMID	GitHub	https://github.com/xxing9703/pepMID_simul
MATLAB R2021b	MathWorks	N/A
Others
PicoLab Rodent Diet 20	LabDiet	Cat# 5053
20% Diet premix	Research Diets	Cat# D11112201Npx2i
20% Amino acids diet	Research Diets	Cat# A11112201Bi

• Gut microbiome feedstocks mapped by isotope tracing into bacteria-specific peptides

• Major contributors are dietary fiber and protein, and host lactate, urea, and mucins

• Microbiome composition shifts towards bacteria fed their preferred nutrients

• Microbial metabolites’ systemic levels reflect dietary precursors reaching microbiome