Profiling the metabolism of human cells by deep ¹³C labeling

Summary

Studying metabolic activities in living cells is crucial for understanding human metabolism, but facile methods for profiling metabolic activities in an unbiased, hypothesis-free manner are still lacking. To address this need, we here introduce the deep labeling method, which combines a custom ¹³C medium with high-resolution mass spectrometry. A proof-of-principle study on human cancer cells demonstrates that deep labeling can identify hundreds of endogenous metabolites as well as active and inactive pathways. For example, protein and nucleic acids were almost exclusively de novo synthesized, while lipids were partly derived from serum; synthesis of cysteine, carnitine and creatine was absent, suggesting metabolic dependencies; and branched-chain keto acids (BCKA) were formed and metabolized to short-chain acylcarnitines, but did not enter the TCA cycle. Remarkably, BCKA could substitute for essential amino acids to support growth. The deep labeling method may prove useful to map metabolic phenotypes across a range of cell types and conditions.

Graphical abstract

eTOC blurb

Grankvist et al. describe a simple, high-throughput method to catalogue endogenous metabolites and detect metabolic pathway activity in mammalian cells, using isotope tracing with a custom-designed ¹³C-labeled medium. This technique can be used to survey the metabolic state of a range of cell types and tissues in various experimental conditions.

Introduction

To understand the role of cellular metabolism in human health and disease, measuring the activity of enzymes or pathways in living human cells and tissues is central. Metabolism is highly cell type- and cell state-specific (Jain et al. 2012; Shlomi et al. 2008; Agren et al. 2012), suggesting opportunities for diagnosis and treatment of human disorders. Besides the well-known metabolic disorders of obesity, diabetes, and the metabolic syndrome, mutations in metabolic enzymes underlie a variety of genetic diseases, and metabolism is increasingly appreciated as a key mechanism in disorders of cell proliferation as well. In cancer biology, for example, metabolic enzymes associated with transformed cells may present new targets for drugs that aim to reduce cell proliferation (Vander Heiden 2011), and similarly, metabolic traits of specific immune cell types are being explored in the context of autoimmune disorders (Freitag et al. 2016). Systematic profiling of metabolic activities in human cells is therefore of fundamental importance for understanding the physiology of normal cells and their metabolic derangements in human disease (Hoppe 2012).

While there has been much progress in advanced measurement techniques for detailed studies of specific metabolic pathways (Buescher et al. 2015), a straightforward, high-throughput methodology for systematic, hypothesis-free assessment of metabolic activities in human cells is still lacking. Mass spectrometry (MS) or nucleic magnetic resonance (NMR) based metabolomics methods measure (relative) abundance of cellular metabolites, which can help detect differences in metabolic state, but such abundance data does not by itself inform on metabolic activities (Buescher et al. 2015). Measuring metabolite uptake and release can be done with reasonably high throughput, but provides only limited information on cellular metabolism (Allen et al. 2003; Jain et al. 2012). For a detailed assessment of intracellular metabolic activity in living cells, metabolic tracing with isotope-labeled nutrients is required. However, in mammalian cells, isotope labeling has mainly been used in a targeted, hypothesis-driven fashion, where selected pathways of interest are studied in depth using specific isotopic tracers in combination with sophisticated mathematical modeling (Niedenführ et al. 2015). Such tracer experiments are the gold standard for quantifying metabolic fluxes, but are challenging to perform and require detailed knowledge of the pathway of interest beforehand. On the other hand, several studies have used untargeted metabolomics methods to detect products of specific tracers, notably glucose and glutamine (Kluger et al. 2014; Huang et al. 2014; Capellades et al. 2016; Hiller et al. 2010); but in such cases, products of other medium nutrients are not detected, and only a subset of metabolic activities can be investigated. Indeed, the complement of metabolites synthesized by human cells is still not known (Viant et al. 2017). Comprehensive identification of endogenously synthesized compounds have to our knowledge been reported only in microorganisms that can grow on a single carbon source, where fully ¹³C cells can been generated, allowing global detection of endogenous metabolites by the IROA method (Qiu et al. 2016). However, since mammalian cells require a complex mixture of nutrients, including chemically undefined serum (Niklas et al. 2010), generating fully ¹³C mammalian cells is not feasible. In addition, full ¹³C labeling affords no information on the pathways used to synthesize the measured metabolites.

In this paper, we report a facile method, “deep labeling”, for hypothesis-free discovery of endogenous metabolites and metabolic activities in human cells from a single isotope tracing experiment. By culturing cells in a custom growth medium where the basic precursors of metabolic pathways are ¹³C, while vitamins and serum components are ¹²C, endogenously synthesized metabolites can be identified by incorporation of ¹³C atoms (Figure 1A). Simultaneously, presence of ¹²C atoms in a metabolite indicates incorporation of serum fat, protein or other exogenous compounds, and also serves to identify inactive metabolic pathways. This method is simple to perform, requires no computational modeling or a priori knowledge about metabolism, and is applicable to any cell type that can be maintained in culture conditions.

Figure 1. Cataloguing endogenously synthesized metabolites using deep labeling.

(A) Schematic of the deep labeling experiment strategy. (B) ¹³C enrichment of 1,319 LC-HRMS peaks with a unique predicted formula, in extract of cells grown in ¹²C medium (blue) or ¹³C medium (black). Individual enrichment values for three replicates are shown. (C-E) MIDs of pyruvate, acetyl-CoA and a novel pentose acid (ribonate; see text) in cells grown in ¹²C medium (blue) or ¹³C medium (black). (F-H) MIDs of pentose acid (ribonate), gulonate and ribose-5-phosphate/ribulose-5-phosphate in cells cultured with 1-¹³C₁-glucose. Ribose-5-phosphate and ribulose-5-phosphate could not be separated by chromatography. Error bars denote standard deviation. MI fractions not shown were negligible. See also Figure S1 and Figure S2.

Results

Cataloguing endogenously synthesized metabolites using deep labeling

To obtain wide coverage of endogenous metabolites and metabolic activities in human cells, we synthesized a growth medium where glucose and all amino acids were fully ¹³C, while choline and the various vitamins were ¹²C (Figure 1A). An atom-level computational analysis of a large human metabolic network model (Nilsson & Jain 2016) covering 342 metabolites and 659 reactions indicated that in these conditions, ¹³C isotopes should reach at least one atom in every metabolite (except the vitamins and choline itself), suggesting that the design is feasible. As a proof-of-principle study, we chose to characterize the HCT116 colorectal carcinoma cell line, a near-diploid, p53-positive epithelial-like cell type widely used as a model system (Brattain et al. 1981). We cultured these cells in the ¹³C medium for at least 6 population doublings to ensure that > 98 % of cellular material was newly synthesized, and analyzed polar cell extracts with liquid chromatography-high resolution mass spectrometry (LC-HRMS). We detected 3,299 metabolite ions that were reproducibly observed across replicate cultures but absent in blanks, which likely represent cellular (but not necessarily endogenously synthesized) metabolites. Since we were interested in mass isotopomer distributions, which can only be calculated when the number of carbons is known, we focused on 1,319 metabolite ions for which we could assign a unique sum formula based on exact mass (Table S1; STAR Methods text). About 70 % of these ions acquired some ¹³C atoms in the ¹³C cultures, but not in ¹²C control cultures, demonstrating that they represent endogenous metabolites (Figure 1B). We were able to annotate 958 ions (73 %) with putative HMDB identifiers (Wishart et al. 2009) based on accurate mass, and we confirmed 105 out of 112 selected annotations (94%) against pure standards, suggesting that error rates are quite low. The majority of metabolites were small (64 % had 10 carbons or less), but we also detected larger species of up to 64 carbons, most of which were likely phospholipids (Figure S1A). Most endogenous metabolites (68 %) contained nitrogen, 18 % were phosphorylated and 17 % contained sulfur, while 22 % were pure carbohydrate (C_xH_yO_z) compounds (Figure S1B). Endogenous metabolites exhibited a wide range of ¹³C enrichment (Figure 1B), with small metabolites such as pyruvate containing only nutrient-derived ¹³C atoms (Figure 1C), while more complex metabolites such as coenzyme A were mosaics of ¹²C vitamin moieties and nutrient-derived ¹³C atoms (Figure 1D), in agreement with known synthesis pathways in human cells.

One important application of the deep labeling methodology is identification of novel metabolites synthesized by various human cell types, given that only a fraction of human metabolites is known (Viant et al. 2017). As an example of discovery of endogenous metabolites, we observed a ¹³C₅-labeled ion in cell extracts (Figure 1E) with formula C₅H₁₀O₆, consistent with a pentose acid. This ion was not present in the medium (Figures S2A–S2C) and had no co-eluting compound (Figure S2D), indicating it was not an in-source fragment. The known pentose acids of biological origin are arabinonate and ribonate, which derive from the pentose sugars arabinose and ribose, respectively; and xylonate and lyxonate, which derive from the hexose acid gulonate. Culturing the HCT116 cells and two additional cell lines with a 1-¹³C₁-glucose (Figures1F–H) or 2-¹³C₁-glucose (Figures S2E–G) tracer resulted in clear ¹³C labeling of the metabolite ion (Figures 1F and S2E), demonstrating that it derives from glucose. Moreover, the mass isotopomer distribution (MID) of gulonate (Figures 1G and S2F) differed markedly from that of the pentose acid, rendering xylonate/lyxonate unlikely candidates, while the MID of ribose-5-phosphate/ribulose-5-phosphate (Figures 1H and S2G) was similar that of the pentose acid, suggesting that the compound is ribonate. A pure standard for ribonate eluted at the same retention time (Figure S2H) and gave an identical MS/MS spectrum (Figure S2I), supporting this conclusion. Ribonate has previously been observed in human plasma and was recently associated with risk for high-grade glioma (J. et al. 2017), but has to our knowledge not previously been shown to be synthesized by human cells, and its function in humans is unknown. These data exemplify that deep labeling experiments can uncover novel metabolites in human cells.

Profiling active and inactive metabolic pathways

Because deep labeling results in ¹³C incorporation into virtually any endogenously synthesized metabolite, it allows determining active and inactive metabolite pathways based on the observed mass isotopomers (Figure 1A). Metabolites in glycolysis, the pentose phosphate pathway, the TCA cycle, amino acid metabolism as well as de novo purine and pyrimidine synthesis all showed the expected ¹³C mass isotopomers, indicating that these central metabolic pathways are active (Table S1). Nucleobases were mostly ¹³C₄, indicating active turnover of nucleotide and deoxynucleotide pools, but little salvage of medium (¹³C₀) nucleotides (Figure 2A). S-adenosylmethionine (SAM)-dependent methylation was active, indicated by ¹³C₁₄ SAM and ¹³C₁₃ SAH (Figure 2B), while cystathionine was ¹³C₀, indicating that the “transsulfuration” pathway that converts methionine into cysteine and taurine was inactive (Figure 2B). Polyamines were both synthesized and degraded, indicated by ¹³C₁₀ methylthioadenosine and ¹³C₄ acetylputrescine, respectively (Figure 2B), consistent with the notion that polyamine metabolism oscillates over the cell cycle (Bettuzzi et al. 1999).

Figure 2. Profiling active and inactive metabolic pathways.

(A-E) MIDs for selected metabolites with carbon number in parentheses (left) are shown as stacked bar charts (middle) with numbers indicating mass isotopomers, together with simplified schemas of the corresponding pathways (right). White regions of bars charts denote the ¹³C₀ MI. Gray arrows denote inactive pathways. Numbers at the right side of the bars indicate ¹³C enrichment. Metabolites in parentheses were not measureable. Nonstandard abbreviation are hxan, hypoxanthine; thym, thymine; 5mta, 5-methiothioadenosine; aptrc, acetyl-putrescine; cyst, cystathionine; taur, taurine; tmabutn, trimethylammoniobutanoate; crn, carnitine; acrn, acetylcarnitine; orn, ornithine; creat, creatine; pcreat, phosphocreatine, gudac, guanidinoacetate; 5htrp, 5-hydroxytryptophan; kynr, kynurenine; ncam, nicotinamide. For enzymes indicated, see text.

Identifying inactive pathways in human cells is of great interest to predict dependency on metabolites provided by the cells’ environment, which could be exploited to prevent cell growth. In the HCT116 cells, we found that the fatty acid carrier carnitine was ¹³C₀ and hence not synthesized, although the initial steps of this pathway leading to ¹³C₇ trimethylammoniobutanoate were found to be active, suggesting that the carnitine-synthesizing BBOX1 enzyme was absent. Yet, acetylcarnitine was ¹³C₂ (Figure 2C), indicating that the cells utilize ¹²C carnitine from the serum for carrying ¹³C acyl groups, and hence may be dependent on extracellular carnitine. Similarly, creatine and its downstream metabolites phosphocreatine and creatinine were present but ¹³C₀, indicating that HCT116 cells lack the creatine synthesis pathway (Figure 2D).

We also observed synthesis of signaling molecules, bridging between metabolism and regulation. For example, the tryptophan-derived signaling metabolite kynurenine was de novo synthesized (¹³C₁₀) but also acquired from the medium (¹³C₀) (Figure 2E). Kynurenine is generated by some cancers and is thought to suppress immune cell responses (Platten et al. 2012). In agreement with a signaling role of this metabolite, kynurenine was not used for downstream synthesis of NAD, since a ¹³C₂₀ mass isotopomer of NAD was not observed; instead, cells incorporated ¹²C nicotinamide from the medium to form ¹³C₁₄ NAD (Figure 2E). We also detected synthesis of ¹³C₁₁ 5-hydroxytryptophan (Figure 2E), the initial metabolite in the branch of tryptophan metabolism that gives rise to serotonin and related hormones; as well as ¹³C₄ gamma-aminobutyrate (GABA; Table S1), a signaling metabolite that can inhibit proliferation in colon cancer cells (Ortega 2003). Taken together, these data show that deep labeling can reveal activity of a variety of metabolic pathways from a single experiment.

Carbon sources for oxidation and biomass synthesis

Deep labeling experiments can also to some extent identify the carbon sources used by cells for oxidation and biomass synthesis. In the HCT116 cells, the acetyl group of acetyl-CoA was 97 % ¹³C (Figure 3A), indicating that the contribution of medium-derived fatty acids or other unlabeled material to acetyl-CoA was very small. Alpha-ketoglutarate, the other major carbon source for the tricarboxylic acid (TCA) cycle, was also nearly completely ¹³C, indicating that most oxidized carbon was derived from the medium nutrients. Yet, citrate/isocitrate showed a prominent ¹²C₁mass isotopomer (Figure 3A), suggesting substantial incorporation of ¹²CO₂ into TCA carbon by isocitrate dehydrogenase (IDH), similar to previous reports (Metallo et al. 2011). Fumarate, malate and aspartate also exhibited notable ¹²C₁ mass isotopomers, indicating that ¹²CO₂ propagates to these metabolites as well, possibly via citrate lyase (Figure 3A). We note that these ¹²C isotopomers may be due to exchange fluxes at IDH, and do not demonstrate net carbon fixation into biomass. In addition, succinate exhibited a prominent ¹³C₀ isotopomer that was not present in other TCA cycle metabolites, suggesting a substantial, separate succinate pool derived from an unlabeled substrate. A possible candidate is peroxisomal oxidation of fatty acids, which may form succinate via the ACOT4 thioesterase (Westin et al. 2005).

Figure 3. Carbon sources for oxidation and biomass synthesis.

(A) MIDs for TCA cycle metabolites and a simplified schema of the TCA cycle, as in Figure 2. For enzymes indicated, see text. (B) MID for lysophosphatidylcholine 16:0. (C) Plot of ¹³C enrichment vs. number of carbons for phospholipids, for extracts of HCT116 cells grown in the ¹³C medium. (D) MID for phosphophatidylcholine 34:1 Notable mass isotopomers are indicated. Percentages denote sums of indicated MI fractions, representing labeled/unlabeled acyl chains (see text). The fraction of heavy chains for PC 34:1 is given by (0.55 + 2 × 0.28) / 2 = 0.55. (E) MID for glycerol-3-phosphocholine. Error bars denote standard deviation. See also Figure S3.

The source of protein and lipids, the major constituents of cell biomass, remains an open question for most human cell types (Hosios et al. 2016). In the present data set, cellular amino acid ¹³C enrichment was very similar to that of the ¹³C medium (Table S2), indicating that amino acids are fully derived from medium nutrients in HCT116 cells, and suggesting that proteins are endogenously synthesized rather than obtained from the serum. A proteomics analysis of ¹³C-labeled cells confirmed this prediction (Figure S3). In contrast, the observed MIDs of phospholipids indicated a mix of endogenous synthesis and consumption. For example, for LPC 16:0, we estimated from the observed MID that 63 % of the fatty acyl chains were de novo synthesized (Figure 3B); the remaining acyl chains are presumable obtained from serum fat. Interestingly, larger lipids generally had lower ¹³C enrichment, suggesting that longer acyl chains are preferentially obtained from the serum (Figure 3C). Moreover, the MID of phosphatidylcholine exhibited two distinct modes corresponding to either one (C₁₆/C₁₈) or two (C₃₄) de novo synthesized acyl chains (Figure 3D), indicating that consumed lipids are actively remodeled by cells. Consistent with this, the phospholipid catabolism product glycerol-3-phosphocholine was 30 % ¹³C₀ (Figure 3E), indicating turnover of the glycerol backbone from serum-derived lipids. It is clear from this data that de novo fatty synthesis contributes substantially to cellular lipids in cultured HCT116 cells, consistent with studies showing that knockdown of fatty acid synthase (FASN) reduces cell growth (Currie et al. 2013), but exogenous lipids are an important source as well.

Branched-chain keto acid metabolism

Branched-chain amino acid (BCAA) catabolism has recently been identified in several cancer types (Hattori et al. 2017; Mayers et al. 2014), but its function remains unclear. In HCT116 cells, we observed several metabolite ions with accurate mass and labeling pattern matching products of the branched-chain amino acids (BCAAs), including the corresponding branched-chain keto acids (BCKAs) and several branched-chain acylcarnitine species (Figure 4A), indicating that BCAA catabolism is active in HCT116 cells (Figure 4B). We confirmed the identity of the BCKA using pure standards (Figure S4), and further pursued these observations with individual isotope tracing experiments using either ¹³C₆-isoleucine, ¹³C₆-leucine, or ¹³C₅-valine. In each case, the amino acid was taken up by cells and transaminated to its corresponding keto acid, confirming that a branched-chain amino acid transaminase is active in HCT116 cells (Figure 4C). A total of 169 ions acquired at least 5% ¹³C enrichment from either of the BCAA tracers (Table S3). All three branched-chain acylcarnitine species were formed, and isoleucine additionally gave rise to ¹³C₃ propanoyl-carnitine, demonstrating that the complete isoleucine degradation pathway is active in these cells (Figure 4D). However, we observed no ¹³C labeling of TCA cycle metabolites, suggesting that the function of BCAA metabolism in these cells is unrelated to oxidation for energy. The branched-chain acylcarnitines have recently been shown to be precursors of fatty acid synthesis in human preadipocytes (Green et al. 2016; Crown et al. 2015; Hirosuke et al. 1994), and are associated with growth phenotypes in C.Elegans (Han et al. 2016), but have not previously been observed in cancer cells.

Figure 4. Branched-chain keto acid metabolism in HCT116 cells.

(A) MI fractions of BCKA and branched-chain acylcarnitine species from deep labeling tracing experiment. (B) Simplified schematic of the relevant BCAA catabolism pathways. Numbers in parentheses denote number of carbons, excluding the CoA or carnitine moieties. (C) MI fractions from isotope tracing experiments with either ¹³C₆-isoleucine (black), ¹³C₆-leucine (blue), or ¹³C₅-valine (red) for 48 h. The isoleucine and leucine peaks have similar retention times, resulting in some overlapping labeling pattern regarding both the amino acids and keto acids. (D) MIDs of isobutyryl-, propanoyl-, methylbutyryl- and isovaleryl-carnitine in extracts of cells cultured in (left to right) unlabeled medium, with ¹³C-isoleucine, ¹³C-leucine, or ¹³C-valine. Methylbutyryl- and isovaleryl-carnitine were not separable by LC-HRMS, and are therefore observed as a single peak that acquires labeling from both ¹³C-isoleucine and ¹³C-leucine. (E) MI fractions from cells labeled with ¹³C-isoleucine, ¹³C-leucine and ¹³C-valine simultaneously for 48 h, in presence (gray) or absence (black) of unlabeled BCKA. Error bars denote standard deviation. (F) Cell growth curves from cells cultured the absence or presence of the BCAA and BCKA for 72 h and 120 h. Data represent mean±SD. See also Figure S4.

Branched-chain keto acids are present in human plasma at 30—50 uM, comparable to the BCAA levels (Wishart et al. 2013), suggesting that utilization of BCKA could be advantageous to proliferating cancer cells as an alternative source of BCAA for protein synthesis, particularly in poorly vascularized tumors where nutrients are scarce. To investigate whether BCKA can be utilized by HCT116 cells, we performed isotope tracing using ¹³C BCAA in medium supplemented with ¹²C BCKA. We found that the fraction of labeled BCAA declined in the presence of unlabeled BCKA (Figure 4E), demonstrating that BCKA were indeed taken up by the cells and transaminated into BCAA. Finally, we found that BCKA could fully substitute for BCAA to support growth of HCT116 cells: at equal nutrient concentrations, cells grew at the same rate on keto acids as on the corresponding amino acids, demonstrating that BCKA are efficiently utilized (Figure 4F). Taken together, these data show that deep labeling experiments can reveal novel metabolic phenotypes of human cells.

Discussion

In this paper, we have introduced deep labeling as a simple experimental strategy for profiling cellular metabolism, and applied it to chart the metabolism of a human cancer cell line, as an example. The resulting data shows that a single deep labeling experiment allows detecting hundreds of metabolites, assessing the activity of metabolic pathways, and determining the source of biomass and metabolic fuels. We therefore anticipate that deep labeling could be broadly applicable to systematically study the metabolism of cultured human cells in a variety of experimental conditions.

In isotope tracing experiments, the precise choice of tracer is a key factor determining the information gained about pathway activity. Our medium design, where glucose and the amino acids are ¹³C while serum-derived nutrients are ¹²C, allows us to detect de novo synthesis of most cellular metabolites (Figure 1A). A drawback is that, since the resulting MIDs depend on the metabolic state of the cells, metabolite formulas must be identified from other sources, such as the HMDB database. In contrast, the Isotope Ratio Outlier Analysis (IROA) methodology (Qiu et al. 2016) allows identifying the number carbon number directly from mass isotopomer distributions; however, this requires complete ¹³C labeling of all biomass, which is only feasible in microorganisms that can grow on a single carbon source, and the resulting data yields no information about pathway activity. It must also be noted that our method cannot distinguish between usages of alternative pathways leading from a labeled nutrient to a labeled product, unless there is at least one unlabeled intermediate. To resolve such cases, metabolic flux analysis experiments are needed. Also, deep labeling requires that the cell culture reaches isotopic steady state, which means that cells need to be grown in the deep labeling for several cell doublings so that biomass is fully renewed (here we ensured at least 6 doubling, so that 1 – 2⁻⁶ = 98% of biomass is newly synthesized). Hence, our methodology is not suitable for studying rapid metabolic responses to acute stimuli.

In the HCT116 cells, we find that both endogenous fatty acid synthesis and serum fat contributes to lipids in cultured cancer cells, consistent with previous evidence that fatty acids can be obtained from serum (Kamphorst et al. 2013). The observation that endogenous fatty acyls are more common in shorter acyl chains could reflect de novo synthesis of palmitate (C₁₆) by fatty acid synthase (FASN), but lack of further fatty acid elongation; or may be related to acyl chain specificity of lipid remodeling enzymes. We note that, since culture medium has lower lipid content than plasma, cancer cells in vivo would be expected to have better access to exogenous lipids. On the other hand, our data indicates that cell protein is fully derived from medium nutrients in HCT116 cells, and not from serum protein, in contrast to previous studies in pancreatic and bladder carcinoma cells (Commisso et al. 2013; Kamphorst et al. 2015). Hence, protein consumption from the medium appears to be cell type-dependent.

A notable finding was that HCT116 cells transaminate branched-chain amino acids (BCAA) to BCKA, which were further metabolized by the mitochondrial oxidative pathway into branched-chain carnitines, but did not enter the TCA cycle. The BCKA are similar to (branched) short chain fatty acids, and their metabolism involves enzymes also utilized for fatty acid metabolism. BCKA metabolism has recently been shown to be important for adipocyte differentiation (Green et al. 2016), where BCKA carbon is incorporated into other fatty acids (Crown et al. 2015). We found no incorporation of BCKA into phospholipids, which indicates that BCKA are not a major contributor to lipid synthesis overall. In addition, adipocytes metabolized BCAA into citrate, while the HCT116 cells did not. While the ultimate function of BCKA metabolism remains unclear, the finding that BCKA can fully substitute for the corresponding amino acids suggests that this is a major metabolic pathway in HCT116 cells. In line with this, another recent study highlighted BCKA as a possible alternative nutrient to branched-chain amino acids in leukemia cells (Hattori et al. 2003). Also, glioma (Tönjes et al. 2013) and non-small lung carcinoma cells (Mayers et al. 2016) are thought to utilize BCAA as a nitrogen source, while pancreatic carcinoma cells did not (Mayers et al. 2016). Overall, it appears that BCAA/BCKA metabolism is a flexible system that can satisfy several metabolic needs of proliferating cancer cells. The diversity of metabolic modes seen in these cell types further underscores the need for facile methods for systematic profiling of metabolism in human cells. We hope that the deep labeling strategy will prove useful for this purpose.

STAR Methods text

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Roland Nilsson (roland.nilsson@ki.se).

Experimental Model and Subject Details

Cell culture conditions

HCT116 (human colorectal cancer, male), HeLa (human cervix carcinoma, female) and MCF7 (human breast adenocarcinoma, female) cells were cultured in custom made RPMI 1640 medium (Table S2) supplemented with 5 % dialyzed heat inactivated fetal bovine serum (FBS), 100U/ml penicillin-streptomycin (Life Technologies) and 1000x phenol red (Sigma-Aldrich) at 37 °C and 5 % CO₂. The FBS was dialyzed in SnakeSkin 10K MWCO dialysis tubing (Thermo Fisher Scientific #88245).

Method Details

Stable isotope tracing experiments

For deep labeling, cells were cultured in 10 cm tissue culture dishes for 120 h in a custom made RPMI 1640 labeled medium containing the following stable isotopes (Cambridge Isotope Laboratories, Inc): ¹³C₆-glucose (CLM-1396-0), ¹³C₆-arginine:HCl (CLM-2265-H), ¹³C₄-asparagine:H₂O (CLM-8699-H), ¹³C₄-aspartic acid (CLM-1801-H), ¹³C₃-cysteine (CLM-4320-H), ¹³C₅-glutamic acid (CLM-1800-H), ¹³C₅-glutamine (CLM-1822-H), ¹³C₂-glycine (CLM-1017-0), ¹³C₆-histidine:HCl:H₂O (CLM-2264-0), ¹³C₆-isoleucine (CLM-2248-H), ¹³C₆-leucine (CLM-2262-H), ¹³C₆-lysine:2HCl (CLM-2247-H), ¹³C₅-methionine (CLM-893-H), ¹³C₉-phenylalanine (CLM-2250-H), ¹³C₅-proline (CLM-2260-H), ¹³C₃-serine (CLM-1574-H), ¹³C₄-threonine (CLM-2261-0), ¹³C₁₁-tryptophan (CLM-4290-H), ¹³C₉-tyrosine (CLM-2263-H) and ¹³C₅-valine (CLM-2249-H).

To evaluate the labeling capability of the three BCAA and effects of the BCKA, cells were cultured in 6-well plates with custom made RPMI 1640 medium with the tracers; ¹³C₆-isoleucine, ¹³C₆-leucine and ¹³C₅-valine, individually or simultaneously w/wo co-culture of 0.3 mM per BCKA for 48 h.

For glucose isotope tracing experiment, HCT116, MCF7 and HeLa cells were cultured in 6-well plates with custom made RPMI 1640 media with the stabled isotope 1-¹³C₁-glucose or 2-¹³C₁-glucose for 48 h.

For proteomics analysis, HCT116 cells were cultured in custom made RPMI 1640 medium with the tracers ¹³C₆-arginine:HCl and ¹³C₆-lysine:2HCl for 120 h. Cells were collected by trypsinization and the pellet were washed with phosphate buffered saline prior analysis.

Common for all tracing experiments; cells were seeded at a density to achieve ~85 % confluence at the time of extraction. Separate wells with identical plating were utilized for confirmation of cell numbers. Three independent biological replicates were used for all cell culture experiments and conditions. As controls, additional cells were cultured in an unlabeled custom made RPMI 1640 medium (unlabeled nutrients from Sigma-Aldrich).

Metabolite extraction and mass spectrometry

For metabolomics studies, HCT116 cells were analyzed by liquid chromatography-high resolution mass spectrometry (LC-HRMS), as previously described. At the time of extraction, spent medium was collected and cells were rapidly washed twice with cold phosphate buffered saline to remove contaminating medium. We estimate that two rounds of washing reduces medium concentrations by a factor ~10⁶, bringing glucose (the most severe contaminant) from 11 mM to 11 uM. Metabolites were then extracted using 4.5 ml (10 cm tissue culture dishes) or 1 ml (6-well plates) of −80 °C pre-cooled HPLC grade methanol. The microplates were kept on dry ice and cell material was scraped in the cold methanol, collected, vortexed for 1 minute and stored in −80 °C until analysis.

For LC-HRMS analysis, cellular material was centrifuged at 10000 RPM for 10 min at 4 °C and the cell pellets were resuspended in solution followed by three freeze-thaw cycles, with alternating 37 °C and −80 °C liquid baths in 60 second intervals. To ensure complete cell lysis, samples were vortexed for 30 seconds and sonicated for two minutes. Samples were then dried down in vacuo using a vacuum concentrator and re-suspended in 25 μL of 80:20 methanol:water. Resuspended samples were mixed by aspiration, sonicated for 2 minutes, vortexed for 30 seconds, centrifuged at 14,000 g × 10 minutes at 4 °C and supernatants were transferred to LC-HRMS vials containing a 200 μl glass insert. An injection volume of 5 μL, which contained metabolic material from ~80,000 cells, was used for all cell samples. LC-MS/MS based metabolomics analysis was performed using a Thermo QExactive orbitrap mass spectrometer coupled to a Thermo Vanquish UPLC system. Chromatographic separation of metabolites was achieved using a Millipore (Sequant) Zic-pHILIC 2.1×150 mm 5 um column maintained at 25 °C. Compounds were eluted via a 19 minute linear gradient starting from 90:10 acetonitrile:20 mM ammonium bicarbonate to 45:55 acetonitrile:20 mM ammonium bicarbonate. A Thermo Q-Exactive orbitrap mass spectrometer was operated in positive and negative ion modes using a heated electrospray ionization (HESI) source at 35,000 resolution, 100 ms ion trap time for MS1 and 17,500 resolution, 50 ms ion trap time for MS2 collection. Data was collected over a mass range of m/z 67-1000, using a sheath gas flow rate of 40 units, auxillary gas flow rate of 20 units, sweep gas flow rate of 2 units, spray voltage of 3.5 and 2.5 kV for positive and negative ion modes (respectively), capillary inlet temperature of 275 °C, auxillary gas heater temperature of 350 °C and an S-lens RF level of 45. For MS2 collection, MS1 ions were isolated using a 1.0 m/z window and fragmented using a normalized collision energy of 35. Fragmented ions were placed on dynamic exclusion for 30 seconds before being allowed to be fragmented again.

LC-HRMS data processing

Primary Thermo .raw data files were processed using in-house software to detect all LC-HRMS peaks with an apex intensity of at least 70,000. Peak detection was done using a variation of the centWave algorithm (Tautenhahn et al. 2008), with the following modifications: (1) a chromatogram was defined by initializing with the highest intensity m/z peak, and iteratively adding any m/z peaks located within 10ppm and 3 scans from the peaks already included in the chromatogram; (2) after wavelet decomposition of a chromatogram, each wavelet component was scaled so as to not exceed the original chromatogram intensity at any point in the time domain, in order to reduce false positives from low-frequency components. Only metabolite ions (LC-MS peaks) that were reproducible detected in both unlabeled and labeled cells and reach > 5x the blank intensity were retained. Metabolites ions were matched to (putative) identifiers in the Human Metabolome Database (HMDB) ¹ by m/z with a tolerance of 5 ppm and considering H⁺, H⁻, Na⁺, K⁺, Cl⁻ and NH4⁺ adducts, while for ions with no matches in HMDB, any molecular formula resulting in a theoretical m/z within 5 ppm of the measured m/z was considered. Where possible, final metabolite identification was made based on agreement with the observed retention time of pure standards, as well as appearance of the expected isotopomers in ¹³C-labeled samples. For concentration estimation, the LC peak shape of the observed endogenous (labeled) compound was also verified in this manner.

Ions with a unique predicted formula were considered for mass isotopomer analysis (given the predicted number of carbons), resulting in a total of 16,949 isotopic peaks. We restricted ourselves to ions of with a unique formula to ensure that the carbon number is known, which is required to account for all possible mass isotopomers peaks and fully determine the MIDs (although it is possible to obtain relative mass isotopomer abundances without knowing the number of carbons). To avoid false isotopic peaks (due to contamination by other ions of similar m/z), we discarded mass isotopomers whose intensity ratio over the base (¹³C₀) mass isotopomers exceeded 3% of the expected ratio (from natural MI distribution). For metabolite of interest, false isotopomers were also excluded by examining the chromatographic shape of isotopic peaks, which must closely match that of the base isotopomer. The ¹³C enrichment E was defined for a compound with n carbons and MI fraction x₀,x₁,…x_n as

$$E=\frac{1}{n}\sum _{k=0}^n{\mathit{kx}}_k$$

Protein extraction and digestion for mass spectrometry analysis

Cell pellets were suspended in 8 M urea (Sigma-Aldrich), 50 mM ammonium bicarbonate and 10 % acetonitrile (ACN) resulting in protein yields 516-528 μg. The samples were sonicated using VibraCell probe (Sonics & Materials, inc) for 1 min with pulse 2/2, at 20 % amplitude, and sonicated in bath for 10 minutes. Ten μg of each sample were subjected to tryptic digestion following protein reduction in 5 mM DTT at 37°C for 45 min and alkylation in 15 mM iodoacetamide for 30 min at room temperature in the dark. Trypsin was added in an enzyme to protein ratio of 1:30 and digestion was carried out over night at 37°C.

Liquid chromatography tandem mass spectrometry

Tryptic peptides were cleaned on C18 StageTips (Thremo Scientific) and 25 % of the resulting peptide mixture was injected into a nano LC-1000 system on-line coupled to a QExactive Plus mass spectrometer (Thermo Scientific, Bremen, Germany). The chromatographic separation of the peptides was achieved using a 50 cm long C18 EASY spray column (Thermo Scientific, Bremen, Germany), with the following gradient: 4-26 % ACN in 180 min, 26-80 % ACN in 5 min and 80 % ACN for 8 min all at a flow rate of 300 nl/min. The MS acquisition method was comprised of one survey full scan ranging from m/z 300 to 1650, acquired with a resolution of R= 140,000 (at m/z 200), followed by data-dependent HCD fragmentations from maximum 16 most intense precursor ions with a charge state ≥ 2. The tandem mass scans were acquired with a resolution of R=17,500, targeting 2e5 ions, setting isolation width to m/z 4 and normalized collision energy to 26.

Cell growth experiments

Cells were cultured in custom made RPMI 1640 medium w/wo the BCAA (Sigma-Aldrich #I7403, #L8912, and #V0513) and w/wo the BCKA (Toronto Research Chemicals, #M326420, #M326425 and #K193500), both individually and in combination, for 72 h and 120 h. After exposure, cells were washed with phosphate buffered saline, trypsinized with Trypsin-EDTA (0.05 %) (Thermo Fisher Scientific) and suspended in growth medium. Cell numbers were measured in the cellsuspension with TC20 automated cell counter (Bio-Rad Laboratories, Inc.) and/or Scepter cell counter (Merck Millipore).

Quantification and Statistical Analysis

Mathematica scripts were used for calculations of peak areas and building plots. GraphPad Prism version 7 software was used for bar charts.

Data and Software Availability

Mass spectrometry data presented in this study is available from www.nilssonlab.se.

Additional Resources

N/A

Supplementary Material

Table S1.

Related to Figure 1. Peak list of 1,319 metabolites from deep labeling tracing experiment

Table S2.

Related to Figure 1 and STAR methods text. Custom made RPMI 1640 medium formulation

Table S3.

Related to Figure 4. Peak list from untargeted analysis

Key Resources Table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Bacterial and Virus Strains
Biological Samples
Chemicals, Peptides, and Recombinant Proteins
Methanol (JT Baker), HPLC grade	VWR	Cat#BAKR8402.250 0
Ammonium Bicarbonate	Honeywell	Cat#40867-50G
LCMS Grade Water	Optima	Cat#W64
LCMS Grade Methanol	Honeywell	Cat#BJLC230-4
LCMS Grade Acetonitrile	Honeywell	Cat#BJLC015-4
8M Urea	Sigma-Aldrich	Cat#U4883
RPMI 1640 vitamins solution 100x	Sigma-Aldrich	Cat#R7256
Trypsin-EDTA (0.05%), phenol red	Gibco	Cat#25300062
PBS, pH 7.4	Gibco	Cat#10010056
Fetal bovine serum, qualified, heat inactivated US origin	Gibco	Cat#16140071
Penicillin-Streptomycin (10,000 U/mL)	Gibco	Cat#15140122
Phenol red solution	Sigma-Aldrich	Cat#P0290
D-Glucose (dextrose)	Sigma-Aldrich	Cat#D9434
Glycine	Sigma-Aldrich	Cat#50046
L-Arginine	Sigma-Aldrich	Cat#A8094
L-Asparagine	Sigma-Aldrich	Cat#A4159
L-Aspartic acid	Sigma-Aldrich	Cat#A7219
L-Cysteine	Sigma-Aldrich	Cat#C7352
L-Glutamic acid	Sigma-Aldrich	Cat#G8415
L-Glutamine	Sigma-Aldrich	Cat#G8540
L-Histidine	Sigma-Aldrich	Cat#H6034
trans-4-Hydroxy-L-proline	Sigma-Aldrich	Cat#H54409
L-Isoleucine	Sigma-Aldrich	Cat#I7403
L-Leucine	Sigma-Aldrich	Cat#L8912
L-Lysine monohydrochloride	Sigma-Aldrich	Cat#L8662
L-Methionine	Sigma-Aldrich	Cat#M3508
L-Phenylalanine	Sigma-Aldrich	Cat#P5482
L-Proline	Sigma-Aldrich	Cat#P5607
L-Serine	Sigma-Aldrich	Cat#S4311
L-Threonine	Sigma-Aldrich	Cat#T8441
L-Tryptophan	Sigma-Aldrich	Cat#T8941
L-Tyrosine	Sigma-Aldrich	Cat#T8566
L-Valine	Sigma-Aldrich	Cat#V0513
Calcium nitrate tetrahydrate	Sigma-Aldrich	Cat#C2786
Sodium phosphate dibasic	Sigma-Aldrich	Cat#S0876
Magnesium sulfate	Sigma-Aldrich	Cat#M2643
Potassium chloride	Sigma-Aldrich	Cat#P5405
Sodium bicarbonate	Sigma-Aldrich	Cat#S5761
Sodium chloride	Sigma-Aldrich	Cat#S7653
L-Glutathione reduced	Sigma-Aldrich	Cat#G6013
4-Methyl-2-oxovaleric Acid Sodium Salt (4mop)	Toronto Research Chemicals	Cat#M326425
α-Keto Isovaleric Acid Sodium salt (3mob)	Toronto Research Chemicals	Cat#K193500
3-Methyl-2-oxovaleric Acid Sodium Salt (3mop)	Toronto Research Chemicals	Cat#M326420
D-Glucose (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-1396-0
Glycine (U-13C2, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-1017-0
L-Arginine (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2265-H
L-Asparagine (U-13C4, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-8699-H
L-Aspartic acid (U-13C4, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-1801-H
L-Cysteine (U-13C3, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-4320-H
L-Glutamic acid (U-13C5, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-1800-H
L-Glutamine (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-1822-H
L-Histidine (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2264-0
L-Isoleucine (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2248-H
L-Leucine (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2262-H
L-Lysine hydrochloride (U-13C6, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2247-H
L-Methionine (U-13C5, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-893-H
L-Phenylalanine (U-13C9, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2250-H
L-Proline (U-13C5, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2260-H
L-Serine (U-13C3, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-1574-H
L-Threonine (U-13C4, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2261-0
L-Tryptophan (U-13C11,99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-4290-H
L-Tyrosine (U-13C9, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2263-H
L-Valine (U-13C5, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-2249-H
D-Glucose (2-13C, 99%)	Cambridge Isotope Laboratories, Inc	Cat#CLM-746-PK
D-Glucose (1-13C)	Omicron Biochem. Inc.	Cat#GLC-018
Critical Commercial Assays
Deposited Data
Experimental Models: Cell Lines
HCT116	ATCC	Cat#ATCC® CCL-247™
MCF7	ATCC	Cat#ATCC® HTB-22™
HeLa	ECACC	Cat#93021013
Experimental Models: Organisms/Strains
Oligonucleotides
Recombinant DNA
Software and Algorithms
mzAccess (http://www.mzaccess.org/)	(Lyutvinskiy et al. 2017)	N/A
Mathematica v10	Wolfram Research	N/A
GraphPad Prism version 7 software	GraphPad Software	N/A
Other
SnakeSkin™ Dialysis Tubing, 3.5K MWCO, 35 mm	Thermo Fisher Scientific	Cat#88244
Cell scraper with two-position blade, scraper length: 25 cm	Sarstedt	Cat#83.1830
Cell culture dish 10 cm	Sarstedt	Cat#83.1830
Cell culture plates, 6-well	Sarstedt	Cat#83.3920

Highlights.

• Rich data on endogenous metabolites and active pathways from a single cell culture

• In HCT116 cancer cells, hundreds of synthesized metabolites were unknown

• Cells scavenge nutrients from serum and lack central synthesis pathways

• Branched-chain keto acids were metabolized and could substitute for amino acids

Significance.

As cellular metabolism is increasingly recognized as a key factor for a range of human disorders, charting the metabolic activity of various human cell types is of great importance. Measuring activity of metabolic enzymes and pathways in living cells or tissues is challenging, and there is currently no systematic method for unbiased profiling of metabolism in mammalian cells. In this paper, we propose such a method, based on a custom-designed ¹³C-labeled growth medium, and demonstrate that a single experiment using less than 10⁶ cells provides rich data on endogenous metabolites, including hundreds of still unknown species, and in many cases also allows identifying active / inactive metabolic pathways. We believe that this method will be valuable for profiling the metabolism of a variety of cell systems in an unbiased manner. Such experiments are needed to catalogue and understand the metabolic phenotypes of human cells in normal and pathological conditions.