Introduction

In this post-genomic era, besides transcriptomics, proteomics and metabolomics, another omics, i.e. fluxomics (the study of flow of metabolites in the metabolic network), is rapidly gaining ground. Except for the most peripheral fluxes, such as uptake and secretion, direct monitoring of fluxes in a metabolic network is not possible and hence inferences are often based on stoichiometric models. Biological systems contain a large number of individual components, which undergo dynamic and interactive responses in time and space. Analysis of these responses can help to unravel the information encrypted in the genome. Emerging technologies that allow for a comprehensive analysis of metabolites are likely to make a vital contribution by detecting many of the phenotypic changes that result from alterations of the genotype (Fiehn 2002). High throughput profiling technologies suffer, however, from a serious blind spot. Although significant progress has been made in understanding the metabolite transport, we still have only a partial picture of the highly dynamic changes of metabolite levels in time and space. We lack knowledge about major parts of the transport machinery, especially about the plant vacuolar importers/exporters, cellular efflux mechanisms, and regulation of transport activities by metabolites. Because pathways may be distributed among cells, the metabolism and metabolite levels in groups of cells and tissues, that form an organ, are not necessarily uniform but possibly differ even between immediately adjacent cells, as has been shown for ion concentrations in leaves (Karley et al. 2000). In the last two decades, three major approaches for metabolic flux analysis, namely, dynamic flux modeling (Yu et al. 1995), flux balance analysis (Vallino and Stephanopoulos 1993) and 13C metabolic flux analysis (Marx et al. 1996), have been developed. Many biological functions involved in the formation of protein–protein complexes are studied by methods based on genetic, biochemical and biophysical concepts. These techniques are unable to address the sub-cellular localization, protein–protein interaction and real time analysis of analytes.

Owing to methodological problems in analyzing the dynamic distribution of metabolites, measurement of metabolite levels is limited primarily to analysis of average concentrations over whole tissues or organs in both animals and plants (Lu et al. 2002). Some recent studies have addressed this limitation. A major advance was the development of single-cell-sap sampling, which enables metabolite levels to be determined and correlated with gene expression in individual cells (Arlt et al. 2001). To partition tissue extracts of metabolites into extracellular and sub-cellular pools, apoplasmic wash fluids were used to determine the averaged levels of metabolites in the apoplasm (Lohaus et al. 2001). Non-aqueous fractionation was developed for compartmental analysis within the cell (Leidreiter et al. 1995). This technique was then combined with mass spectrometry for the analysis of a wide spectrum of metabolites (Arrivault et al. 2014).

None of these methods, however, enables real-time visualization of metabolites. Non-invasive imaging techniques would have significant advantages over these methods in providing a better understanding of the compartmentalization of metabolic reactions, transport, and metabolite sensing. Positron-emitting tracers have been used in combination with positron-emission tomography imaging for spatial resolution in the range of 1 mm (Myers 2001). 13C-Imaging, using indirect detection techniques, permits 13C-labeled compounds to be used as tracers at a similar spatial resolution (Köckenberger 2001). Spectroscopic techniques, nuclear magnetic resonance (NMR) imaging and liquid chromatography-mass spectroscopy (LC–MS) are used for in vivo metabolite imaging (Lanza et al. 2010), but can be applied only to the analytes present in high concentrations (Ratcliffe and Shachar-Hill 2001). Recently, high-speed atomic force microscopy (AFM) has shown promise to provide simple and prompt answers to how and what structural changes occur, even with reference to the working of individual biomolecules. AFM can directly visualize biomolecules in physiological solutions at a nanometer resolution, but in a high-speed mode AFM, tip-sample interaction force cannot be reduced to a negligible level for samples having delicate biomolecular interactions, hence invasive in nature (Ando et al. 2008). Another promising technique for the detection of biomolecules is surface plasmon resonance imaging, which lacks in the spatial resolution and involves destruction of the tissue (Qavi et al. 2009).

The following questions were still unanswered, as was asserted by Looger et al. (2005):

  • Do we have tools to determine the physiological concentration of biomolecules?

  • Can we visualize the concentration of metabolites from cell to organ?

  • How do metabolites change with stimuli?

  • Can we monitor these changes at real time?

Keeping the above in view, Tsein’s group of Howard Hughes Medical Institute, USA (Miyawaki et al. 1997) and Formmer’s group of Carnegie Institution For Science, Stanford, USA (Fehr et al. 2003) attempted to develop fluorescence resonance energy transfer (FRET)-based genetically-encoded nanosensors by using the combination of fluorescent proteins. The genetically-encoded nanosensor can function non-invasively. This review elucidates the development and application of genetically-encoded nanosensors for various metabolites, and their utility in understanding the metabolic and regulatory networks through novel high throughput analyses.

Components of a FRET sensor

The genetically-encoded sensors require (i) fluorescent proteins, (ii) fluorescence resonance energy transfer, and (iii) ligand-sensing domains, as the basic components:

Fluorescent proteins (FP)

Fluorescent proteins are a family of homologous proteins from various organisms, whose major characteristic is to display bright fluorescence in entire visible spectrum (Zhang et al. 2002; Pakhomov and Martynov 2008; Okumoto 2010). Green fluorescent protein (GFP), isolated from Aequorea victoria, is the best known fluorescent protein (Tsien 1998). With plenty of mutants, it has markedly facilitated understanding its photochemistry and photophysics (Fig. 1). The fluorescence of FP involves an internal, autocatalytic and post-translational modification of chromophore. Fluorescent proteins are now the fundamental tools in current cell biology research, because (i) these are genetically encoded, and (ii) it is possible to fuse them to virtually any desired protein target to monitor intracellular processes (Zimmer 2002; Giepmans et al. 2006).

Fig. 1
figure 1

Structural motifs of GFP and its variants

Full size image

Fluorescence resonance energy transfer

Fluorescence resonance energy transfer (FRET) is a distance-dependent physical process by which energy is transferred non-radiatively from an excited molecular fluorophore (donor) to another fluorophore (acceptor) by means of intermolecular long-range dipole–dipole coupling (Förster 1965). It is not always necessary for the acceptor molecule to be a flourophore. FRET can be an accurate measurement of molecular proximity at nanometer distances (1–10 nm) and highly efficient if the donor and acceptor are positioned within the Förster radius. Thus within the range of protein dimensions, FRET efficiency depends on the inverse sixth power of intermolecular separation (Joseph and Lakowicz 1999). It is a sensitive method for investigating a variety of biological activities that produce changes in molecular proximity (dos Remedios et al. 1987). Advances in light microscopy imaging combined with the fluorescent proteins offer the tools necessary to acquire temporal distribution of protein associations inside living cells (Heim and Tsien 1996; Day 1998; Elangovan et al. 2003).

Fluorescence resonance energy transfer as a tool

FRET refers to a quantum mechanical effect between a given pair of chromophores, consisting of a fluorescence donor and a respective acceptor. The primary conditions for FRET, as enunciated by Vogel et al. (2006), include:

  1. 1.

    Proximity of donor fluorophores and acceptor molecule;

  2. 2.

    Emission spectrum of donor overlapping the excitation spectrum of acceptor;

  3. 3.

    Mutually parallel alignment of donor and acceptor;

  4. 4.

    High-quantum yield of donor;

  5. 5.

    High-absorption coefficient of acceptor.

Measurement and interpretation of FRET

FRET can be measured by multiple methods through donor fluorescence; i.e. by the value of acceptor fluorescence, by analysis of emission of both fluorophores, and by the change in emission value of donor fluorophore (Jares-Erijman and Jovin 2003). However, to determine the donor fluorescence in the presence and absence of acceptor is a widely used method. The fluorescence of the donor should decrease in the presence of acceptor, indicating the occurrence of FRET. The emission-intensity ratio of donor to acceptor or vice versa is the most suitable method. Permutation of filters for fluorescent wavelength selection is also critical to the success of detection of FRET (Vogel et al. 2006). Excitation filter for the donor must be able to excite selectively the donor molecule, while minimizing the direct excitation of the acceptor molecule. An appropriate filter set is required to get the correct FRET image. Another important concern regarding the detection of FRET is analyte concentration. Only those molecules that interact with one another will result in FRET; if a large number of donor and acceptor molecules are present but do not interact, the amount of FRET taking place would be quite low. Negative FRET signal does not mean that the two molecules are not interacting with each other; their dipole may not be aligned parallel with each other. Relative emission of the acceptor is calculated upon donor excitation. When the fusion molecules interact, the acceptor fluorophore is brought into vicinity of the donor and the acceptor molecule is excited by resonance energy transfer (Lalonde et al. 2005). The interaction is detected by increased emission of the acceptor, simultaneous with a decrease in donor emission intensity. Various techniques beyond the basic analysis of the relative emission ratio are available for reliable determination of FRET changes (Jares-Erijman and Jovin 2003). Due to the presence of so many variants of the GFP, the use of FRET phenomena is consistently on the increase.

Ligand-sensing domains

An important requirement for the development of FRET-based nanosensor is the selection of a suitable ligand-sensing protein. The ligand-binding proteins, which may be membrane proteins, periplasmic-binding proteins (PBP) and other types of ligand-sensing domain, should possess the Venus-flytrap movement. The hinge-like twisting and bending motion of bacterial PBPs from E. coli make them an ideal ligand-binding protein for the development of FRET-based sensors (Fehr et al. 2002; Mohsin et al. 2013). These proteins are highly versatile and recognize a number of substrates with high affinity and specificity.

These PBPs are the representative members of a protein superfamily, that mediate in chemotaxis and solute uptake responses (Felder et al. 1999). Periplasmic-binding proteins that recognize a huge variety of ligands (carbohydrates, amino acids, anions, metal ions, dipeptides and oligopeptides), have been identified (Tam and Saier 1993).

Ligand-binding site located at the interface between the two domains, these proteins with two domains are linked by the hinge region that has two conformations as shown in Fig. 2. The unbound state is the open form and the ligand-bound forming i.e. the closed form, arises through a large bending motion the hinge region interconvert them (Quiocho and Ledvina 1996). A ligand-mediated structural change should be large enough to perturb the surroundings of an attached fluorophore for the construction of an optical biosensor. Therefore, linked fluorophores can be positioned (i) in the substrate binding site, forming direct contacts with the ligand; (ii) at the perimeter of the binding site, differentially contacting the two domains in the open and closed conformations; and (iii) in a cleft that opens and closes in consonance with the conformational change.

Fig. 2
figure 2

Maltose-binding periplasmic-binding protein; in the absence of maltose, the open conformation is formed and the closed form occurs in the maltose-bound state (Quiocho and Ledvina 1996)

Full size image

Some genetically-encoded FRET-based sensors

Several sensors have been constructed by using the PBPs (Fig. 3) and other membrane proteins as the ligand-sensing domain. FRET-based nanosensors developed so far are described below:

Fig. 3
figure 3

The periplasmic-binding protein superfamily: diversity among a specific set of bacterial PBPs. Here, LBP leucine-binding protein, LIVBP leucine-, isoleucine-,valine-binding proteins, DPBP dipeptide-binding protein, OPBP oligopeptide-binding protein

Full size image

Maltose

Maltose-binding protein (malE), a periplasmic-binding protein without a signal sequence, was used as ligand-binding protein for the development of maltose nanosensor. Enhanced cyan-fluorescent protein (ECFP), acting as a donor, was fused to the N-terminus and the enhanced yellow fluorescent protein (EYFP) was fused to the C-terminus of maltose-binding protein. EYFP acted as an acceptor fluorophore. By changing the concentration of maltose, alteration in the FRET ratio was observed, reflecting maltose uptake. This showed that binding of maltose to MBP brings the two chromophores closer to each other, thus resulting in the FRET. This nanosensor was named as FLIPmal (Fehr et al. 2002). The K d of this sensor is 2.3 μM.

Glucose

Glucose homeostasis is important for human beings. To develop nanosensors for glucose measurement, two GFP variants were attached; CFP at N-terminus and the YFP at C-terminus, to the glucose–galactose binding protein (GGBP) obtained from E. coli. The nanosensor was named as FLIPglu. When glucose binds to the FLIPglu-170n sensor, FRET decreases. The K d of the protein was 170 nM, with the maximum FRET ratio being 0.23. An affinity mutant was created with 600 µM K d, showing the glucose dynamics in COS-7 cells (Fehr et al. 2003).

Sucrose

No specific sucrose-binding periplasmic binding proteins have been found in bacteria. Since micro-organisms in the rhizosphere developed sucrose-sensing protein, several candidates from Rhizobium and Agrobacterium were tested, and a homologue of the rhizobial protein (ThuE) from A. tumefacians was found to bind sucrose (Lager et al. 2006). Sucrose-binding protein, ThuE, was then fused to the ECFP and EYFP. This showed FRET change upon binding to sucrose. The sensor was called as FLIPsuc. The K d of this sensor is 4 μM. Mutant of nanosensors showing different affinity towards sucrose.

Phosphate

A synechococcus phosphate (Pi)-binding protein (PiBP) was fused to ECFP and Venus at the N-terminus and C-terminus respectively. The nanosensor was named as FLIPPi. Upon binding of phosphate to sensor protein, the indicator showed the Pi concentration-dependent increase in FRET (Gu et al. 2006). This nanosensor, with K d of 840 nM, can be used for real time measurement of phosphate in living cells.

Zinc

FRET-based genetically-encoded sensor has been developed to measure zinc, which is a modification of previously-developed sensor CALWY: CFP-Atox1-linker-WD4-YFP, which lacks the large change in FRET ratio. This CALWY has two metal-binding domains with zinc-binding pocket. Cerulean and citrine were used as the donor and acceptor pair and fused at the N- and C-terminus of Atox1-linker-WD4. Upon introducing the mutations on the surface of both fluorescent proteins, FRET efficiency improved. This nanosensor showed a high energy transfer in the absence of zinc, but zinc-binding decreased the FRET ratio up to 2.4 fold. Mutation (C416S) in zinc-binding cysteines weakens the Zn2+ affinity 300 times, thus producing an eCALWY-4 a K d of 0.6 nM. An eCALWY-1-6 was developed that measures in the range of picomolar to nanomolar and displaying zinc affinity a two-fold change (Vinkenborg et al. 2009).

Ribose

For developing a ribose nanosensor, ribose-binding protein (RBP) E. coli was taken and flanked by CFP and YFP, at the N-terminus and C-terminus respectively. It was named as FLIPribose. Binding of ribose to FLIPribose decreases FRET efficiency, which convert into ribose concentration with the help of standards (Lager et al. 2003). The K d of the nanosensor is 11.7 mM.

Calcium

Genetically-encoded FRET sensors for calcium were developed by sandwiching calmodulin with CFP at the N-terminus and Venus at the C-terminus. Donor–acceptor-distance detail provides information about the location of the CaMs within the ion channel. FRET ratio was recorded in resting cells as well as in Ca2+ responsive conditions. This sensor can be used to monitor the calcium level in muscles (Evanko and Haydon 2005).

Adenosine 5- triphosphate (ATP)

Since the half life of ATP is few seconds, developing a sensor was a challenging task. The ε- subunit in bacteria is an ATP-binding protein with one N-terminal β-barrel domain and C-terminal α-helices. Indicators for ATP were generated, using the ε-subunit of bacteria as the ligand-binding domain. The ε-subunit was sandwiched by CFP and Venus at N- and C-termini of the ε-subunit. These nanosensors were named as ATeams. By changing the ATP concentration a dynamic FRET ratio is produced. The K d for ATeams is 3.3 mM (Imamura et al. 2009).

Cyclic adenosine monophosphate (cAMP)

A construct was developed by fusing mCerulean and mCitrine at the N- and C-terminus of Epac1, respectively. With the binding of cAMP to the sensor protein, conformation of mCerulean–Epac–mCitrine altered, resulted in FRET signals. Increasing the concentration of cAMP decreases the FRET ratio and vice versa thereby quantifying the cAMP level at the real time (Salonikidis et al. 2008).

Citric acid

Production of citrate is the first committed step of the tricarboxylic acid cycle. Analysis of citrate dynamics calls for resolving concentrations in different compartments of the cell, which could be achieved only by genetically-encoded sensors. A series of genetically-encoded citrate sensors based on FRET has been developed (Ewald et al. 2011). The citrate-binding domain of the histidine sensor kinase Klebsiella pneumoniae CitA, was inserted between Venus/CFP and produced a sensor which was highly specific for citrate. By modifying the residues in the citrate-binding pocket, seven mutants were created with different affinities, enhancing the physiological detection range of three orders of magnitude without losing specificity. K d, the ligand concentration at half-maximal saturation, is 8 mM (Ewald et al. 2011). In an in vivo application, E. coli maintains the capacity to take up glucose or acetate within few seconds even after a long-term starvation.

Lactate

Lactate is an indication of altered metabolism and participates in the pathogenesis of inflammation, hypoxia/ischemia, neurodegeneration and cancer (Vander Heiden et al. 2009). Tumor cells show increases in lactate production in the presence of O2; this is known as the Warburg effect (Brooks 2009). To measure the flux of the lactate within and between the cells it is required to develop a sensitive tool, which can monitor the activities of lactate inside the cell. A FRET-based genetically-encoded lactate sensor was constructed. LldR, a lactate-binding bacterial transcriptional protein was fused between mTFP and Venus at N- and C-termini, named as Laconic. The sensor protein was capable of quantifying lactate levels from 1 to 10 mM (San Martίn et al. 2013). Like other genetically-encoded nanosensors, Laconic can be expressed in vivo in different cell types and possible to target in sub-cellular organelles.

FRET Sensors for amino acids

A glutamate sensor was developed by using the glutamate-binding periplasmic binding protein, ybeJ, from E. coli. ECFP and Venus (an improved variant of YFP) were used as donor and acceptor, respectively, for the construction of this sensor (Okumoto et al. 2005). This sensor is known as FLIPE. Addition of glutamate caused changes in the emission ratio of Venus to ECFP. FRET ratio changes with the concentration of glutamate. Mutant nanosensors were created by point mutation of the ligand-binding site. The Kd of FLIPE is 600 nM (Okumoto et al. 2005).

A glutamine-binding protein (QBP) from E. coli was used as the reporter element for the development of FRET-based sensor for arginine. The ligand-binding protein was fused to the ECFP and YFP at the N- and C-termini, respectively. This nanosensor undergoes a conformational change upon binding to the arginine largely instead of Gln, thus causing changes in the FRET value. The K d of this sensor is 2 mM (Bogner and Ludewig 2007).

The cellular glutamine concentration is temporally dynamic; its control is not easy and involves a complex process. Because of the multiple levels of regulation and the heterologous functions of different cell types, concentration of glutamine differ among cell types. A FRET-based genetically-encoded sensor for reporting the dynamics of glutamine in living cells was constructed. The periplasmic-binding protein from E. coli was amplified and sandwiched between the monomeric teal fluorescent protein (mTFP) and Venus. The sensor construct was named as FLIPQ-TV. The detection range of the sensor proteins varies from 85 nM to 7.6 mM and demonstrates that FLIPQ-TV is capable in determining the physiological glutamine concentration (Gruenwald et al. 2012).

Leucine is a dietary amino acid that stimulates protein synthesis in muscles (Etzel 2004). It can stimulate mRNA-translation initiation via insulin-dependent and independent pathways and facilitate the muscle protein synthesis. Leucine is unique among amino acids for its important regulatory roles in metabolism, including the translational control of protein synthesis. A FRET-based genetically encoded nanosensor was designed to measure the flow of leucine in the eukaryotic and prokaryotic cells (Mohsin et al. 2013). LivK, a bacterial periplasmic binding protein, was sandwiched by CFP and YFP and expressed in E. coli and yeast. Using this sensor, leucine can be quantified between 8 μM and 1 mM. The nanosensors are useful to study leucine uptake and identify the efflux mechanisms.

A genetically-encoded nanosensor, based on the principle of FRET and fluorescent protein combinations, has been developed by our group. A methionine-binding protein (MetN) from E. coli K12 was used as the reporter element. MetN was sandwiched between CFP and YFP and expressed in bacteria and yeast. The sensor protein was named as the FLIPM. FRET changes were specifically observed after addition of methionine. The recombinant nanosensors showed a concentration-dependent change in the fluorescence ratio. The calculated affinity (K d) of FLIPM was 203 mM. These sensors are stable with pH changes within the physiological range (Mohsin and Ahmad 2014). Data suggest that these nanosensors can be a versatile tool for studying the in vivo dynamics of methionine level in living cells.

Conclusions and future perspectives

Assignment of functions to genes encoded by a given genome and their integration into metabolic and regulatory networks are the primary goals of research in the post-genomic era. For the development of this metabolic network, quantitative information on the cellular and sub-cellular dynamics of ions, signaling molecules, and metabolites is critical. However, there are limits in the analysis of metabolite levels at the cellular and sub-cellular levels in the real time mode, because molecular activities are highly dynamic and can occur locally in sub-cellular domains or compartments. Metabolite levels in neighboring cells in the same tissue can exist in different states. Analyses based on mass spectrometry provide a glimpse of the total metabolite and ionic inventory but lack the temporal and spatial resolutions that are necessary to indicate the local concentrations and flux rates in vivo.

Biosensors, the molecules that report analytes or processes in live organisms or in their environment, are now on the horizon to allow us to quantify the local changes of ions, signaling intermediates and metabolites in real time. Application of protein recognition-based genetically-encoded biosensors has already gained ground in research, clinical, environmental and security operations (Medintz et al. 2006). The FRET-based genetically-encoded nanosensors, that use fluorescent proteins as reporters and measure the dynamics of biophysical processes, ion and metabolite concentrations, and macromolecules in live cells, have been a tremendous success in real time monitoring of metabolites (Okumoto et al. 2005; Kaper et al. 2007; Gruenwald et al. 2012). Our knowledge about conditions for expressing these sensors in transgenic animals and plants and about the possible caveats and complications is rapidly increasing. The designing of these sensors exploits the obligatory conformational changes of bacterial ligand-binding proteins upon ligand-binding and the phenomenon of fluorescence resonance energy transfer of fluorescent protein pair.