|Home | About | Journals | Submit | Contact Us | Français|
This protocol provides a method for quantitating the intracellular concentrations of endogenous metabolites in cultured cells. The cells are grown in stable isotope-labeled media to near-complete isotopic enrichment and then extracted in organic solvent containing unlabeled internal standards in known concentrations. The ratio of endogenous metabolite to internal standard in the extract is determined using mass spectrometry (MS). The product of this ratio and the unlabeled standard amount equals the amount of endogenous metabolite present in the cells. The cellular concentration of the metabolite can then be calculated on the basis of intracellular volume of the extracted cells. The protocol is exemplified using Escherichia coli and primary human fibroblasts fed uniformly with 13C-labeled carbon sources, with detection of 13C-assimilation by liquid chromatography–tandem MS. It enables absolute quantitation of several dozen metabolites over ~1 week of work.
This protocol describes methodology for quantitating the concentrations of endogenous metabolites in cultured cells using MS, with the example described herein using liquid chromatography (LC)–tandem MS on a triple quadrupole mass spectrometer. Implementation of the protocol involves growing cells in a medium containing nutrients labeled with stable isotopes and then quenching metabolism and extracting metabolites in a solution spiked with unlabeled standards. Concentrations of metabolites in the cells are then calculated using the ratio of the labeled extracted metabolite to the unlabeled internal standard (see Fig. 1). This method has been used to investigate metabolites including glycolytic and tricarboxylic acid cycle intermediates, amino acids, nucleotides and folates from cells including E. coli, Salmonella enterica, yeast and human fibroblasts1-6 (see also B.D.B. and J.D.R., unpublished data).
Many metabolites turn over very rapidly; thus, correctly measuring intracellular metabolite concentrations requires the ability to sample cells quickly. Otherwise, the measured levels will reflect the metabolic state induced by the handling steps leading up to quenching of metabolism, rather than normal cellular physiology. For adherent cells such as human fibroblasts, metabolism can be quenched with minimal perturbation of the culture simply via quick aspiration of medium and addition of cold organic solvent. The solvent addition stops metabolism (initially due to the temperature drop and subsequently by denaturing enzymes) and simultaneously initiates the extraction process by disrupting the cell membrane.
For nonadherent cells, such as E. coli or Saccharomyces cerevisiae, filter culture7 allows similarly rapid and nondisruptive metabolic quenching. In this method, cells are grown on a membrane filter sitting on top of an agarose plate loaded with media (Fig. 1). The cells are fed by nutrient diffusion from the underlying media up through the filter. For a given medium composition, the growth rate of the cells is comparable with those grown in standard batch culture3. In addition, the cells show comparable responses to nutrient deprivation7,8. To quench metabolism and extract the intracellular metabolites, the membrane is simply moved from the agarose plate to a dish containing cold organic solvent. The move from plate to extraction solution can be done in ~1 s, during which there is little perturbation, as there are still extracellular nutrients contained in the liquid absorbed in the filter. Nevertheless, the cells are efficiently separated from the vast majority of their surrounding media without the delays and disruptions associated with filtering or pelleting a liquid culture.
Alternatives to filter culture include growing cells in standard liquid culture and separating them from their surrounding media by filtration or centrifugation. These approaches are generally acceptable for metabolites that turn over less rapidly. For example, we find that fast filtration (time scale approximately 5–10 s) gives adequate results for metabolites in steady-state cultures of yeast, but not for high flux metabolites like ATP or glutamine in E. coli. Other alternative approaches are described elsewhere9,10. Approaches that involve washing of cells (e.g., with water or PBS) before quenching metabolism, despite facilitating subsequent analytical steps, generally result in unacceptable alterations of cellular metabolite composition and should be avoided9,11.
A spectrum of solvent mixtures for metabolism quenching and metabolite extraction have been tested1,12,13, and one may select an appropriate extraction solvent mix according to experimental needs. For extraction of filter cultures, we recommend a ratio of 40:40:20 acetonitrile/methanol/water in general, and this system with the addition of formic acid to a final concentration of 0.1 M for studies involving nucleotide triphosphates in E. coli. This recommendation is based on experiments showing that, for E. coli, there is a marked decrease in nucleotide triphosphate extraction (and increased conversion to less phosphorylated species) unless both acetonitrile and formic acid are added to the extraction mixture1,14. Methanol extraction is recommended for extracting amino acids in E. coli (100% methanol for the first extraction and 80:20 methanol/water for the subsequent two extractions) and for extracting human fibroblasts (80:20 methanol/water for all three extractions). Extraction using these methods captures a mixture of free and (perhaps with lesser efficiency) macromolecule-bound intracellular metabolites. Accordingly, the present method is insufficient to differentiate free from bound metabolite pools.
Once an extract is obtained, it can be analyzed using a wide variety of chromatography-MS procedures. For example, separation may be achieved by gas chromatography15, capillary electrophoresis16 or LC17, including variants of LC such as capillary monolithic chromatography18-20 and ultra performance LC21-24. In conjunction with capillary electrophoresis or LC, ionization may be achieved using electrospray ionization (ESI), atmospheric pressure chemical ionization or atmospheric pressure photoionization25. We typically use LC coupled to ESI. An advantage of LC is its applicability to a broad diversity of analytes without the need for derivatization. A key advantage of ESI is its efficiency in converting charged compounds into gas phase ions. An important limitation of ESI is that it is a competitive process. Thus, an abundant ion can suppress the signals of less prevalent components (‘ion suppression’). Accordingly, the quality of LC separation is critical to the sensitivity of MS detection. Ion suppression can also lead to quantitative artifacts, as the signal of a compound in a complex mixture will be less than the signal of the same concentration of a pure standard. Such quantitative artifacts are corrected for by mass ratio-based analysis of the type described here, as both the labeled and unlabeled forms of the compound will be subject to quantitatively identical ion suppression.
There are a number of mass spectrometer options for analysis of the gas phase ions produced by LC-ESI or other chromatography–ionization approaches. These include quadrupole (most useful when arranged in series in a triple quadrupole instrument), ion trap, time-of-flight (TOF), Fourier transform ion cyclotron resonance and Orbitrap26. Different types of analyzer can also be combined to form a hybrid mass spectrometer, such as a quadrupole-TOF instrument. Any of these mass spectrometer types can in concept be used to conduct the experiments described herein, with triple quadrupole MS the approach used by us to date and TOF representing an economical alternative.
For the present purposes, a triple quadrupole instrument is run in multiple reaction monitoring (MRM) mode. Instrumentally, three quadrupoles are arranged in series. The first selects a parent ion m/z of interest, the second fragments the parent ion and the third isolates a product ion m/z predetermined on the basis of preliminary experiments using purified standard of the metabolite of interest. To analyze multiple metabolites and isotope-labeling states, this process is repeated in a cyclic manner, with a dwell time of ~50 ms for each compound. The MRM approach benefits from excellent sensitivity and broad linear dynamic range. Its main disadvantages are the need to prespecify each compound of interest and a limit to the number of metabolites/labeling states that can be reliably analyzed in a given chromatography interval. It is most readily used to measure only fully labeled and fully unlabeled compounds. Partially labeled forms can also be measured, but it takes a longer scan time, as different locations of isotope label may result in different product ion masses (i.e., the labeled carbon may or may not end up in the product ion, depending on its position in the parent). An ion-trap instrument can be used to conduct MRM, but with less sensitivity and dynamic range.
TOF involves a conceptually simpler MS approach: every ~1 s, a readout is taken of whatever ions are present. Selectivity, which is obtained in the MRM approach via the MS/MS, is obtained in TOF by mass accuracy (i.e., separation of compounds of the same nominal m/z by their exact mass). Full-scan methods like TOF enable monitoring of all isotopic forms of a metabolite. The main disadvantages of TOF relative to MRM are somewhat reduced sensitivity and dynamic range. In addition, information about the location of the isotope label within a partially labeled species is not available. Orbitrap and Fourier transform ion cyclotron resonance and provide similar data to TOF but are currently more expensive.
Irrespective of the LC-MS approach that is used, it is ultimately necessary to quantify the intensities of mass-specific chromatographic peaks. Software for peak intensity quantitation is included in most commercial LC-MS packages and generally allows quantitation on the basis of peak area or peak height at the user’s discretion. Both give equivalent results for Gaussian peaks. Typically, we quantify on the basis of peak height, as it is somewhat less sensitive than peak area to baseline fluctuations. Alternative metrics, such as the mean height of the top few points in an MRM peak, are also equally valid and may offer subtle improvements in reproducibility.
A detailed flowchart of the experimental workflow is illustrated in Figure 2. Cells are grown in isotopically labeled medium to the desired density, at which point the cells are quickly transferred to cold organic solvent spiked with standards. The presence of the internal standards in the quenching/extraction solution is critical, as the internal standards experience similar conditions to the cellular metabolites, especially to the free intracellular metabolites, thereby controlling for degradation and adsorption during the extraction process. Standards are spiked into the extraction solution only during the first step, with unspiked solvent used for subsequent steps. The isotopic standards in the resulting final extract control for ion suppression and other sources of LC-MS variability. Best results are obtained if standard concentrations are close to the concentration of the metabolite in the extract. Intracellular concentrations are then determined by the ratio of the peak sizes of the cellular metabolite to the standard, as originally described by J.J. Heijnen6 and colleagues.
If labeling of the cellular compound of interest is substantially complete, then analysis is straightforward. Figure 3 shows a hypothetical example involving a 4-carbon compound, such as malate or aspartate, and labeling with uniformly 13C-glucose. For simplicity, the example assumes that the quantity of spiked unlabeled compound (blue) exactly equals that of endogenous compound (yellow). Figure 3a exemplifies the ideal case of complete labeling. The fully labeled and fully unlabeled peaks are almost equal, except that a small amount (~4%) of the spiked unlabeled compound has a single 13C atom due to the natural abundance of 13C. The ratio R, defined as the intensity of the fully labeled peak divided by the fully unlabeled peak (corrected for losses of the fully unlabeled form due to the natural abundance of 13C), is 1 (see Step 8).
We find that extended growth in uniformly 13C-labeled glucose will lead to complete labeling of many metabolites in E. coli or yeast (as shown in Fig. 3a)4,5,17. For many other metabolites, however, partially labeled forms are also found, primarily due to assimilation of unlabeled carbonate (e.g., by the reactions generating oxaloacetate from pyruvate or phosphoenolpyruvate)5. This results in labeling patterns like the one shown schematically in Figure 3b. In this case, even though the spiked standard (blue) and endogenous compound (yellow) are equally abundant, the ratio R is less than 1. This necessitates a correction factor (termed L) reflecting that fraction of the endogenous compound in the fully labeled form. If data on partially labeled forms are available, this factor can be determined as the fraction of the labeled form that is fully labeled (see Step 11A). If MRMs that provide information solely on the fully labeled and fully unlabeled forms are being used, then L can be calculated by comparing the peak of the fully labeled form (in cells grown on labeled nutrient) with the fully unlabeled form (in cells grown on unlabeled nutrient) (see Step 11B).
For mammalian cells, calculations are further complicated by persistence of fully unlabeled endogenous forms of certain compounds. For example, feeding with both U-13C-glucose and U-13C-glutamine over ~5 d and 2+ doublings leads to incorporation of 13C into most metabolites in cultured mammalian fibroblasts, but a nontrivial amount of certain metabolites persist in the fully unlabeled form. This results in labeling patterns like the one shown schematically in Figure 3c, in which the peak of the spiked, unlabeled compound is ‘contaminated’ with unlabeled endogenous compound. The extent of contamination can be estimated by measuring the relative peak intensities of the fully unlabeled and fully labeled forms in cells fed with labeled nutrient and extracted in the absence of any spiked standard. We term this ratio Z. The product R × Z represents the fraction of the unlabeled signal observed in the experiment for determining R, which arose from unlabeled endogenous compound (i.e., the fraction of the peak with zero labeled carbons in Fig. 3 that is ‘yellow’). A large value of R × Z potentially leads to large error in the calculation of the metabolite concentration, and as such, we suggest repeating any measurement for which R × Z > 0.25. If Z is small, this correction will generally not be necessary, as the error caused by this small amount of unlabeled metabolite will not be significant when compared with the overall error. For example, in Figure 3a and b, and in general in experiments with microbes, Z is ~0.
All calculations in the procedure are done using geometric means, as is appropriate when dealing with ratios. In the protocol, error is calculated as the variance of the logarithm until the final determination of a concentration confidence interval (which is nonsymmetrical, being larger in the upward than downward direction). A step-by-step guide for calculating the errors of the determined concentrations is provided. Typically, we recommend conducting ~4 replicates of each measurement, which generally results in 95% confidence intervals on metabolite concentrations spanning an ~2- to 4-fold range. Tighter confidence intervals are generally found for stable, abundant compounds and broader ranges for less stable or low-abundance compounds. For compounds present in cells in concentrations <100 μM, quantitation is often difficult, unless the compound ionizes well. Similarly, quantitation of compounds with intrinsic decay half-times of <6 h is difficult, unless suitable preservatives can be identified (e.g., ascorbate for folates27).
In the initial application of this protocol, it may be useful to begin by replicating literature results (e.g., if working with E. coli, regarding glutamate, glutamine and adenosine nucleotide concentrations1,8). Once the protocol is up and running, it can be a useful tool for addressing basic questions such as the following: which are the most abundant metabolite species in cells? Along specific pathways? How do intracellular substrate concentrations compare to the Michaelis constants (Km) of enzymes; that is, which enzymes are saturated or not saturated with substrates in live cells? Answers to these questions are essential for the quantitative analysis of metabolism and its regulation, and ultimately for the rational manipulation of metabolism to meet medical and bioengineering objectives.
Before initiation of an experiment, cells should be handled as per typical laboratory protocols tailored to the cell type of interest. To initiate this protocol, a starter culture is required for microbes and 106–107 cells in culture for mammalian cells.
Wash the Ultrapure agarose three times with HPLC-grade water to remove trace impurities. For 30 g agarose, use 1 liter of water for each wash. For each wash, stir the agarose–water mixture for 10 min and allow to settle for ~1 h. Aspirate the water with care to avoid loss of agarose. The resulting washed agarose can be used to make minimal media plates with 1.5% agarose.
Combine sterile salts, U-13C glucose, water and agarose according to your media recipe. (The exact composition of the complete minimal media we use is as follows: KH2PO4 4.7 g liter−1, K2HPO4 13.5 g liter−1, K2SO4 1 g liter−1, MgSO4 · 7 H2O 0.1 g liter−1, NH4Cl 10 mM, glucose 0.4%; make the media with U-13C-glucose and separately with unlabeled glucose.)
Combine minimal liquid medium (with U-13C-labeled glucose when appropriate) with 1.5% agarose. Autoclave and pour into the sterile Petri dishes to make plates. Use 15–25 ml of agarose-medium mixture per 10-cm plate.
Add U-13C-labeled glucose and glutamine (and, separately, unlabeled glucose and glutamine) to the DMEM media without glucose and glutamine, to obtain complete labeled and unlabeled DMEM media. Final glucose and glutamine concentrations should be 4.5 g liter−1 and 584 mg liter−1, respectively.
Different groups of metabolites are extracted with different efficiency depending upon the extraction solution mixture1. Choice of extraction solution should be made according to which metabolites are of the greatest interest. Among the solution systems that we have tested for extracting filter cultures, 40:40:20 acetonitrile/methanol/water solvent system works the best for extracting filter cultures in general; addition of formic acid to a final concentration of 0.1 M provides additional protection of nucleotide triphosphates against degradation1. Methanol (100% methanol on dry ice for the first round, 80:20 methanol/water at 4 °C for the two subsequent rounds) extracts amino acids efficiently while extracting fewer other components than acetonitrile/methanol/water; it is accordingly preferred for studies focused solely on amino acids. For extracting human fibroblasts, we have obtained adequate results with 80:20 methanol/water for all three extractions. Pending more definitive studies, we recommend this solvent mixture for them.
Standards of each metabolite for which quantification is desired should be made to a high concentration (typically 0.1–1.0 mg ml−1) in the solvent to be used for extraction. If you are using an acidic extraction solution, omit the use of the acid while making up the standard solutions. Stock solutions should be stored at −80 °C after preparation.
Solvent A: 20 mM ammonium acetate + 20 mM ammonium hydroxide in 95:5 water/acetonitrile, pH 9.45; Solvent B: acetonitrile. Note that this is our mobile phase of choice when working with aminopropyl column in hydrophilic interaction chromatography mode17; there are many other chromatography choices available28-30. Many of these have important advantages relative to the aminopropyl approach for certain classes of compounds.
The components are listed below (for filter cultures), as shown in Figure 4:
Hydrophilic interaction chromatography is performed on a 2-mm inner diameter column packed with 5-μm aminopropyl resin to 250 mm in length, using an LC-10A HPLC system (Shimadzu). The column is maintained at 15 °C with a solvent flow rate of 0.15 ml min−1, and the gradients are as follows: t = 0, 85% B; t = 15 min, 0% B; t = 28 min, 0% B; t = 30 min, 85% B; t = 40 min, 85% B. Other chromatographic approaches and/or HPLC systems can be used depending on their availability and the metabolites of interest.
A Finnigan TSQ Quantum Ultra triple quadrupole mass spectrometer (Thermo Electron Corporation) is run in MRM mode and coupled to the HPLC via ESI. ESI spray voltage is 3,200 V in positive ionization mode and 3,000 V in negative ionization mode. Nitrogen is used as sheath gas at 30 psi and as the auxiliary gas at 10 psi, and argon as the collision gas at 1.5 mTorr, with a capillary temperature of 325 °C. The mass spectrometer and LC system with autosampler are shown in Figure 5.
Reactions should be optimized for metabolites of interest using standards before the quantification experiment. A list of reactions used in our experiments has been published previously17. Optimization of the product ion and collision energy for a given unlabeled metabolite is achieved by infusing purified compound standard into a triple quadrupole mass spectrometer. Collision energy should be identical for labeled and unlabeled forms. For 13C labeling, the parent ion mass should be increased by the number of carbon atoms in the metabolite, and the product ion mass by the number of carbon atoms in the product ion (product ion structures can be obtained from the literature for common metabolites, or otherwise estimated on the basis of common routes of fragmentation and confirmed experimentally by MS/MS of the labeled forms). For partially labeled forms, more than one product ion mass may be possible for each parent ion mass. The different product ion masses arise from labeling at different positions within the parent. As an example, consider the possibilities for a 4 carbon compound that gives a 3 carbon product ion (e.g., aspartate): with 0 × 13C in the parent, there cannot be 13C in the product ion; with 1 × 13C in the parent, there can be 0 or 1 × 13C in the product ion; with 2 × 13C in the parent, there can be 1 or 2 × 13C in the product ion; with 3 × 13C in the parent, there can be 2 or 3 × 13C in the product ion; with 4 × 13C, there must be 3 × 13C in the product ion.
It is important to note than many of these steps have times dependent upon the growth rate of the cell culture. As such, these times may vary substantially with your cultures.
Check the LC-MS method to determine that settings are correct to detect compounds of interest. Concentrate sample to increase concentration.
Allow cells to grow for more time in the labeled medium. Some cell lines may require additional labeled nutrients in addition to glucose and glutamine. If it is not possible to get near-complete isotopic labeling of a metabolite, the error may be reduced by increasing the concentration of standards used in determining R. This will decrease the value of R (by increasing its denominator), leading to smaller value of RavgZavg, decreasing the contribution of the error from (1 − RavgZavg) to the total error. We recommend keeping the value of RavgZavg below 0.25.
In cases where unlabeled carbon can enter metabolism via only limited routes (e.g., bicarbonate), it is not necessary to determine L for every metabolite independently, as L must be equal for metabolites with the same carbon skeleton (e.g., glutamine, glutamate; NAD+, NADH; ADP, ATP) or containing the same carbon atoms (e.g., carbamoyl aspartate, dihydroorotate). For metabolites produced by unidirectional condensation of two substrates, Lproduct = Lsubstrate1 × Lsubstrate2 (e.g., Lorotidine-5′-phosphate = LPRPP × Lorotate). These relationships eliminate the need to determine L for each species individually. They can also be used to reduce error in estimating L, by averaging across compounds for which L must be equal (e.g., NAD+, NADH).
Here, we go through one example of the calculation of the intracellular concentration of uridine-5′-triphosphate (UTP) in E. coli. Following Steps 1–7, the following peak heights for the 13C-labeled endogenous UTP and the 12C UTP standard were determined for four replicates:
|Sample no. 1||Sample no. 2||Sample no. 3||Sample no. 4|
Representative chromatograms (from Sample no. 2) are shown in Figure 8.
Ri is determined for each sample, with no. of carbons = 9, as described in Step 8:
|Sample no. 1||Sample no. 2||Sample no. 3||Sample no. 4|
Ravg is calculated from the four values above, according to Step 9:
, the variance of R, is calculated according to Step 10:
For the calculation of L, the ratio of UTP peak sizes of 12C-glucose- to 13C-glucose-grown cultures was determined, as described in Step 11B:
|Sample no. 1||Sample no. 2||Sample no. 3||Sample no. 4|
L is determined for each sample, according to Step 11B, with no. of carbons = 9:
|Sample no. 1||Sample no. 2||Sample no. 3||Sample no. 4|
Lavg is calculated, according to Step 12:
This relatively small value of L is reasonable, as there are two separate reactions that can incorporate unlabeled carbonate into UTP: those catalyzed by carbamoyl-phosphate synthetase and phosphoenolpyruvate (PEP) carboxylase (which yields oxaloacetate, which forms aspartate and then condenses with carbamoyl phosphate to form carbamoyl aspartate, the committed compound of de novo pyrimidine biosynthesis).
is calculated according to Step 13:
The total cellular volume of the culture, F, is calculated according to Step 17, based on the measured culture dry weight of 0.8 mg:
Cavg is calculated from the above values according to Step 18, with Z determined to be zero in previous experiments17 and with S, the concentration of the internal standard in the first extraction step, equal to 8.28 × 10−7 M:
The variance of Cavg in logarithmic space, on the basis of variances of L and R, is calculated according to Step 18:
The standard error of Cavg, in logarithmic space, is calculated according to Step 20:
The upper and lower bounds of the determined concentration are determined, according to Step 21:
A final concentration of 3.3 mM UTP, with a 95% confidence interval of 2.5–4.4 mM, is reported.
This research was supported by the Beckman Foundation, National Science Foundation–Dynamic Data-Driven Application Systems grant CNS-0540181, American Heart Association grant 0635188N, National Science Foundation CAREER award MCB-0643859, National Institutes of Health grant AI078063 and National Institutes of Health grant GM071508 for the Center of Quantitative Biology at Princeton University. We thank Wenyun Lu for contributing LC-MS expertise.