Biochemical Journal

Review article

In vivo biochemistry: quantifying ion and metabolite levels in individual cells or cultures of yeast

Clara Bermejo , Jennifer C. Ewald , Viviane Lanquar , Alexander M. Jones , Wolf B. Frommer


Over the past decade, we have learned that cellular processes, including signalling and metabolism, are highly compartmentalized, and that relevant changes in metabolic state can occur at sub-second timescales. Moreover, we have learned that individual cells in populations, or as part of a tissue, exist in different states. If we want to understand metabolic processes and signalling better, it will be necessary to measure biochemical and biophysical responses of individual cells with high temporal and spatial resolution. Fluorescence imaging has revolutionized all aspects of biology since it has the potential to provide information on the cellular and subcellular distribution of ions and metabolites with sub-second time resolution. In the present review we summarize recent progress in quantifying ions and metabolites in populations of yeast cells as well as in individual yeast cells with the help of quantitative fluorescent indicators, namely FRET metabolite sensors. We discuss the opportunities and potential pitfalls and the controls that help preclude misinterpretation.

  • imaging
  • metabolic signalling
  • metabolic pathway
  • sugar signalling
  • yeast


In recent years, ionomics and metabolomics have made major steps forward [15]; however, further progress in this area is hampered by the lack of information on the spatial and temporal distribution of metabolites within and between cells. A similar situation existed for transcriptomics, which was overcome by the development of tools for systematic analysis of gene expression at the cellular level such as cell sorting of fluorescently tagged cells [6], TRAP-seq [7] or INTACT (isolation of nuclei tagged in specific cell types) [8]. Accurate knowledge of the subcellular levels of a given ion, metabolite or signalling intermediate, and many other biochemical and biophysical parameters, is needed in order to obtain a systems level understanding of the cellular processes that involve the analyte. Moreover, we still lack knowledge of the function of many genes, even in model organisms with comparatively low gene numbers (i.e. in the range 500–6000 as compared with over 20000 in higher plants and animals). In the yeast Saccharomyces cerevisiae there are still over 900 ORFs (open reading frames) with uncharacterized function, plus a similar number of ORFs classified as ‘dubious’. In particular, many metabolic signalling networks still need to be elucidated, an effort that could be significantly advanced using high-throughput screening methods [911]. Thus new tools are required for advancing and accelerating the speed of gene discovery, even in organisms such as Escherichia coli or yeast. Analysis of the metabolome and ionome at the cellular and subcellular level, and with high temporal resolution, promises to provide novel insights into metabolic networks and into metabolic control. Such ‘in vivo biochemistry’ data made possible with new analytic tools are essential for systems level analysis of biology in plants, animals and micro-organisms. Both in single cell organisms as well as multicellular organisms, these tools will help us to understand whether and how individual cells differ in their biochemistry and how they render decisions in a complex environment [1215]. For a detailed review of methods available for single cell analysis, see Amantonico et al. [16].


Fluorescence of chemical dyes and organic chromophores has been widely used as a tool in biology. Over the past 20 years, FPs (fluorescent proteins), specifically variants of the GFP (green FP), have revolutionized not only the field of cell biology, but have had an impact on almost all areas of biology [17]. Fluorescent proteins respond in a very specific manner when excited by light: photons from adequate excitation light can be absorbed by an electron in the fluorophore, raising the energy level of the electron to an excited state. During this short excitation period, some of the energy is dissipated by molecular collisions or transferred to a proximal molecule, and then the remaining energy is emitted as a photon to relax the electron back to the ground state. Because the emitted photon usually carries less energy and therefore has a longer wavelength than the excitation photon, the emitted fluorescence can be distinguished from the excitation light.


In the context of the quantitative analysis of ions and metabolites, FRET (sometimes also called fluorescence resonance energy transfer) sensors are easy to use fluorescence tools that simply require determination of the relative emission intensities of a pair of fluorophores (Figure 1). The underlying physics is more difficult to understand as it refers to a quantum mechanical effect between a given pair of fluorophores, i.e. a fluorescent donor and an acceptor, where, upon excitation of the donor, energy is transferred from the donor to the acceptor in a non-radiative manner by means of dipole–dipole coupling [18,19]. As a result of FRET between a defined donor and an acceptor that is in close vicinity (see below), part of the energy absorbed by the donor is emitted in a spectral window that is characteristic for the acceptor. The efficiency of the energy transfer (E), which is defined as the fraction of the photons absorbed by the donor and transferred to the acceptor, is a function of the inverse sixth power of the distance (r) between the two fluorophores: E=R06/(R06+r6). The distance at which energy transfer is 50% of the maximum is defined as the Förster distance, R0. R0 is not universal, but is a unique property of a specific FRET pair, e.g. ~5 nm for eCFP (enhanced cyan FP) and eYFP (enhanced yellow FP). R0 depends on the extent of spectral overlap (overlap integral) between donor emission and acceptor absorption, the quantum yield of the donor QD, the refractive index (n) of the medium and the relative orientation of the dipole moment (κ) of the donor and acceptor: R0=9.78×103 [Qd κ2·n−4·J]1/6 Å, [18,20]. A nice tool for calculating FRET efficiency can be found at Because of its dependence on molecular proximity, FRET has been described as a molecular ruler [21,22], which typically operates in a range of 1–10 nm, a distance relevant for most molecules engaged in complex formation or conformational changes. Although the contribution of the dipole orientation compromises FRET as an accurate measure of molecular distance, FRET is capable of resolving molecular interactions and conformations with a spatial resolution exceeding the inherent diffraction limit of conventional optical microscopy [18]. For a detailed description of the role of the dipole orientation, see van der Meer [23] and van der Meer et al. [24].

Figure 1 Schematic models of FRET sensors for metabolites

(A) The recognition element is depicted as a round or ellipsoid shape, its ligand (analyte) is depicted as a peanut shape, the short-wavelength fluorophore (SWF) is a blue barrel, the long-wavelength fluorophore (LWF) is a yellow barrel and dark grey bars represent linkers. (B) Excitation with short wavelength (e.g. indigo for CFP/YFP sensors). Indigo excitation provides maximal excitation of CFP, and thus high cyan emission intensity, but provides minimal excitation of YFP leading to minimal yellow emission (also called bleed-through emission). (C) Excitation with long wavelength (e.g. green for CFP/YFP sensors). Green excitation provides minimal excitation of CFP, and thus minimal cyan emission intensity, but provides maximal excitation of YFP leading to high yellow emission intensity. (D) Upper panel: fusion of the SWF and LWF via a rigid linker (theoretical). When the CFP is excited with indigo light and the fluorophores are in Förster distance (1–10 nm), some of the energy is transferred from the SWF to the LWF, resulting in reduced cyan emission and increased yellow emission. Lower panel: SWF and LWF are fused to the N- and C-termini of the recognition element. In this example, in the absence of the analyte the energy transfer is ~75%. Note that the peptide bonds in the linker are rotatable. Thus since typically many molecules are analysed over comparatively long time intervals relative to FRET, the fluorophores will occupy a wide range of positions, leading to averaging. When the analyte binds, conformational changes occur that can either lead to lower energy transfer, as shown in the example here (~50%), or can lead to increased energy transfer (not shown). Structural information on the recognition element does not appear to help in sensor design, since observed ratio changes do not typically follow predictions. w/o, without.


It is easy to understand from the above that FRET between two spectral variants of GFP that are fused via a sequence-specific protease cleavage domain can be used to monitor protease activity [25,26]. In the absence of the protease, the two FPs will be close enough to exhibit FRET. Once the protease cleaves the fusion, the two FPs will diffuse away from each other, leading to a loss of FRET. The measurement is comparatively simple: a shorter wavelength donor (for example eCFP) is excited at 434 nm, and in the absence of FRET emits light with maxima at 477 and 501 nm. When eYFP is in Förster distance, FRET will lead to reduced emission of eCFP and detectable emission of eYFP with a maximum at 527 nm. The ratio of eYFP/eCFP will thus decrease proportionally to protease activity. A FRET tension sensor uses the same principle, i.e. detection of a change in r, in a slightly different way. In this case, the two FPs are fused to each other via a flexible domain and each of the FPs is additionally coupled to a cellular protein [27]. When the tension increases, the distance between the FPs increases, and consequentially FRET decreases. FRET sensors for small molecules such as ions or metabolites work on a slightly different principle (Figure 1). Here, the two FPs are attached to an analyte-binding protein, domain or polypeptide termed the recognition element. Conformational changes caused by binding of an analyte of interest result in FRET changes that serve as a proxy for the analyte concentration; when analytes bind the recognition element, the position of the two FPs relative to each other changes, leading to a change in distance and orientation of the two FPs, and thus to a FRET change. It is of course critical that the properties of the binding protein used as a recognition element are suitable for the desired in vivo biochemistry; specifically, the affinity of the recognition element has to match the range in which the ion or metabolite fluctuates in the compartment of interest. Analytes can be, for example, ions, metabolites or signalling intermediates. It is important to note that the measurements performed here do not use single molecule analysis and rather rely on ensemble behaviour. Flexible domains are thus expected to lead to higher averaging of the signal, compared, for example, with intramolecularly inserted FPs that are attached at both their N- and C-termini to the recognition element [28]. In accordance with this, the large analyte-induced conformational change observed for a common type of recognition element, the family of periplasmic binding proteins, does not appear to be relevant [29]. This suggests that the orientation factor κ is much more important. One possibility is that the FPs scan for changes in the surface properties of the binding proteins; a small change in these surface properties translates into a change in the probability space in which the fluorophore can move, thus affecting FRET. Since at present such surface property changes cannot be rationally designed, both the design and optimization processes for sensors are empirical. The success rates are, however, reasonably high; in a number of cases, the first constructs made yielded usable FRET sensors, and in other cases, variants had to be constructed. The big advantage of such a subtle mechanism is, however, that proteins that undergo only subtle conformational changes upon ligand binding can be used in FRET sensors.

One of the major advantages of these FRET sensors is that they can be encoded genetically, and thus are relatively easy to introduce into many bacteria, fungi, plants and animals. The proteins can be expressed in specific cells using cell-specific promoters. Moreover, the proteins can be fused to a signal sequence in order to target them to specific cellular compartments or even to the surface of specific membranes [26,3033]. The temporal resolution is limited only by the exposure time and the on/off rates for the binding of the analyte to the recognition element and can reach the sub-second range (e.g. 120 frames per second and 2 ms [34]).


FRET sensors can be used to quantify the metabolite content in complex solutions such as beer or soy sauce [35,36], determine blood glucose levels [37], or follow metabolite accumulation for example in E. coli, yeast cells or in intact plant roots [3842]. FRET sensors have successfully been used to measure ribose or tryptophan transport in mammalian cell lines, as well as to characterize the transport of glucose across the ER (endoplasmic reticulum) membrane, or the release of glutamate from neurons [33,4345].

Importantly, the FRET sensors can be used for systematic mutant screens (see below), and to identify genes that encode yet unknown functions. For example, a novel expression clone screening system based on the use of FRET glucose sensors in mammalian HEK-293T [human embryonic kidney-293 cells expressing the large T-antigen of SV40 (simian virus 40)] cells was developed [33]. The same FRET glucose sensors used to study transport of glucose across the ER membrane in mammalian cells [33] were then used to screen a clone library for transport activity. This system was successfully used to identify the SWEETs, a novel class of sugar transporters found in plant, Caenorhabditis elegans and human genomes [46].

The simplest type of analysis involves the determination of steady-state ion or metabolite levels in specific cells or cellular compartments. Since it is possible to monitor ion and metabolite levels with FRET sensors in real time, the sensors also provide information on rate changes, for example the rate of accumulation or elimination of a sugar in/from a cell as affected by cellular uptake and efflux rates, biosynthetic and metabolic rates, and rates of compartmentation. For further detail, the reader is referred to an extensive review of the potential use of FRET sensors for the analysis of metabolites [32].

Neurobiologists were the first to implement FRET sensor tools, specifically a suite of FRET calcium sensors. In the meantime, these sensors have been used in the brains of transgenic worms, flies and mice [4750]. Plant biologists also adopted the technology early on, whereas the use of such sensors in micro-organisms is still in early stages [38,42,5155].


Genetically encoded sensors for calcium

The development of calcium FRET sensors has been driven mainly by interest in monitoring neural activity, specifically action potentials and the activity of glutamate receptors that generate transients of Ca2+ during synaptic transmission. Although synthetic indicators combine high Ca2+ sensitivity with a relatively rapid association/dissociation rate, the injection procedure is damaging and invasive, and the dyes are difficult to target to specific cell types and subcellular localizations, such as pre- or post-synaptic sites. Several GECIs (genetically encoded calcium indicators) have been generated to resolve rapid Ca2+ fluctuations with kinetics approaching those of synthetic dyes. The latest generation of GECIs combines brighter FPs, improved kinetics, improved sensitivity, better signal-to-noise ratio, low toxicity, compatibility with in vivo imaging and easy targeting to cellular compartments. They are, therefore, good alternatives to synthetic dyes. To our surprise, we did not find any references describing the use of GECI sensors to study calcium signalling in yeast. GECI application in yeast could substantially benefit the understanding of Ca2+ signalling and Ca2+-dependent processes in this organism.

The most successful Ca2+ recognition element used in the development of GECIs is the EF hand motif. Unlike structural EF hands, regulatory EF hands such as those in CaM (calmodulin) or TnC (troponin C) will undergo a conformational change upon the binding of calcium. The canonical EF hand motif contains a 12-residue pentagonal bipyramidal Ca2+-binding site (for more details see [56]). EF hands generally appear in pairs in nature (TnC and CaM are normally formed by four EF hand motifs).

The Cameleons were the first Ca2+ FRET sensors used to measure in vivo Ca2+ dynamics; it was used to visualize free Ca2+ dynamics in the cytosol, nucleus and ER of individual HeLa cells [57]. The recognition element of the original Cameleon consisted of CaM fused to its binding peptide M13, this fusion being flanked by eBFP (enhanced blue FP) and eGFP. Different amino-acid substitutions on the CaM-binding loop led to generation of Cameleon versions with Kd values ranging from 70 nM to 700 μM. Several modifications in FPs followed: switching to CFP/YFP to reduce FP bleaching, using Citrine as the acceptor FP to reduce the pH sensitivity while maintaining proper folding at 37 °C, and incorporating cpGFP (circularly permutated GFP), which greatly improved the signal-to-noise ratio. Despite several improved Cameleon versions, there are still important physiological problems caused by the use of CaM-derived sensors such as inactivation of the sensor by high endogenous cellular levels of CaM or phenotypes caused by interaction of the sensor, with cellular CaM-binding proteins.

To avoid interactions with CaM-binding proteins when using CaM-derived sensors, calcium FRET sensors based on TnC were generated. TnC specifically regulates muscle contraction, reducing the potential interference by important cellular factors such as CaM. Multiple parameters have to be considered to evaluate and compare different sensors. In the present review we use a simple metric to compare ratiometric sensors, namely Rmax/Rmin (Rmin is the ratio in the absence of analyte, Rmax is the ratio when the sensor is saturated). A high Rmax/Rmin ratio may indicate that a given sensor may provide not only a large dynamic range, but also high signal-to-noise ratios. However, when selecting the best possible sensor for an application, there are many critical factors (e.g. signal-to-noise ratio), and it is important to note that ratiometric and intensity-based sensors have fundamentally different properties and each may have advantages for a specific application. The Rmax/Rmin of Cameleon D3cpv (a CaM-based sensor) is ~5.1 and the sensor can be used in a Ca2+ range between 0.01 to 10 μM [58]. For the TnC-based sensor TN-XXL, Rmax/Rmin is ~3.7, covering a Ca2+ range from 0.01–40 μM. TN-XXL is the latest improved version of the original TN-L15 using TnC from chicken skeletal muscle [59,60]. As for the Cameleon FRET sensors, the improvement process in the TnC-derived sensors also included mutagenesis in the EF hand motif, leading to variants with different affinities, and also changes in the fluorophores with Citrine leading to the fastest rise and decay signal. Additional mutagenesis cycles were necessary to eliminate the Mg2+-binding properties of TnC by eliminating the acidic amino-acid side chains (z-acid pairs) of EF hands III and IV [59].

Pericam [61] and the GCaMP sensor series [62] are Ca2+ sensors using an environmentally sensitive single fluorophore that exhibits a calcium-dependent change in fluorescence intensity. In Pericams and the GCaMPs, a switch in the protonation status of the fluorophore causes changes in fluorescence, thus these sensors report changes in both Ca2+ levels and pH [56]. GCaMP3, an improved version of the original GCaMP, has been reported to show higher signal-to-noise ratio and photostability compared with other Ca2+ FRET sensors [47,63].

Genetically encoded sensors for transition metals

Transition metals are present as trace elements in the environment and play central roles in all living cells as cofactors for enzymes, as redox acceptors or as structural elements necessary for adequate folding of proteins. Organisms such as bacteria, fungi and plants take up transition metals from the environment and concentrate metals such as iron (Fe), copper (Cu), manganese (Mn) and zinc (Zn) by several orders of magnitude [64]. Most probably as a means of protecting cellular mechanisms from damage by metal-induced free radical formation, these metals are highly compartmentalized in eukaryotic cells and can even differ between adjacent cells in the same tissue [14,15]. Therefore metal concentrations vary from femtomolar to micromolar concentrations between subcellular compartments. Inductively-coupled plasma MS has revolutionized the analysis of metal ions in yeast and plant cells [65]; however standard ionomics does not provide for cellular and subcellular resolution and has limited temporal resolution. Synchrotron X-ray fluorescence and X-ray microanalysis only provide static information on cellular and subcellular ion levels [15,66]. To be able to measure metal levels dynamically in the cytosol, synthetic fluorescent sensors such as Zinpyr have been used, but their use is limited by cell permeability and targeting to the desired subcellular compartment [67]. Genetically encoded FRET sensors provide an opportunity to measure dynamic changes in subcellular ion levels over a wide concentration range [68]. Critical considerations for designing metal FRET sensors include ion selectivity (i.e. competition with more abundant cations such as Ca2+ or Mg2+) as well as selectivity for speciation forms (e.g. Fe2+/Fe3+). Moreover, a suite of sensors with a range of affinities is required to match the broad range of metal levels in different subcellular compartments. A variety of transition metal sensors have been constructed and FRET-based zinc sensors have successfully been implemented for studying zinc partitioning in yeast cells (see below).

Genetically encoded zinc sensors

Zinc is an important transition metal in all organisms. At least six different sets of FRET-based zinc sensors have been developed [6978].

A simple approach for designing a divalent transition metal FRET sensor has been the use of a hexahistidine tag added either to the N-terminus (His–CLY9 sensor), or to both N- and C-termini (CLY9–2His sensor) of an eCFP–linker–eGFP fusion protein. In these examples, a flexible peptide consisting of glycine and serine residue repeats was used as the linker [71]. Addition of Zn2+ induced dimerization of the His–CLY9 sensor, whereas in the case of the CLY9–2His sensor, Zn2+ induced the formation of an intramolecular complex. In both cases, zinc binding led to a measurable change in the FRET ratio [71]. Since hexa-histidine tags are not specific for zinc, changes in other metal ions may affect in vivo measurements. In an alternative approach, Zn2+-binding sites were engineered into fluorescent proteins to produce sensors [e.g. ZinCH-x or (Y39H–S280C)–CFP/YFP) [72,73]. Addition of zinc to these sensors in vitro induced 2–4-fold (Rmax/Rmin) FRET changes. Although not all of these engineered sensors had a high selectivity for zinc, they could still prove useful for in vivo analyses.

Two sets of Zn2+ FRET sensors have been constructed using C2H2 ZF (zinc finger) domains of Zap1 or Zif268 transcription factors. The Zap1 set was originally created to study the binding properties of four ZF domains, and were named ZF1/2 and ZF3/4 sensors [74]. ZF1/2 and ZF3/4 sensors were used to monitor changes in cytosolic Zn2+ levels of yeast grown in various zinc concentrations. To our knowledge, this is the first description of the use of FRET sensors for monitoring ion levels in yeast. For the analysis of zinc levels with the FRET sensor, the authors developed a fluorimetric assay in which they analysed zinc levels in populations of yeast cells. Although the ZF1/2 sensor was expressed under a strong constitutive yeast promoter (~50000 sensor molecules per cell), zinc homoeostasis in these cells was not affected as inferred from expression of a zinc reporter gene, and the sensor did not appear to compete for Zn2+ bound to endogenous proteins in yeast [74]. Given that the in vitro binding affinity and ion selectivity of ZF1/2 have not been determined, and due to its Rmax/Rmin of 1.3 being relatively close to 1 compared with the eCALWY sensors described below, the ZF1/2 sensor may require further optimization before becoming suitable for quantitative in vivo zinc analysis. The second set of ZF domain FRET sensors, Cys2His2 and His4, were constructed with the ZF domains from the mammalian transcription factor Zif268 [75]. Cys2His2 was used to monitor in vivo Zn2+ levels in the cytosol and in mitochondria of HeLa cells. The isolated ZF domain had an affinity of 10 nM for Zn2+. However, sandwiched between the fluorophores, Zn2+ affinity was reduced to the micromolar range. Despite the relatively low affinity, cytosolic Zn2+ levels appeared to be affected by the expression level of the Cys2His2 sensor [75].

In an attempt to develop a copper sensor, van Dongen et al. [76] constructed a sensor that responds to Zn2+, Pb2+, Co2+ and Cd2+. This sensor is based on the delivery of copper by the Atox1 chaperone to the Golgi membrane Cu-P type ATPase ATP7B and uses both scaffolds to coordinate copper [76]. Surprisingly, optimization of this sensor resulted in high-quality zinc sensors. The original sensor consisted of two independent polypeptides, one fused to eCFP, the other to eYFP. Further improvements of the original sensor include: generation of a single protein sensor by fusion of the Cu+-binding domain of Atox1 to the WD4 domain of ATP7B via a flexible domain [77], replacement of eCFP and eYFP by modified Cerulean and Citrine versions, a point mutation that suppresses Cu+ binding, and fine tuning of the linker length to modulate the binding affinity for Zn2+ [78]. Apparently, modification of the linkers affects the relative positioning of the binding pockets of the two zinc-coordinating polypeptides, thus leading to differences in Zn2+ affinity [65]. Altogether, a ‘tool box’ of six high-sensitivity sensors, named eCALWY, was built with Zn2+ affinities ranging from 2 pM to 3 nM and an Rmax/Rmin of 2.4 [78]. These sensors were successfully used to measure Zn2+ levels in the cytosol and in secretory vesicles of pancreatic β cells. Careful analysis of the cells indicates that zinc homoeostasis remained unaffected [78].

Genetically encoded sensors for copper

Recently, a Cu+ sensor based on the Cu+-binding domain of Amt1, a Candida glabrata Cu+ transcriptional regulator, was developed for measuring copper levels. Amt1-FRET is characterized by an Rmax/Rmin of 1.2 and high Cu+ affinity (Kd 2.5×10−18 M). Amt1-FRET also appears to bind zinc and iron although with lower apparent affinity. Amt1-FRET was successfully used in CHO-K1 mammalian cells to monitor Cu+ levels [79].

Many of these ion sensors were initially tested in mammalian or other non-yeast cells; their application to measuring critical ion dynamics in yeast should also prove effective as previously described [74].


Over the past 10 years, an ever-increasing number of FRET metabolite sensors have been developed. At present, sensors for pentoses (ribose and arabinose), hexoses (glucose and galactose), disaccharides (maltose and sucrose) and a variety of amino acids (arginine, glutamate, glutamine and tryptophan) have been published [33,3537,41,44,45,8084] (Table 1).

View this table:
Table 1 FRET sensors for ions and metabolites


Besides calcium, a rich source of FRET indicators has been engineered for other signalling elements such as GPCRs (G-protein-coupled receptors) [85], cAMP [86], cGMP [87] and PIs (phosphoinositides) [88]. Excellent summaries of sensors for signalling processes are available [8992], and this field will not be reviewed further here.


The first implementation of the FRET sensors for in vivo measurements was the expression of the maltose sensor in yeast cells [81]. The dynamic range of the sensor was relatively small, the imaging system was not extremely sensitive and the analysis was complicated by only partial immobilization of the cells (see below). The maltose sensors have in the meantime been further improved [93] and have been used for monitoring maltose levels in yeast cultures. As mentioned above, zinc levels were analysed in yeast cultures using a FRET zinc sensor [74]. We have also used a fluorimeter equipped with an injection device for monitoring rapid glucose changes in yeast cell cultures [39,40].


One of the major advantages of using FRET sensors is that they provide the opportunity to measure analyte levels in individual cells using complex perfusion protocols (e.g. with rapid changes in external analyte levels). Such protocols have successfully been used with mammalian cell lines, which grow adherent to cover slips. Thus analysis of individual mammalian cells using FRET assays is comparatively simple. However, yeast cells cannot be easily immobilized during perfusion experiments. Various approaches to immobilize yeast cells have been used in the past, but most proved inadequate for rapid kinetic studies. Immobilization in agar or alginate does not appear suitable since it would limit exchange rates at the cell surface. We have tried the use of surgical glue as used for Arabidopsis roots [55], as well as coating of coverslips with ConA or poly-L-lysine. However, in all cases cells were only weakly attached to the glass, and cells detached at high perfusion rates or during long perfusion periods. As an alternative, we used commercial microfluidics to trap yeast cells [39,40]. The microfluidic system proved to be effective since perfusion rates were much higher compared with those in typical laminar flow perfusion chambers and cells remained in position over hours of perfusion.


FRET sensors have recently been used to determine the levels of free glucose in the cytosol of S. cerevisiae cells and to identify the hexose transporter Hxt5p as a dominant transporter present during carbon starvation [40]. To determine cytosolic glucose levels, a suite of FRET glucose sensors with affinities covering the nanomolar to millimolar range were expressed in yeast. Cytosolic glucose levels varied with extracellular supply and dropped to the nanomolar level in cells starved for glucose as determined with an ultra-high affinity FRET glucose sensor [40]. Figure 2 shows an example for the analysis of yeast cells expressing the high sensitivity FRET glucose sensor FLII12Pglu700 μδ6 trapped in a microfluidic device [40]; shown is the average FRET change from several individual yeast cells during perfusion with square pulses of glucose at three different concentrations. Using a similar approach, FRET ATP sensors with Kd values of 1.2 and 3.3 mM were expressed to study the cytosolic ATP levels after starvation or after glucose resupply. The cytosolic levels of ATP increased upon the addition of glucose and reached saturating levels at external glucose concentrations of 2 mM [94]. Taken together, these results support four major conclusions: (i) glucose levels can drop several orders of magnitude below the Km of the enzymes responsible for metabolism of glucose, namely hexokinases (~40 μM [95]); (ii) after resupply of glucose to starved cells, cytosolic glucose and ATP levels increased proportionally with increasing external glucose concentrations; (iii) a steep gradient of glucose was maintained across the cell membrane, most probably due to efficient metabolic conversion; and (iv) significant amounts of ‘free’ glucose were detectable over the nanomolar to millimolar range, indicating that substrate channelling, if present, is not preventing detection of free glucose [96]. The almost instantaneous glucose transport capacity of starved yeast cells resulting in rapid accumulation of glucose indicates the presence of a glucose transporter even in cells that have not seen significant levels of glucose for several hours [40]. This transport capacity, termed the ‘ajar pathway’, indicates that glucose-starved yeast cells are prepared for sudden changes in nutrient availability, e.g. under natural conditions, when cells present on a low-carbon medium such as soil are transferred to a glucose-rich environment such as a fruit.

Figure 2 In vivo kinetics of glucose accumulation

In vivo measurement of FRET responses in wild-type cells expressing FLII12Pglu-700 μδ6 trapped in a microfluidic platform. Cells were grown in glucose, glucose-starved in SCeth, trapped in a microfluidic device, washed with Mes buffer and exposed to 4 min pulses of glucose with increasing concentrations (shading). Error bars represent values±S.D. (n=30 cells). For experimental details see [39,40].

It is well known that in the absence of fermentable carbon sources, many of the 18 hexose transporters of yeast are repressed, whereas the HXT5 gene is induced [97,98]. To identify the main players, we carried out an unbiased screen of hxt-knockout mutants using yeast cells expressing the FRET sensor FLII12Pglu-700 μδ6 [40]. Among the 13 available hxt-knockout mutants tested, only the hxt5Δ mutant showed a significant reduction in ajar pathway activity: hxt5Δ mutant cells perfused with 100 mM glucose accumulated glucose more slowly and at lower rates compared with the wild-type strain (Figure 3). The remaining activity indicates that other Hxts also contribute to the uptake potential during starvation [40]. Not surprisingly, the reduction in glucose accumulation coincides with reduced cytosolic ATP levels as determined using a FRET ATP sensor [94] (Figure 3).

Figure 3 Glucose and ATP cytosolic accumulation are reduced in hxt5Δ mutants

Yeast cells expressing either AT1.03 (A) AT1.03YEMK (B) or FLII12Pglu-700 μδ6 (C) were glucose-starved for 5 h in SC−C, washed and transferred to microplates. Glucose was added after time point 2 at the indicated concentrations of 0–10 mM, and time-dependent glucose responses were analyzed in wild-type and hxt5Δ. After glucose addition, fluorescence intensities for CFP and YFP channels were measured for eight additional cycles (each cycle ~100 s). Dark orange labels for glucose (glc) and ATP indicate the ligand being measured using FRET sensors. Error bars represent values±S.D. (n=6 transformants). The x-axis is broken to indicate interruption of the time course during glucose addition (grey bar). For experimental details see [39,40].

Taken together, the FRET glucose sensors provide us with a powerful tool to characterize not only the glucose transport mechanisms, but to also help decipher regulatory networks in yeast. The measurement of cytosolic glucose/ATP levels in yeast cultures can be performed in small volumes (e.g. in 96-well microplates) with an appropriate fluorimeter device in a few minutes at medium throughput. The system can thus be used for functional screens of the yeast knockout collection to identify the regulatory networks that specifically induce HXT5 expression during starvation. Preliminary results from our laboratory indicate that multiple signalling pathways are involved in the activation of Hxt5 [40].


While being minimally invasive, overexpression of a binding protein for a metabolite creates a new buffer in the cell and must necessarily affect the absolute binding status of the metabolite relative to control cells. This has been discussed elsewhere in the context of selecting an appropriate zinc sensor or appropriate calcium dye concentrations for analysis [99,100]. The sensor or dye concentration needs to be sufficient for high signal-to-noise measurements, but should not reach levels that affect intracellular analyte levels. Owing to the buffering capacity of the introduced sensors, they can affect steady-state levels and rate changes of free analyte, e.g. they can eliminate calcium oscillations. This is an inherent problem in this type of experimental observation, namely it is impossible to measure metabolite levels or flux in a completely non-invasive way. It is thus important to be aware of this fact, and where possible to attempt to keep sensor or dye concentrations below the cellular levels of analytes (more difficult of course for signalling molecules), and to determine whether sensor expression leads to side effects. Careful phenotypic analysis and more detailed physiological analyses such as isotopomer-based flux analysis [101] can be used to exclude potential artifacts. For example, behavioural tests in animals expressing calcium or glutamate sensors, or careful growth-curve comparison of cells/organisms expressing various sensors to wild-type controls. In case of problems, optimization of signal-to-noise ratios will allow the use of weaker promoters to reduce interference. Also, the sensors can be expressed under inducible promoters to avoid potential effects on growth and development. Such artefacts can be excluded if the recognition element is derived from the organism to be analysed and if the sensor is used to replace the endogenous protein. Ideally, sensors should be calibrated for each in vivo system. An interesting option is that such sensors may even be used intentionally as tools to modulate cellular metabolite levels.

To determine the effect of expressing FRET glucose sensors at high levels in yeast cells, which in some cases possibly exceeds steady-state levels of glucose in the cell, we measured growth rates, biomass yield, glucose consumption and ethanol production (Figure 4). Despite overexpression from the very strong PMA1 promoter fragment in pDR-GWf1, we did not observe significant differences between cells transformed with empty vector or cells expressing a high- or a low-affinity glucose sensor, namely FLIPglu170nΔ13 (where FLIP is fluorescent indicator protein) and FLII12Pglu-700 μδ6Δ13. Furthermore, we measured metabolic flux using 13C-labelling and did not observe significant differences in the distribution of fluxes in central carbon metabolism (experiment performed once in duplicate; results not shown). We can thus assume that the physiology of the cell was largely unaffected by expression of the sugar sensors.

Figure 4 Effect of glucose sensors on physiology and growth of yeast cells

Yeast BY4743 cultures expressing a high-affinity glucose sensor (FLIPglu170nΔ13, blue) or a low-affinity glucose sensor (FLII12Pglu-700 μδ6, red) or an empty vector (black) were grown in 2% SCD liquid medium [40] for 12 h (mean±SD, n=3). (A) Growth rates were determined by measuring attenuance at D600 increase during exponential growth. (B) Biomass yield (dry weight, plain bars) and ethanol yield (hatched bars) during exponential growth at 30 °C (means±S.D., n=3): glucose and ethanol concentrations were determined in the culture supernatant using commercial enzyme assays (Bioassay); biomass dry weight was determined by filtering and drying culture samples in mid-exponential phase.


As outlined above, the FRET sensors exploit protein conformation as a proxy for the concentration of the metabolite. Obviously, protein conformation is sensitive to a variety of factors, e.g. ionic conditions, pH, redox potential and interaction with other cellular factors. Thus all FRET sensors may respond to a change in the cellular environment even in the absence of a change in concentration of the analyte. The sensitivity of all components of the sensor to environmental changes needs to be considered: (i) effects on the conformation of the recognition element; (ii) effects on the structure of the linkers between the core structure of the recognition element and the core structure of the fluorophores; and (iii) effects on the fluorophores themselves.

In particular, when measuring analytes in compartments that are known to rapidly change their pH such as vesicles or the vacuole, additional controls may be required. In this case, it may be useful to determine the in vitro pH dependency of the sensor. pH changes will have different kinetics compared with the analyte and the effect of pH changes may affect the ratio in the opposite direction relative to the ligand-induced ratio change. In this case it is easy to evaluate the concentration of the analyte even if pH changes affect the response. For some of the metabolite sensors tested so far, a ratio change can be induced by pH because of the differential sensitivity of two fluorophores to the pH, and not because of the change in FRET efficiency [38]. It is thus recommended to use GFP variants such as Citrine or Venus in which pH and halide sensitivity have been minimized [102]. Although even the improved GFP variants are not insensitive to pH and halides, careful analysis of the fluorescence intensity of the two fluorophores can be used to identify pH-induced ratio changes [38]. An excellent control is the use of affinity mutants of a FRET sensor. We have constructed a wide range of affinity mutants by introducing point mutations into the recognition element. For example, for the glucose FRET sensors we have an affinity series with Kd values of 170 nM, 2 μM, 30 μM, 600 μM and 3.2 mM [40,55]. When cells are perfused with different glucose concentrations, observed ratio changes for only some of the sensors is indicative of changes in cytosolic glucose concentrations, because ratio changes caused by pH effects on the fluorophores would be observed for all sensors. The affinity mutants can therefore be used to exclude artefacts [31,45].

pH sensors such as pHluorins and the Ptilosarcus GFP can serve as ratiometric dyes for measuring pH in vivo [103,104]. A tandem fusion of eYFP and eCFP named Clomeleon can be used to monitor chloride changes [105,106]. Variants of GFP with enhanced redox sensitivity (roGFP) report on the oxidation state of the FP and thus the redox potential [107]. These sets of sensors for pH, halides and redox status may serve as additional controls to exclude artefacts and to calibrate the in vivo responses to the in vitro behaviour.


Though it may be useful for many applications to obtain absolute quantification, in most cases relative changes will be sufficient. However, when absolute values are needed, e.g. for metabolic modelling, it is necessary to calibrate the sensors in vivo. Calibration can be achieved by using the combination of: (i) a drug that permeabilizes the cells to determine the FRET ratio at full saturation of the sensor (e.g. for calcium sensors a calcium ionophore, and for zinc sensors a zinc ionophore); and (ii) a high-affinity chelator to determine the FRET ratio for the sensor in its unbound form (reviewed in [108]). Although ionophores and chelators are available for metal ions {TPEN [N,N,N′,N′-tetrakis(2-pyridylmethyl)ethylenediamine] for zinc, or neocuproine for copper}, no such compounds are available for example for glucose [78,79]. Detergents that permeabilize the membrane might represent a potential alternative. However, the results have to be interpreted with caution because the intracellular environment will be altered drastically by permeabilization.


The difficulty of measuring metabolites with cellular and subcellular precision has been a major roadblock. Genetically encoded sensors allow quantitative measurement of steady-state levels of ions, signalling molecules and metabolites and their respective change over time. FRET sensors exploit conformational changes in polypeptides as a proxy for analyte levels. Subtle effects of analyte binding on the conformation of the recognition element are translated into a FRET change between two fused GFP variants, enabling simple monitoring of analyte levels using fluorimetry or fluorescence microscopy. FRET sensors provide an efficient tool for minimally invasive and quantitative analysis of ion and metabolite levels in the cytosol of yeast cells as well as in animal and plant cells. We show here that high- and low-affinity FRET glucose sensors do not appear to affect biomass production, glucose consumption or ethanol yield. The sensors have successfully been deployed for gene discovery. Preliminary experiments demonstrate that the sensors can be used to screen mutant collections. The sensors will therefore also be well suited for drug screening. The sensors can be targeted to subcellular compartments, providing genetically defined subcellular resolution without requiring high optical resolution. The sensors, in conjunction with microfluidic trapping devices, also provide high temporal resolution, thus permitting analysis of rate changes when glucose is added or removed. This technique is simple to use, can be implemented wherever a fluorimeter is available and represents a significant advance over previous techniques. Although fluorimetry provides information averaged over cell populations, microscopy in conjunction with microfluidic trapping devices detects differences between cells. Confocal microscopy ultimately permits observation of gradients or local differences within a compartment.

The FRET sensor technology can provide us with functional information under minimally invasive conditions. FRET technology yields information that is complementary to data from stable isotope-based fluxomics and can provide us with valuable new insights into metabolic regulation when combined with other large data sets including transcriptomics and interactomics [109,110].


Further improvements in the sensitivity of the sensors and measurement technologies will provide advantages such as increased dynamic range, the option to reduce sensor levels in the cell to minimize potential interference with metabolic flux, and further improved quantification, especially in the context of single-cell analysis or characterization of local concentration changes. Such improvements are possible, e.g. the brightness of eCFP is relatively low with an extinction coefficient of 20000 and a quantum yield of 0.15. Compared with eCFP, AcGFP1 is much brighter with an extinction coefficient of 32500 and a quantum yield of 0.82 ( FRET sensors for glucose have been constructed with AcGFP1/mCherry [37]; however, the energy transfer for this current set of sensors with ratios of ~0.1 is still comparatively low. The impressive success in generating sensors for many ions, metabolites and signalling-related processes indicates that it will be possible to develop sensors for most molecules of interest, and, in combination with other technologies, help to significantly advance our understanding of metabolism and its regulation.


This work was made possible by grants from the National Institutes for Health/National Institute of Diabetes and Digestive and Kidney Diseases [grant number 1RO1DK079109] and a Marie Curie International Outgoing Fellowship [grant number FP7-people-2007-4-1-IOF].


We thank Stefan Christen and Nicola Zamboni, ETH Zürich, for the [13C] flux analysis.

Abbreviations: CaM, calmodulin; ER, endoplasmic reticulum; FLIP, fluorescent indicator protein; FRET, Förster resonance energy transfer; FP, fluorescent protein; eCFP, enhanced cyan FP; eYFP, enhanced yellow FP; GECI, genetically encoded calcium indicator; GFP, green FP; ORF, open reading frame; TnC, troponin C; ZF, zinc finger


View Abstract