Single-Cell Metabolomics: Analytical and Biological Perspectives

See allHide authors and affiliations

Science  06 Dec 2013:
Vol. 342, Issue 6163, 1243259
DOI: 10.1126/science.1243259

Structured Abstract


In recent years, there has been a surge in the development and application of single-cell genomics, transcriptomics, proteomics, and metabolomics. The metabolome is defined as the full complement of small-molecule metabolites found in a specific cell, organ, or organism. The most interesting potential application of single-cell metabolomics may be in the area of cancer—for example, identification of circulating cancer cells that lead to metastasis. Other fields where single-cell metabolomics is expected to have an impact are systems biology, stem cell research, aging, and the development of drug resistance; more generally, it could be used to discover cells’ chemical strategies for coping with chemical or environmental stress. Relative to other single-cell “-omics” measurements, metabolomics provides a more immediate and dynamic picture of the functionality (i.e., of the phenotype) of a cell, but is arguably also the most difficult to measure. This is because the metabolome can dynamically react to the environment on a very short time scale (seconds or less), because of the large structural diversity and huge dynamic range of metabolites, because it is not possible to amplify metabolites, and because tagging them with fluorescent labels would distort their normal function.

Embedded Image

Single-cell analysis uses a wide variety of imaging and chemical analysis methods to study vastly different cell types and sizes. (A) Closterium acerosum (algal cells, ~300 μm × 40 μm; optical micrograph). (B) Euglena gracilis (algal cells, diameter ~20 μm); Raman image of β-carotene distribution (left) and fluorescence emission from proplastids (right). (C) Baker’s yeast (diameter ~5 μm); optical micrograph. (D) Escherichia coli (diameter ~0.75 μm, length 1 to 3 μm); fluorescence micrograph (image courtesy of M. Heinemann, University of Groningen).


Although deep biological insight based on single-cell metabolomics has not yet been obtained, important steps have been taken toward this goal. Advances in mass spectrometry (MS), MS imaging, capillary electrophoresis, optical spectroscopy, and in the development of fluorescence biosensors now allow the simultaneous determination of hundreds of metabolites in a single cell, with sensitivities in the attomole range. Modern array formats, in particular microfluidic platforms, contribute to our ability to perform such measurements rapidly and with high throughput. Several recent studies show how novel biological insight can be extracted from single-cell metabolomics. Substantial differences in the metabolomes of different snail neurons—for example, in B1 and B2 type neurons—have been found, immediately after isolating them and after overnight culturing. Glycosphingolipids could be labeled with a fluorescent tag, and in lysates of neurons incubated with such conjugates, all metabolic products derived from them were fluorescent and could be identified. Phosphorylation of 3′-deoxy-3′-fluorothymidine in lymphoma cells and solid tumors could be followed after treatment with cancer drugs. The biological effect of treating yeast cells by 2-deoxy-d-glucose (2DG) on the metabolome could be followed. The fact that single-cell measurements exhibited a much larger spread in metabolite concentrations than population measurements was exploited to determine many metabolite-metabolite correlations, which were altered in 2DG-treated yeast cells relative to controls.


The metabolome is an excellent indicator of phenotypic heterogeneity and has been recognized as a key factor in rare-cell survival when populations are subjected to major chemical or environmental challenges. Metabolomics at the single-cell level, however, is only just coming of age. Improvements leading to more complete coverage of the metabolome, better and faster identification of metabolites, and nondestructive measurement are anticipated.


There is currently much interest in broad molecular profiling of single cells; a cell’s metabolome—its full complement of small-molecule metabolites—is a direct indicator of phenotypic diversity of single cells and a nearly immediate readout of how cells react to environmental influences. However, the metabolome is very difficult to measure at the single-cell level because of rapid metabolic dynamics, the structural diversity of the molecules, and the inability to amplify or tag small-molecule metabolites. Measurement techniques including mass spectrometry, capillary electrophoresis, and, to a lesser extent, optical spectroscopy and fluorescence detection have led to impressive advances in single-cell metabolomics. Even though none of these methodologies can currently measure the metabolome of a single cell completely, rapidly, and nondestructively, progress has been sufficient such that the field is witnessing a shift from feasibility studies to investigations that yield new biological insight. Particularly interesting fields of application are cancer biology, stem cell research, and monitoring of xenobiotics and drugs in tissue sections at the single-cell level.

Every person is different, and this individual phenotype is based on genetic, epigenetic, developmental, and environmental differences. What about cells in a clonal or isogenic culture, whose members are all derived from a single common ancestor? Even in this case there are differences in phenotype, attributable to environmental influences (including the cell cycle), the age of an original cell, and other factors that affect the phenotype of individual cells in the culture. Imagine a hypothetical situation in which all cells in a clonal population are perfectly synchronized and grown in exactly the same conditions. Even then, different phenotypes will develop in the population after some time, frequently as a result of stochastic biological processes (Fig. 1A). “Stochastic” in a biological context refers to irregularities or “noise” in the rates at which biochemical reactions run. This is quite pronounced for processes that involve very few or single molecules—for example, DNA transcription and protein expression (1). To assess how different the resulting phenotypes are, single-cell measurements are necessary; population measurements yield only average values (Fig. 1B). Stratification in tissues may also warrant single-cell measurements. For example, neurons are a cell type in which individual differences are required for function. Excitation of an individual neuron in a mammal can initiate the movement of a whisker, change learning, and thus affect the entire organism’s behavior. Even in a complex ensemble of sensory neurons, individual neurons have different receptor fields and are therefore distinct.

In recent years, there has been a surge in the development and application of single-cell molecular profiling [for reviews with a focus on metabolomics, see (28)]. The metabolome can be defined as the full complement of small-molecule metabolites (with molecular weights of less than ~2 kD) found in a specific cell, organ, or organism (9). For the present purposes, the metabolome (Fig. 1C) includes endogenous as well as exogenous small molecules [e.g., pyruvate, lactate, sugars, adenosine monophosphate (AMP), adenosine diphosphate (ADP), adenosine triphosphate (ATP), etc.], drugs and their metabolites (often referred to as xenobiotics), and lipids, as well as peptides that are not degradation products of proteins (e.g., neuropeptides that are important in cellular signaling). Nucleic acids with short sequences may qualify in terms of their molecular weight, but they are considered part of the cellular transcriptome rather than the metabolome. Salts are not considered metabolites.

Fig. 1 Development of different phenotypes in cell cultures.

(A) Reasons for phenotypic variations can range from genetic differences to stochastic processes. (B) Hypothetical histogram of a population average (gray bars) and a single-cell measurement (white and black bars) of the same parameter (e.g., a metabolite concentration) revealing bistability only in the case of the single-cell determination. (C) Some examples of metabolites, illustrating the large structural diversity of compounds covered by metabolomics.

The genotype of a cell describes its “potential” whereas the phenotype describes its function, but the link between the two is often obscure (10). Relative to single-cell genomics, transcriptomics, or proteomics, metabolomics provides the most immediate and dynamic picture of the functionality (i.e., the phenotype) of a cell, but is arguably also the most difficult to measure: Whereas the genome is more or less static, and the transcriptome and proteome change on a time scale of minutes to hours (11), the metabolome reacts to environmental influences on a time scale of seconds or even milliseconds (12). This fast dynamics is one of the major challenges for single-cell metabolomics. It requires protocols that quench the metabolism of cells to be studied, because they experience sample preparation itself as an environmental stress and will react to it (13). Other challenges include the large structural diversity of the compounds that encompass the metabolome (Fig. 1C), the large dynamic range (from a few molecules per cell to 1010 molecules for major metabolites in larger cells), our inability to amplify metabolites (as is commonly done with DNA), and the need to refrain from fluorescent labeling: Very few metabolites are autofluorescent, and attaching a label would in most cases prevent a metabolite from performing its biological function.

Detecting and understanding cancer cells is among the most interesting potential applications of single-cell metabolomics (14, 15). The detection of cancerous cells that display abnormally high metabolic rates among many others with normal metabolism, including circulating cancer cells that lead to metastasis, would be one such application. In the framework of cancer therapy, one might identify cells in a tissue that develop resistance to a drug treatment, or, more generally, discover the chemical strategies of how some cells successfully cope with chemical or environmental stress, whereas others die (16). Other potential uses of single-cell metabolomics (relative to other “-omics” methodologies) are to obtain input and output data for mathematical models of cellular metabolism (17), to learn more about aging (18, 19), and to predict the developmental fate of stem cells.

Have single-cell metabolomic data already yielded deep biological insight? I would argue that this is not yet the case. One reason is that stochasticity cannot generally be followed by studying metabolites; there are only indirect and often small effects on the metabolome from stochastic biochemical processes. Moreover, important metabolites appear in numerous nodes of the metabolic network in cells; that is, either multiple correlations between major metabolites are needed to obtain any insight, or, alternatively, precise measurements on rare and low-concentration metabolites with highly specialized roles need to be performed, which is difficult. However, a few studies discussed below (2023) have already enhanced our knowledge about biological systems.

This review focuses on the rapidly developing analytical methodology, but also discusses what can be learned about a biological system if metabolomic data are collected at the single-cell level rather than from population averages (Fig. 1B), on how the variations in the metabolome among many single cells can give new information about the causes and consequences of cell variability, and on whether these variations are adaptive, an epigenetic phenomenon, or a basic property of biochemistry.

General Considerations for Single-Cell Metabolomics

Unless the sample is a cell suspension, the first step of sampling (2, 6) is to isolate appropriate cells from an organism, which is sometimes done manually, under microscopic observation. Once cells are isolated, some approaches sample an individual cell directly; others involve cell culture, for example, in a microfluidic device. A major issue, as explained above, is a suitable sample preparation that does not upset the metabolism of the cells to be investigated. One way to cope with this problem is to keep the cells in a native environment as long as possible. A number of highly successful microfluidic chips that gently trap cells have been presented in the literature (2427). Key functions of these microfluidic platforms are to isolate cells, culture them under well-controlled conditions, inject highly defined amounts of chemicals into the growth medium, and selectively release cells for analysis, which may involve an on-chip lysis step (28). Another option is to shock-freeze cells before subjecting them to measurement (23) to quench the metabolism.

Another issue is the required sensitivity. Cell sizes vary widely. Typical mammalian cells have diameters around 10 μm (volume = 1 pl); the giant neurons of the sea slug Aplysia californica, which have frequently been used in early single-cell studies because they can be manipulated manually under a microscope, can reach 500 μm. On the other end of the size spectrum are model organisms such as yeast (diameter ≈ 5 μm) and bacteria with diameters on the order of 1 μm (volume = 1 fl). Assuming a metabolite concentration of 1 mM, the absolute amounts that need to be detected in these tiny volumes are thus in the range of 1 amol to 1 fmol, which is challenging, even for major metabolites. Interestingly, the concentration sensitivity is less of an issue: The presence of a single molecule inside a 6-fl bacterial cell translates into a concentration of 0.28 nM; that is, it will generally not be necessary to measure concentrations lower than nanomolar.

Furthermore, cells are usually grown in medium rich with molecules that are similar or even identical to metabolites. Thus, it is critical to differentiate between the metabolites in the surrounding medium (footprinting) and the metabolites within the cell (fingerprinting).

Finally, high-throughput formats for sampling cells are clearly necessary; an isolated measurement on one single cell may be less meaningful in a biological context (although if a single cell could be precisely and continuously analyzed in its morphological and molecular aspects, biological insight could be obtained from that single cell). To generate statistically significant data, hundreds or thousands of measurements are generally necessary, which presents another challenge. A number of strategies that enable high-throughput interrogation of many single cells have been developed. Classical high-throughput formats are flow cytometry and modifications thereof, such as fluorescence-activated cell sorting (FACS; Fig. 2A). Flow cytometry measurements are, however, generally not at all linked to metabolites and may be used to separate a cell culture into two or several subpopulations that are subsequently analyzed. Special formats of cytometric measurements have thus been developed—for example, a mass spectrometric readout following flow cytometric sample delivery (29). Other high-throughput formats include microarray printing of controlled numbers of cells (30) and many different lab-on-a-chip devices (31). One such platform developed by Di Carlo and Lee that allows gentle trapping is shown in Fig. 2B (24). Many U-shaped hydrodynamic trapping structures allow both arrayed culture of individual adherent cells and simultaneous control of fluid perfusion with uniform environments for individual cells. Cell loading can be achieved in less than 30 s.

Fig. 2 High-throughput methods for preparing single cell for chemical analysis.

(A) Flow cytometry of human bone marrow cells, using fluorescent antibodies that bind to cell surface antigens (here, CD45RA versus CD4). Each dot in the plot originates from a single cell. [Reproduced from (92)] (B) Microfluidic single-cell trapping array. Upper left: Increasingly higher magnifications depict the cell-trapping device as a whole (scale bar, 500 μm; cells and media flow enter from the left). Lower left: High-resolution bright-field micrograph of the trapping array with trapped cells. Lower right: Magnification showing the details of one trapped cell. Trapping is a gentle process, and no cell deformation is observed for routinely applied pressures. Upper right: Diagram of the device and mechanism of trapping (not drawn to scale). [Reproduced, with permission, from (24)] (C) Microarray for mass spectrometry (MAMS) chip. Scale bars, 1440 μm; diameter of individual wells, 300 μm. On a MAMS array the size of a 1” x 3” microscope slide, 2860 wells can be placed. Chlamydomonas reinhardtii algal cells; chlorophyll fluorescence readout measured with a LS 400 scanner (Tecan, Männedorf/Switzerland). [Image courtesy of Jasmin Krismer, ETH Zürich, and Jens Sobek, Functional Genomics Center, Zürich]

High-density chips are also becoming available to allow measurements by mass spectrometry (MS). As described below, MS is one of the most successful methods for single-cell metabolomics—which, however, means that high-throughput preparation of single-cell samples becomes the bottleneck. One possible solution involves MAMS (microarray for mass spectrometry) chips (32), as shown in Fig. 2C. A very attractive feature of MAMS chips is that the hydrophilic wells surrounded by an “omniphobic” polysilazane coating on the surface allow automated isolation of small volumes containing single cells from cell suspensions, by simply dragging the liquid over the surface of the chip. Current versions of such MAMS chips allow 10,000 to 50,000 wells to be filled and subsequently analyzed; the number of cells per well is given by a Poisson distribution.

Preparing Single-Cell Samples for Analysis


Although microfluidics is not an analysis method per se, microfluidic devices are extremely useful in presenting single-cell samples for readout by optical spectroscopy, mass spectrometry, or other means. The job of the microfluidic device is to transport, immobilize, culture, infuse with reagents, hold for observation, and retrieve single cells in a high-throughput fashion. Formats include patch-clamp array (33), dynamic single-cell culture array (24), and integrated microfluidic array plate (iMAP) (34), in all cases using microscale soft lithography with polydimethylsiloxane (PDMS) as the material. The dynamic single-cell culture array (Fig. 2B) allows an arrayed culture of many individual adherent cells (~100 in a field of view of ~1 mm2). Trapping is passive and self-terminating in the sense that once a site is filled with a cell, the altered hydrodynamic flow around the filled site prevents other cells from entering it. This device has not been used for chemical measurements so far, but it appears to provide a very gentle growth environment, as shown for the high rate (95%) of survival of HeLa cells after 24 hours of perfusion culture on the array. The iMAP array is based on gravity-driven flow and sedimentation to capture the cells, with close to 100% capture rate. It features open access for fluid exchange; gene expression, protein immunoassay, and cytotoxicity data can be accessed in parallel.

Another strategy is the dynamic microfluidic array based on fluidic resistance (25). The flow channel has a meander-like shape with some additional small “bays” (hydrodynamic traps) where cells get trapped passively. To selectively release beads or cells from such a bay, a microbubble is generated by laser heating of an Al pattern near an individual trap. The authors wrote that the heating is of no concern for the integrity of the beads that were investigated (the Al structure reaches temperatures of >130°C), but one would probably have to worry about the transient temperature jump as an environmental influence to which a cell’s metabolome would react. Although this device has so far been developed and operated only with beads rather than with cells, 100 objects could be trapped and individually addressed. For single-cell analyses, such a microfluidic device would be operated with medium as the circulating liquid and be used to deliver individually addressable cells to a subsequent analysis (e.g., by mass spectrometry).

Microchamber arrays can be used not only for single-cell isolation but also for the analysis of intracellular biomolecules. This design is based on PDMS valves that encapsulate single cells in circular reservoirs with volumes of ~625 pl. These microchambers can be opened and closed rapidly and reversibly; such a design allows incubation, washing, labeling, and lysis steps to be done with single cells (35). The lysate remains contained in the small volume of the microchamber; although dilution still occurs, it is controlled and limited such that various target analytes can be directly studied, cell by cell. This format has been used for analyzing the cofactors NADH (reduced nicotinamide adenine dinucleotide) and its phosphate NADPH, and for quantitative assays of a number of intracellular biomolecules, including compounds such as cyclic AMP (cAMP) in human embryonic kidney (HEK) T-Rex cells, production of which is stimulated by the hormone lutropin (36). Attomole amounts of cAMP (between 250 and 1000 amol, increasing with the level of stimulation by lutropin) were detected. Because a competitive enzyme-linked immunosorbent assay (ELISA) was used for detection, this format provides the equivalent of a single-cell immunoassay.

There are many other microfluidic formats for single-cell trapping, culturing, and handover to analysis [for further information, see (31, 3739)], although in many cases these are not specially designed for obtaining chemical information on metabolites. Some formats allow very interesting microscopic observations to be conducted—for example, the observation of aging processes of budding yeast cells throughout their life span (19), which revealed remarkable age-associated changes in phenotypes and substantial heterogeneity in cell aging and apoptosis. This is, however, not a metabolomics study, and the same is true for most other microfluidic platforms: In connection with metabolomics, their usefulness is to observe and classify individual cells, stimulate them inside the microfluidic device, and deliver them in a rapid but controlled fashion to a subsequent analysis step that identifies the metabolites.

Nanoscale Devices

Nanoscale devices can be used to manipulate single cells or deliver chemicals into cells in a controlled fashion. A nanowire waveguide-based approach allows single-cell optical endoscopy (40), a hollow atomic force microscopy (AFM) probe can be inserted through the cell wall and used to deliver liquids into the cytoplasm (41), and a nanochannel has been used to deliver precise amounts of biomolecules into living cells (i.e., a kind of precision transfection technique) (42). Nanoscale devices are also being developed to analyze cells, as summarized in (43)—for example, optically through near-field methods, through AFM, or electrochemically by scanning conductance microscopy. Although much of the relevant literature cites the potential and especially the need for high-throughput operation, most of these nanoscale approaches are at present slow, serial, and difficult to control.

Analytical Methodologies for Identifying Metabolites in Single Cells

Mass Spectrometry

Mass spectrometry (MS) is rapidly becoming one of the most widely used methods for ultrasensitive and simultaneous detection of many metabolites at the single-cell level. Label-free, highly sensitive, and information-rich, MS has contributed since the early days to this field. MS can be used as a detector for flow cytometry (29), with a capability of infusing hundreds of cells in a few minutes. Histamine (0.75 ± 0.33 fmol) and serotonin (0.11 ± 0.06 fmol) have been detected in rat peritoneal mast cells, however, without any obvious correlation of the amounts of the two amines in each cell. Giant neurons of A. californica (20, 44) provide the ease of handling very large cells. Principal components analysis of large data sets with more than 300 distinct cell-related signals revealed significant differences in the metabolomes of various Aplysia neurons—for example, in B1 and B2 type neurons—immediately after isolating them and after overnight culturing. It was also possible to determine absolute concentrations of several metabolites in the isolated neurons. For example, intracellular glutamic acid concentration was 11 mM in one type of neuron and 4 mM in others. This particular strategy involved both capillary electrophoresis (CE) and MS, which affords a quite powerful way to identify and detect metabolites (44).

Among the modern ionization methods, matrix-assisted laser desorption/ionization (MALDI) has sufficient sensitivity for single-cell analyses. However, the presence of intense matrix signals in the ≤500-dalton range is of great concern for metabolomics. A number of strategies have thus been developed to circumvent this problem. Matrix-free ionization methods based on nanophotonic effects are available—for example, using desorption-ionization from porous silicon (DIOS) or silicon nanopost arrays (NAPAs), which in the hands of some laboratories feature a sensitivity in the sub-femtomole or even sub-attomole range (45, 46). NAPA-MS revealed metabolic differences in stressed and control yeast cell populations (47). The sensitivity was enough to determine metabolites in samples consisting of 1 to 80 cells in a semi-quantitative manner, and more than 100 peaks could be assigned to various metabolites, representing ~9% of the ~1200 known yeast metabolites. Following oxidative stress, up-regulation of glutathione, cysteinylglycine, glutamylcysteine, and urate (among others) was observed, whereas compounds related to folate biosynthesis, such as amino-4-deoxychorismate or dihydroneopterin phosphate, were down-regulated, indicating that the cells redirected resources from growth to fighting stress.

Another strategy is to use MALDI matrices that generate only few well-defined signals. An excellent compound is 9-aminoacridine (9-AA), which has the added benefit of promoting formation of negative ions (48). This is useful for detection of metabolites in their deprotonated form, such as small organic acids or phosphorylated compounds. Single-cell detection sensitivity for metabolites in yeast has been shown using 9-AA (49), and the same matrix has been used in subsequent single-cell studies (4, 50, 51). Negative ion mode MALDI and 9-AA allowed detection of 26 different metabolites in single yeast cells (23). Hundreds of cells were measured in this study with a MAMS chip, and the effect of a chemical perturbation by 2-deoxy-d-glucose (2DG) on the metabolome was measured (Fig. 3A). The fact that single-cell measurements exhibited a much larger spread in metabolite concentrations relative to population measurements was exploited to determine many metabolite-metabolite correlations, which were altered in the case of 2DG-treated yeast cells versus controls.

Fig. 3 Analyses of metabolites at the single-cell level by mass spectrometry, showing the capability of this method.

(A) Full-scan negative-ion mode MALDI mass spectrum of a single YSBN.6 (S. cerevisiae) cell (m/z = 70 to 900) treated with 2DG; the measurement took place 1 min after the spike of 2DG into the growth medium occurred. Inset shows two spectra: (i) m/z = 100 to 180; (ii) m/z = 330 to 350. [Reproduced, with permission, from (23)] (B) LDI mass spectrum of a single yeast cell from a NAPA in the mass range of metabolites. Four of the ~24 putatively assigned metabolites are labeled. Inset: AFM image of a single yeast cell deposited on a NAPA substrate before LDI-MS analysis. [Reproduced, with permission, from (46)] (C) Example of full spectrum (m/z range 100 to 1000) and photograph of Pelargonium zonale single-cell contents extraction by the nanoelectrospray tip from a leaf cell. [Reproduced, with permission, from (55)] (D) MALDI time-of-flight (MALDI-TOF) imaging detects strong neuropeptide signals in MS and MS/MS modes from cell culture samples of cultured pedal neurons. The automatically generated MS/MS peak assignment for m/z 1540 yields the sequence that corresponds to the known amino acid sequence for Aplysia pedal peptide, PLDSVYGTHGMSGFA. [Reproduced, with permission, from (60)]

Alternatively, electrospray ionization (ESI)–based methods can be used. A method called laser ablation electrospray ionization (LAESI; Fig. 3B) was developed for in situ mass spectrometric analysis of individual cells at atmospheric pressure (52). Single-cell probing was achieved by delivering mid-infrared (IR) laser pulses (that are absorbed by the water in the sample) through the etched tip of a GeO2-based glass fiber, and the laser ablation products were post-ionized in a pure solvent ESI plume. Metabolic analysis was achieved from single cells and small cell populations of Allium cepa and Narcissus pseudonarcissus bulb epidermis, as well as single eggs of Lytechinus pictus. Among the 332 peaks detected for A. cepa, 35 were assigned to metabolites with the help of accurate mass and MS-MS measurements. Differences in the metabolite profile of adjacent cells with different pigmentation could be discerned for N. pseudonarcissus; A. cepa showed metabolic differences as a function of cell age. The cells of these plant species, however, were rather large, on the order of 70 × 400 μm. A completely different ESI-based approach dubbed live single-cell video mass spectrometry (53, 54) involves inserting an electrospray needle into a cell under video microscopic observation, removing some of the cytoplasm, and then directly electrospraying the liquid via the same needle into an ambient inlet mass spectrometer (Fig. 3C). This method is even able to distinguish between cytoplasm-specific and granule-specific signals, showing, for example, differences in the subcellular metabolite distribution of a rat leukemia mast cell line (RBL 2H3) model. Around 1000 peaks belonging to metabolites were distinguished in plant tissue (from Pelargonium zonale, geranium), and a combination of high-resolution MS, tandem mass spectrometry, and database searches yielded tentative assignments for more than 20 signals (55). This method, however, can hardly be called high-throughput. Sampling the contents of the cells is still best done manually and is difficult, and mass spectrometric analysis is performed off-line, in a second step. At best, a few cells can be analyzed per hour.

Mass Spectrometry Imaging

Mass spectrometry imaging (MSI) is dealt with separately here because of the completely different manner in which samples are prepared and subjected to analysis (56). Both MALDI and secondary ion mass spectrometry (SIMS) imaging are used. On research-grade MALDI-MS instruments, MSI can now be done with a spatial resolution of <1 μm (57, 58), at ambient pressures, and with very high mass accuracy and mass resolution (59). This corresponds to subcellular resolution for many kinds of cells. For the sake of generating enough signal per spot, the spatial resolution is often chosen to be somewhat lower, in the range of 5 to 30 μm, which is also the spatial resolution achieved by commercial MALDI instruments with imaging capabilities (60). MSI is capable of detecting and identifying a broad range of metabolites and is often applied to tissue sections, where a cell-by-cell distinction is feasible. One of the premier applications of MALDI-MSI is the visualization of drugs and their metabolites in tissue section or even whole animal thin sections. Although single-cell resolution is in principle available, it is often not used in such studies, for the sake of increasing the speed of data acquisition, or because it is simply not required. For example, in tuberculosis-infected rabbit lung tissue sections, the distribution of the second-line tuberculosis drug moxifloxacin was imaged at several time points after dosing (61). The drug was observed to accumulate in granulomatous lesions in amounts higher than those in the surrounding lung tissue, with an inhomogeneous distribution within the granulomas; very low amounts were observed in the caseum relative to the cellular granuloma regions.

Nanostructure-initiator mass spectrometry (NIMS) imaging has been used to follow the reaction of individual cancerous cells and cancer xenografts to chemotherapy (22). Phosphorylation of 3′-deoxy-3′-fluorothymidine (FLT, a thymidine analog) to FLT monophosphate (FLT-MP) was followed in lymphoma cells and solid tumors after treatment of the specimens with rapamycin or FLT. Although FLT is readily transported out of cells, FLT-MP is not, and its accumulation serves as an indirect indicator of drug-induced metabolic changes of treated cells. Complementary liquid chromatography–tandem MS (LC-MS/MS) measurements on extracts of many cells were also conducted, with consistent results.

Secondary ion mass spectrometric imaging is capable of generating chemical maps of complex surfaces, such as tissue sections or cell cultures, with subcellular resolution [down to <100 nm (6264)]. In SIMS, a compromise often must be found between very high spatial resolution and the ability to detect intact molecular species: SIMS with highly focused primary ion beams capable of sub–100 nm resolution generally operates in the “dynamic SIMS” mode (i.e., a high primary ion dose is necessary to generate enough signal). This, however, turns out to be detrimental to the integrity of larger organic molecules. There have been reports of studies that combine MALDI imaging with SIMS, the latter providing the higher spatial resolution if required (65). Both allow label-free and simultaneous determination of the identity and distribution of xenobiotics and their metabolites as well as endogenous substances in tissue. This is an interesting extension to whole-body microautoradiography, eliminating the need for radiochemistry and providing molecule-specific information.

Mass spectrometric analyses have in common that among all the analytical methods presented, they yield the most detailed information about the compounds that are detected. A good strategy is to use accurate mass measurement (59), which often restricts the possible elemental compositions of a signal to only a handful or ideally, to a unique sum formula (see Box 1). Sometimes tandem MS (MS/MS) can even be performed on signals from a single cell (55), although normally this is not possible because of an insufficient signal-to-noise ratio. Alternatively, MS/MS analysis can be done on the same, accurately measured mass-to-charge ratio (m/z) signal from a larger sample, assuming that the same compound appears in the single-cell data. MS/MS may not always give the information sought. For example, differently phosphorylated carbohydrates exhibit very similar MS/MS spectra and therefore may be indistinguishable by mass spectrometry. Finally, an increasing number of databases list metabolites for various species (Box 1). Looking up putative metabolites in a database definitely has its merits, in particular for automated data interpretation, although it helps little in the quest to identify new metabolites.

Box 1.

Identification of metabolite signals from mass spectrometry data and database searches.

Nominal mass: 180 daltons → 141 different possible elemental compositions that include C, H, N, O, S, Cl, Br, and I; very large number of isomers. Includes the elemental composition C6H12O6. Some elemental compositions are not chemically sensible, for example, HO9Cl, C12HCl, or C15.

Accurate mass: 180.06339 → 2 possible elemental compositions with less than 1 ppm deviation, C5H6N7O (chemically unlikely) and C6H12O6 (a carbohydrate). 32 different isomers (fructose, galactose, gulose, sorbose, inositols, etc., including stereoisomers: d and l forms).

MS/MS information: Glucose (d-glucose and l-glucose indistinguishable by tandem MS)

Selection of Databases on Metabolites and Mass Spectrometry

Human Metabolome Database ( a freely available electronic database containing detailed information about small-molecule metabolites found in the human body.

Pubchem ( provides information on the biological activities of small molecules.

Metabolights ( a database for metabolomics experiments and derived information. The database is cross-species, is cross-technique, and covers metabolite structures and their reference spectra.

Kegg ( a database resource for understanding high-level functions and utilities of the biological system, such as the cell, the organism, and the ecosystem, from molecular-level information.

Metlin ( a repository of metabolite information and tandem mass spectrometry data, for use in metabolomics experiments.

Yeastnet ( a portal to the consensus yeast metabolic network as reconstructed from the genome sequence and literature.

The Yeast Metabolome Database ( a manually curated database of small-molecule metabolites found in or produced by Saccharomyces cerevisiae. It covers metabolites described in textbooks, scientific journals, metabolic reconstructions, and other databases.

ChemBioFinder ( an online chemistry and biology reference database with almost 2 million compounds indexed and linked to other Web sites.

Chemspider ( free chemical structure database providing fast text and structure search access to more than 28 million structures from hundreds of data sources.

Massbank ( public repository of mass spectral data, useful for the chemical identification and structure elucidation of chemical compounds detected by mass spectrometry.

Fluorescence-Based Detection, Fluorescence Biosensors, and FRET Biosensors

Fluorescence microscopy is in principle an excellent, sensitive, nondestructive, and widely available method to image single cells, and in combination with fluorescent tags, it can also provide chemical specificity. With modern image processing, it is straightforward to get information in a high-throughput fashion. However, very few metabolites are autofluorescent; examples include the detection of carotenoids in red yeast (66), the distinction of tumor cells in a tissue through autofluorescence spectroscopy and decay kinetics (67), or the localization of chlorophyll pigments related to the photosynthetic reaction center II and internal antennae of photosystem II in algal cells (68). Attaching fluorescent labels to low–molecular weight metabolites turns out to be too invasive: The fluorescent tag itself will disturb the biochemical function of the metabolite. Thus, methods that provide indirect fluorescent readout as a function of the presence or concentration of a metabolite are required. “Smart” designs of fluorescent proteins (FPs) can be exploited in various ways for use as biosensors (69): by activating the fluorescence by small-molecule ligand binding to the FP, by inducing a conformational change of the FP in the presence of a metabolite, by oligomerization-dependent fluorescence of more than one FP moiety, or through Förster resonance energy transfer (FRET) of two different FPs, activated by binding of a metabolite. Often such formats are referred to as “nanosensor probes.”

Numerous fluorescence-based methods exist to detect AMP, ADP, or ATP—for example, an assay in which a dipicolylamine ligand, a naphthalimide chromophore, and a Zn(II) center form a fluorescent complex in the presence of ATP (70). High Stokes shift (>70 nm), high fluorescence quantum yield, good selectivity against other organic and inorganic phosphates, and a sensitivity down to 1 μM ATP were some of the key characteristics. One problem with fluorescence detection is that a fluorescence signal is readily perturbed by changes in the environment (e.g., by changes in pH, temperature, or solvent polarity). Ratiometric measurements that involve simultaneous determination of two fluorescent signals at different wavelengths can circumvent these problems and yield greater precision for quantitative determinations. Binding-induced modulation of FRET is a possible embodiment of a ratiometric measurement. Such a FRET biosensor has been useful for detecting nucleoside polyphosphates (71). It showed good sensitivity (<<1 μM for ATP) and excellent linearity; the authors were able to follow the energy charge of cells in a human cancer (HeLa) cell line after treatment with 20 mM 2-deoxyglucose and 1 mM KCN. A problem, however, was the limited selectivity: Binding of a range of nucleoside polyphosphates to the FRET sensor molecules was reported.

There are a number of difficulties that render fluorescence microscopy less suitable for true metabolomics: Genetically encoded nanosensor probes exist for only a handful of specific small molecules (72), and it is relatively complex to express the biochemical machinery for indirect fluorescence detection. Often tedious optimization is required, the specificity and optical contrast are sometimes not satisfactory, and the genetic modification may affect the native physiology of such cells. Also, the number of different fluorescent reporter tags that can be simultaneously imaged is limited (to around 10). In fact, the literature does not seem to include any report of simultaneous detection of multiple target metabolites; however, dozens to hundreds would be required for true metabolomics. On the other hand, fluorescence is extremely useful for ultrasensitive detection, down to the level of a single molecule. This has been shown for proteins in individual cells that were handled through microfluidics (1, 73) and was a key method to detect stochasticity in expression of small–copy number proteins in single cells. It is also very powerful for ultrasensitive detection after CE separation of the chemical contents of cells (21, 74, 75).

Vibrational Spectroscopy

In some cases, vibrational spectroscopy can be used to distinguish certain metabolites in single cells (76). As opposed to detection through fluorescent tags, vibrational spectroscopy is label-free and thus amenable to spectroscopically active compounds (e.g., pigments) as long as they are present in sufficiently high concentration. An example is the localization of β-carotene by its 1150 and 1515 cm−1 Raman bands with subcellular resolution (~550 nm per pixel) in Euglena gracilis algal cells (68). Complementary single-cell MS data were also recorded in this case. A colocalization of β-carotene and the plastids containing internal antennae of photosystem II was shown. Synchrotron-based IR spectroscopy, with multiple low-emittance beams and a large focal plane array detector, was used to achieve a pixel size of 540 nm in the mid-IR, about two orders of magnitude smaller than is possible with conventional thermal or synchrotron IR sources (77). Detailed spectroscopic information was available, for example, from images of the CH3 stretching (2950 cm−1) and the amide I (1654 cm−1) modes, with single-cell resolution, on thin sections of cancerous prostate tissue with chronic inflammation. Corneal epithelial cells have been isolated from biopsies of live tissues by FACS; these were divided into putative stem cells (SCs), transit-amplifying (TA) cells, and terminally differentiated (TD) cells (78). DNA regions of the spectra (1080 and 1225 cm−1) and some protein regions (1443 cm−1) primarily distinguished SCs from TA cells and TD cells, whereas amide regions and lipids (1550, 1650, and 1740 cm−1) could be used to distinguish TA cells and TD cells. A stimulated Raman scattering microscopy technique (79) allows visualization of various biomolecules (e.g., lipids) by their saturated C-H vibrations at 2845 cm−1, proteins by their amide I band at 1655 cm−1, or nucleic acids by their 785 and 1090 cm−1 bands. Pixel sizes on the order of 200 nm are possible with this technique; that is, subcellular spatial resolution is possible. The notable feature of vibrational spectroscopy is that absolutely no staining or labeling of the cells is required. On the other hand, the spectra are not really specific for metabolites, but instead reveal all compounds that are highly concentrated and spectroscopically active. This often leads to spectral congestion of IR and Raman spectra, which renders them less suitable as tools for metabolomics.

Separation-Based Chemical Analyses

Capillary electrophoresis and capillary LC have been used without exception for single-cell studies involving separation on a column, for the obvious reason that they can handle very small volumes. It is possible to detect metabolites, including amino acids and neurotransmitters, in quantities of tens of femtomoles in large snail neurons after capillary separation and voltammetric as well as fluorescence detection (80), and even sub-attomole detection limits for serotonin with amperometry have been reported (81). This approach is currently most successfully used for single-cell proteomics (82, 83), although some applications to metabolites have appeared in the literature (8486). In particular, quantitative studies have been reported on glycosphingolipids in single neurons with one- (74), two- (21), and three-color (75) fluorescence detection (Fig. 4). The general strategy is to label one or more glycolipids [for example, mammalian ganglioside GM1 and lactosylceramide in (21)] with fluorophores, incubate cells with these conjugates, and follow glycolipid catabolism and anabolism by the appearance of fluorescent products. These were identified by aspirating single cells into a capillary, lysing them inside it, and separating the metabolic products by CE. Identification was possible by comparison with standard compounds, taking into account the slightly different migration times of the labeled glycosphingolipids. About a dozen metabolites were identified, and individual cells showed vastly different uptake of the labeled glycosphingolipids. A dynamic range of six orders of magnitude and a separation power of ~106 theoretical plates were reached. For glycosphingolipids labeled with boron dipyrromethane (BODIPY) fluorophores, extreme sensitivities of 34 ± 1 to 67 ± 7 molecules could be achieved, depending on the exact nature of the label. Obviously, the greater the number of distinct fluorophores that are introduced, the more metabolic pathways that can be followed simultaneously. The limitations, however, are the same as simultaneous detection by a number of different fluorescence channels, as well as the disturbance of the metabolism itself by the presence of the rather bulky fluorophores.

Fig. 4 Separation-based detection of metabolites in neuronal cells.

(A and B) Separation by capillary electrophoresis and fluorescence detection of metabolic products from the glycosphingolipid LacCer labeled with BODIPY dye, shown at full scale (A) and expanded scale (B). Unknown components are marked with “?”. Vastly different levels of metabolites are found for different cells. [Reproduced, with permission, from (21)]

Combining CE separation with MS detection is also a very powerful technique (20, 44, 48). For example, a sheathless CE-ESI-MS interface allowed detection of close to 20 metabolites in extracts of E. coli with sensitivities ranging from 20 nM (~0.8 fmol) for ADP-ribose to 2.5 μM for α-ketoglutarate (48). Qualitative and quantitative metabolomic investigation has also been reported for single neurons by CE-ESI-MS (20, 44, 87).

Discussion and Outlook

The chemical signature of phenotypic heterogeneity in cellular populations is a property that can only be assessed by single-cell measurements. It appears to be an evolvable trait, with the metabolome being the most immediate indicator of how individual members of a cell culture react to a stimulus or to environmental stress. Biological insight can already be obtained by precise analysis of cell conditions (outside and inside, e.g., by optical microscopy) and comparing these data with metabolomics data. The underlying cause of phenotypic heterogeneity, however, is hardly metabolic, but rather genetic, epigenetic, or based on stochastic processes. It is now well established that phenotypic heterogeneity enhances rare-cell survival (88, 89); in other words, “extreme” phenotypes may rescue an entire population that suffered a major chemical or environmental challenge (such as a dose of medication). Metabolomics is expected to be a good way of identifying such cells, although genomic and proteomic data, ideally from the same cell, should also contribute. However, to extract the genome, the proteome, and the metabolome from one and the same cell currently poses great experimental difficulties; to date, there are no experimental protocols that allow this to be accomplished. An approximation may be to grow a microculture from a single or a few cells that have survived a certain perturbation, and perform metabolomics along with genomics, transcriptomics, and proteomics on aliquots of this microculture. The question is then whether heterogeneity will develop in this microculture, too, and how rapidly. This question can be answered using high-throughput, single-cell analyses (Fig. 5).

Fig. 5 Schematic representation of high-throughput screening of many individual cells exhibiting three different phenotypes (white, gray, black).

After applying stress (e.g., nutrient restriction or drug treatment), only the black cells survive. A new microculture from these survivor cells may develop either into a population with a uniform phenotype (all black offspring cells) or—more likely—into a population with diverse phenotypes.

MS and separations are clearly the most successful single-cell metabolomics methods, both in terms of coverage of the metabolome and speed. Hundreds of metabolite signals have been discerned, although usually from fairly large cells such as snail neurons or plant cells. The identification from smaller cells (yeast, bacteria) is much more difficult. Identification is relatively tedious: Comparisons of elution time of metabolite and standard compound in electrophoretic measurements, and of signals obtained from large numbers of cells (including comparison to standards in MS/MS), are required. Unfortunately, both MS and CE are destructive; fluorescence and spectroscopic methods that detect metabolites nondestructively only cover single or very few metabolites. Moreover, with the exception of some high-resolution imaging methods, current single-cell analyses cannot distinguish molecular locations of compounds in subcellular compartments such as the membrane, cytoplasm, or nucleus. This leads to averaging over a single cell and in that sense presents a limitation of current methodology.

There are a number of further requirements for single-cell metabolomics to become truly useful for systems biology and medical diagnosis: (i) More extensive coverage of the metabolome. Even for model organisms such as yeast [the size of the yeast metabolome is 1168 compounds (90, 91); in the yeast metabolome database,, 2027 small molecules are listed], the coverage of the best methods currently available is <10% of the full metabolome. (ii) Faster identification of metabolites from single-cell data, and in general measurements with high throughput. The recently introduced mass cytometry method (92), in which antibodies marked with unusual elements (lanthanum, cerium, erbium, ytterbium, etc.) are used to encode particular antigens of cells, could be extended to metabolites if suitable antibodies or aptamers can be found. (iii) Protocols for discovery of new or unknown metabolites. (iv) Nondestructive metabolite measurements. A possible strategy for this last challenge would be a two-step approach, where a metabolomic method with a large chemical scope is first used to pinpoint key compounds that indicate, for example, a disease state, followed by development and implementation of nondestructive fluorescence nanosensors that specifically target these key compounds.

References and Notes

  1. Acknowledgments: I thank A. Ibáñez and M. Zenobi-Wong for critical reading of this manuscript and for helpful suggestions, and J. Krismer and J. Sobek for making data available for Fig. 2. I am also indebted to a number of talented individuals: A. Amantonico, S. Fagerer, M. Heinemann, A. Ibáñez, J. Krismer, M. Pabst, R. Steinhoff, and P. L. Urban have all contributed substantially to the success of the single-cell metabolomics project in the author’s laboratory over the past few years. Supported by the Swiss National Science Foundation, ETH Zürich, and the European Union.
View Abstract

Stay Connected to Science

Navigate This Article