A Dual Nanosensor Approach to Determine the Cytosolic Concentration of ATP in Astrocytes

Adenosine triphosphate (ATP) is the central energy carrier of all cells and knowledge on the dynamics of the concentration of ATP ([ATP]) provides important insights into the energetic state of a cell. Several genetically encoded fluorescent nanosensors for ATP were developed, which allow following the cytosolic [ATP] at high spatial and temporal resolution using fluorescence microscopy. However, to calibrate the fluorescent signal to [ATP] has remained challenging. To estimate basal cytosolic [ATP] ([ATP]0) in astrocytes, we here took advantage of two ATP nanosensors of the ATeam-family (ATeam1.03; ATeam1.03YEMK) with different affinities for ATP. Altering [ATP] by external stimuli resulted in characteristic pairs of signal changes of both nanosensors, which depend on [ATP]0. Using this dual nanosensor strategy and epifluorescence microscopy, [ATP]0 was estimated to be around 1.5 mM in primary cultures of cortical astrocytes from mice. Furthermore, in astrocytes in acutely isolated cortical slices from mice expressing both nanosensors after stereotactic injection of AAV-vectors, 2-photon microscopy revealed [ATP]0 of 0.7 mM to 1.3 mM. Finally, the change in [ATP] induced in the cytosol of cultured cortical astrocytes by application of azide, glutamate, and an increased extracellular concentration of K+ were calculated as −0.50 mM, −0.16 mM, and 0.07 mM, respectively. In summary, the dual nanosensor approach adds another option for determining the concentration of [ATP] to the increasing toolbox of fluorescent nanosensors for metabolites. This approach can also be applied to other metabolites when two sensors with different binding properties are available.


INTRODUCTION
The metabolism of the brain is realized by a joint effort of all cell types including neurons, glial cells as well as cells constituting the blood vessels. Almost all aspects of metabolism in the brain involve several types of cells, including energy metabolism (Pellerin and Magistretti, 1994;Guzmán and Blázquez, 2001; Barros and Deitmer, 2010;Nave, 2010a;Hirrlinger and Nave, 2014;Barros et al., 2018b;Díaz-García and Yellen, 2019;Vicente-Gutierrez et al., 2019;Zuend et al., 2020), and neurotransmitter metabolism (van den Berg et al., 1978;Waagepetersen et al., 2003;Bak et al., 2006;Le Douce et al., 2020). Furthermore, brain cells are structurally intermingled, heavily interdigitating their numerous cellular processes (Somjen, 1988;Grosche et al., 1999;Bushong et al., 2002;Nave, 2010b). Therefore, to study the dynamics of brain metabolism in detail, methods allowing to assess metabolites at high spatial and temporal resolution are required. These requirements are not fulfilled by most of the classical biochemical methods, as well as by techniques like PET or fMRI.
Fluorescent nanosensors for metabolites can be used to monitor relative changes of the respective metabolite within a cell or cellular compartment during an experiment by monitoring the time course of the fluorescent signal. However, the fluorescence signal is often not linearly related to the concentration of the metabolite (or only in a small concentration range). Furthermore, all sensors show a basal, metabolite independent fluorescence. Therefore, to calculate the concentration of the metabolite from this fluorescence signal has remained challenging and several different approaches have been employed (San Martin et al., 2014;Yellen and Mongeon, 2015;Barros et al., 2018a). All of these different approaches have their own advantages, but also limitations (see discussion section and Table 1). We here add a novel approach to the toolbox of fluorescent nanosensors for metabolites, which allows estimating the intracellular concentration of a metabolite under basal conditions by combining two nanosensors for the same metabolite with different binding equilibrium constants. This approach is exemplified using two ATP sensitive nanosensors (ATeam1.03, abbreviated AT, and ATeam1.03YEMK, abbreviated ATY) which differ in their dissociation constants (k D ; Imamura et al., 2009). The feasibility of this approach is shown in cultured cortical astrocytes as well as in cortical astrocytes in situ in acutely isolated brain slices.

Ethics Statement
In accordance with the guidelines for the welfare of experimental animals issued by the European Communities Council Directive (2010/63/EU) and with the German Protection of Animals Act (Tierschutzgesetz), mice were bred in the animal facility of the Medical Faculty of the University of Leipzig. Mice were housed in individually ventilated cages in a specific pathogen free environment in a 12 h/12 h light dark cycle with access to food and water ad libitum. Experiments were approved by the animal welfare office of the Faculty of Medicine, University of Leipzig and the governmental authorities of Saxony (Landesdirektion Sachsen, registration number T20/16; TVV62/15).

Cell Culture and Transfection
Plasmids pDR-GW AT1.03, pDR-GW AT1.03YEMK, and pDR-GW AT1.03R122K/R126K (Bermejo et al., 2010) were obtained from Wolf Frommer (via Addgene;plasmids 28003, 28004, and 28005). The open reading frames of ATeam1.03, ATeam1.03YEMK, and AT1.03 R122K/R126K (Imamura et al., 2009) were subcloned into pDEST expression vectors using Gateway cloning. Primary cortical astrocytes were prepared from the brains of newborn mice of the C57Bl/6J background as described (Requardt et al., 2010(Requardt et al., , 2012Winkler et al., 2017). Cells were seeded on glass coverslips (30 mm diameter) with custom made silicon rings, which allow splitting the surface of the coverslip into two independent chambers (Supplementary Figure 1), and were cultured in DMEM/10% FCS/25 mM glucose for 1 week. The medium was exchanged to DMEM/10% FCS/5 mM glucose and cells were further cultured for at least 7 days with exchange of the medium twice per week. Cells on one coverslip were transfected with expression plasmids encoding ATeam1.03 in one chamber and ATeam1.03YEMK in the other chamber using lipofectamine (Thermo Fisher Scientific, Schwerte, Germany) and the standard protocol suggested by the supplier. Cells were used for experiments 1 day after transfection.

AAV Injection and 2-Photon Imaging in Acute Brain Slices
To image the dynamics of ATP in cortical astrocytes in acute brain slices, ATeam1.03 or ATeam1.03YEMK-open reading frames (Imamura et al., 2009) were cloned into a vector for packaging into adeno-associated viruses (AAV) and astrocytespecific expression was driven by GFAP-promoter elements (Lee et al., 2008;Mächler et al., 2016;Stobart et al., 2018). The plasmids were packaged into AAV serotype 5 by the Viral Vector Facility at the Neuroscience Center Zurich, University of Zurich, Switzerland. 0.5 µl of AAV containing solution (virus titer: 2.2 × 10 12 vg/ml (AT); 2.75 × 10 12 vg/ml (ATY) were stereotactically injected into cortex of 2 to 3 month old mice (coordinates: 0.5 mm caudal to bregma, lateral 2.0 mm, and ventral 1.1 mm), i.e., in the primary somatosensory cortex. 4 weeks later, mice were sacrificed and 250 µm thick acute coronal brain slices were prepared (Pätz et al., 2018). Slices were transferred to the stage of a 2-photon laser scanning microscope (Olympus FV1000) equipped with a XLPlan N 25x/1.05 W objective (Olympus) and a Mai Tai DeepSee laser (Spectra-Physics, Darmstadt, Germany) and continuously superfused at room temperature with aCSF solution (in mM): 130 NaCl, 2.5 KCl, 1 MgCl 2 , 2 CaCl 2 1.25 NaH 2 PO 4 , 26 NaHCO 3 , 10 glucose, pH 7.4. The medium was continuously bubbled with carbogen. Imaging was performed within 500 µm of the site of AAV injection. The following imaging parameters were used: excitation: 810 nm; emission 510 DC XR (F33-511 OD, 266685; AHF, Germany); pixel size: 0.552 × 0.552 µm; 512 × 512 pixels; pixel dwell time: 2 µs; time resolution: 60 s; stack of 23 single z-planes. All experiments were started by perfusing the slice with aCSF for 20 min followed by aCSF containing 1 µM TTX (Tocris, Bristol, United Kingdom) to prevent neuronal activity and to define the sensor signals under baseline conditions. Afterward, glutamate (100 µM in aCSF) was applied for 20 min. At the end of each experiment, 1 mM iodoacetate and 10 mM azide was added to deplete the cells of ATP. All solutions were adjusted to same pH and osmolality.

Data Analysis and Analytical Solution
For a detailed description of the variables and indices used for this calculation see Supplementary Table 1. Regions of interest (ROI), each containing a single cell, were defined manually using Fiji (Schindelin et al., 2012). Background subtracted mean fluorescence intensities averaged over all pixels within a ROI were determined for the FRET (I FRET ) and the CFP (I CFP ) channel. The ratio of the intensities (I) was calculated as I = I FRET /I CFP . The relative sensor signal ratio (R) was calculated as with n H being the Hill-coefficient and k D the dissociation constant. B scales between 0 (at [ATP] = 0) and 1 (at full saturation of the sensor; Figures 1A,B). However, while at [ATP] = 0 both, B and R are equal to 0, R = 1 at [ATP] 0 for each experiment but B = 1 at full saturation of the sensor (Figures 1A,B). As [ATP] 0 within a cell is not known, the value of B 0 at baseline conditions is also not known; therefore, a direct calculation of changes in [ATP] (d[ATP]) from changes in R (dR) is not possible.
If R values at full saturation (R max ) can be measured, it is possible to relate R to the metabolite concentration, as has been done for the ATeam-type ATP sensors using purified sensor proteins (Imamura et al., 2009): Different from this R max approach, the dual nanosensor approach introduced here avoids the necessity to determine R max . Yet, it still relies on knowledge about k D and n H ( Table 1). To calculate [ATP] 0 and [ATP] treat from dR AT and dR ATY measured in the experiments, a system of two non-linear equations was used which allowed deriving analytical solutions: Introducing R according to Eq. 3 into Eqs 4a,b the following equations result (for simplicity "n H " is replaced by "n" in Eqs 5a,b and 6a,b): ) − 1 (5b) Solving Eqs 5a,b as a system of non-linear equations using "Solve" of Mathematica 12.1. yielded the following analytical expressions for [ATP] 0 and [ATP] treat : Of note, this analytical solution requires that n H,AT = n H,ATY , a prerequisite met by the ATeam sensors. If n H,1 = n H,2 , solving the system of non-linear equations results in expressions involving complex numbers. In this case, numerical solving of Eqs 5a,b using "NSolve" of Mathematica is preferable.
Having established the mean [ATP] 0 of all cells observed in a given experiment according to Eq. 6a, R values (Eq. 1) can be correlated to sensor occupancy B. First, B 0 is calculated from [ATP] 0 using Eq. 2. Then, all R values are multiplied by B 0 resulting in R values reflecting sensor occupancy with a range of R = 0 at [ATP] = 0 to R = R max = 1 at saturation of the sensor. Consecutively, [ATP] for each single cell at any time point during an experiment (including [ATP] 0 ) is revealed by

Data Processing and Presentation
Microscopic images were processed using Zeiss ZEN software, Fiji and Corel Draw X4 Graphic. Data were analyzed and calculated using Fiji and Microsoft Excel. Analytical solutions of the system of non-linear equations as well as values for [ATP] 0 and [ATP] treat according to Eqs. 6a,b were obtained using Mathematica. Diagrams were generated using Sigma Plot. In the boxplots, the box spans from the 25th to 75th percentile, the whiskers span from the 10th to the 90th percentile and dots highlighting the 5th to the 95th percentile. In addition, in

RESULTS
The ATP sensors ATeam1.03 (abbreviated AT in the following) and ATeam1.03YEMK (abbreviated as ATY) are two cytosolic, ATP specific nanosensors, which differ in their affinity for ATP (Imamura et al., 2009). Dissociation constants (k D ) of 3.3 mM and 1.2 mM were described for AT and ATY, respectively; for both sensors a Hill coefficient n H of 2.1 was found (Imamura et al., 2009). Therefore, the occupancy of the sensor (i.e., the ratio of ATP bound sensor/total sensor; referred to as B here; see Eq. 2) is different for both sensors at a given cytosolic concentration of ATP ([ATP]; Figure 1A). The [ATP] of a cell under baseline resting conditions ([ATP] 0 ) is a very important parameter for cellular energy metabolism, which is difficult to determine and not known in many cases. We reasoned that a treatment of cells, which results in a given change in [ATP] (d[ATP]), will cause different relative changes of the sensor signal for AT (dR AT ) and ATY (dR ATY ; Figure 1B). For example, at [ATP] 0 = 2 mM a change in [ATP] of −0.5 mM will result in dR AT = −38% and dR ATY = −17%; while at [ATP] 0 = 3 mM the same change in ATP will result in dR AT = −20% and dR ATY = −6% ( Figure 1B). Therefore, at a given dR ATY , the corresponding dR AT correlates to [ATP] 0 , or else, depending on [ATP] 0 a change in [ATP] will result in a characteristic pair of values for dR AT and dR ATY (Figures 1C,D). Accordingly, [ATP] 0 can be derived from pairs of dR AT and dR ATY obtained by treating cells with conditions which change [ATP] ( Figure 1E). Importantly, for this approach prior knowledge of the actual d[ATP] caused by these conditions is not necessary.
To test this approach experimentally, primary cultured cortical astrocytes were studied, which have been reported before to respond to different stimuli with changes in [ATP] . After transfection with either AT or ATY, cells were imaged by epifluorescence microscopy in a custom-made system allowing to image cells transfected with the different plasmids in parallel in the same experiment. Cells readily responded to treatments with a change in [ATP] reflected by both AT and ATY (Figure 2A). At the end of each experiment, cells were incubated with iodoacetate (IA) and azide to block cellular ATP production thereby depleting cellular ATP, allowing to normalize the sensor signal to the sensor signal at The calculation of [ATP] 0 using this dual nanosensor approach depends on knowledge of the k D values of the two versions of the sensor. In the original description of ATP sensors of the ATeam family k D values of 3.3 mM and 1.2 mM (and n H,AT = n H,ATY = 2.1) have been reported for AT and ATY, respectively (Imamura et al., 2009). However, based on calibration of the sensors in astrocytes and neurons in organotypic brain slice cultures, recently k D values of 9.4 mM and 2.7 mM were determined (Gerkau et al., 2019;Lerchundi et al., 2020). The latter studies used a Michaelis-Menten equation for fitting the data, which inherently implies n H = 1.
[ATP] 0 calculated using these parameters ( Figure 2F) revealed 18.10 ± 29.23 mM (n = 9 experiments) and 8.04 ± 1.54 mM (N = 3 conditions). These values are substantially higher compared to previously published data on astrocytes and neurons (1 mM to 4 mM; Fukuda et al., 1983;Ainscow et al., 2002;Mollajew et al., 2013;Rangaraju et al., 2014;Toloe et al., 2014;Pathak et al., 2015), suggesting that this set of parameters might not be fully applicable on the experimental system used here. Despite these uncertainties regarding the k D values, these analyses show that the dual nanosensor approach is well suited to estimate the intracellular cytosolic [ATP] within cells. We note, that for further analysis of the data obtained on cultured astrocytes (Figures 4, 5) the original k D and n H  Two crucial assumptions are inherent in this approach: (a) treatment of the cells with iodoacetate + azide depletes the cell of cytosolic ATP; (b) no other factors, like, e.g., pH, affect the sensor signal during treatment of the cells. To test the validity of these assumptions, experiments were repeated with cells expressing AT1.03 R122K/R126K (abbreviated AT MUT ), a mutated version of the ATeam sensors consisting of the same fluorophores and the same ATP binding protein, but which does not bind ATP due to two point mutations within the ATP binding domain (Imamura et al., 2009). First, the signal of AT MUT is very similar to the signal of AT and ATY during application of iodoacetate + azide (Figure 3A), suggesting that cytosolic [ATP] reaches nominally zero and confirming previous observations (Trevisiol et al., 2017;Winkler et al., 2017). Secondly, when cells expressing AT MUT were treated with azide, glutamate or with increased [K + ] e , only a minor change in the AT MUT signal was observed ( Figure 3B; azide: −0.9 ± 2.9%; n = 285 cells; glutamate: −0.2 ± 2.2%; n = 260 cells; [K + ] e : 0.2 ± 2.2%; n = 265 cells). As AT MUT has  Figure 4).
Astrocytes are a heterogenous cell population which also includes heterogeneity of metabolism (Bittner et al., 2010;Matyash and Kettenmann, 2010;Zhang and Barres, 2010;Bayraktar et al., 2015;Farmer and Murai, 2017;Köhler et al., 2018Köhler et al., , 2019Miller, 2018;Morel et al., 2018;Batiuk et al., 2020). Therefore, information on the heterogeneity of [ATP] 0 within the astrocytic cell population is of major interest. While calculating [ATP] 0 using the dual nanosensor method initially results in the mean [ATP] 0 of all cells included in the analysis, [ATP] 0 for each single cell can be obtained in a second calculation step using experiments; ATY: 1.70 ± 1.06 mM; n = 394 cells from N = 9 experiments). However, the ATY-based calculation resulted in more cells with very high [ATP] 0 , most likely because the ATY binding curve (Figure 1A) is flat in this concentration range and even very small differences in R ATY result in a rather large difference in [ATP] 0 .
Another method to obtain single cell data from fluorescent nanosensor imaging is the single point calibration protocol (e.g., Sotelo-Hitschfeld et al., 2012; Fernández-Moncada and Barros, 2014;Arce-Molina et al., 2020). Only one nanosensor is needed for this approach, and k D and n H are derived from other experimental systems like in the dual nanosensor approach ( Table 1). In addition, an estimate for R max needs to be included in the calculation. For AT, a dynamic range of R max = 2.3 × R min has been reported (Imamura et al., 2009). As no explicit information on R max of ATY is available, only data obtained by imaging of AT were reanalyzed using the single point calibration protocol (Figure 5B). This analysis revealed a higher mean [ATP] 0 and a larger variability of [ATP] 0 compared to the dual nanosensor approach (Figure 5B; 7.00 ± 8.08 mM; n = 346 cells from N = 9 experiments).
Calibration of fluorescent nanosensors to the concentration of the metabolite also allows assessing kinetic changes of the concentration of the metabolite. As an example, the rate of the Finally, to validate the dual nanosensor approach in a more intact system than primary cultured cortical astrocytes, AT and ATY were expressed in cortical astrocytes in vivo by stereotactic injection of AAV vectors, in which expression of the sensor is driven by the GFAP promoter. Acute brain slices were prepared from these mice and imaged using 2-photon microscopy (Figures 6A,B). Application of glutamate induced changes in [ATP] and corresponding values of dR AT and dR ATY were recorded ( Figure 6C). As the two sensors were not expressed within the same mice, no pairing of experiments was possible. Based on the two different sets of k D -values (Imamura et al., 2009;Lerchundi et al., 2020), a basal [ATP] 0 of 0.7 mM or 1.3 mM was calculated, respectively ( Figure 6D), showing that the dual nanosensor approach can also be applied to more intact preparations like acute brain slices as well as to other imaging techniques like 2-photon laser scanning microscopy.

DISCUSSION
Fluorescent nanosensors for metabolites have strongly contributed to a much deeper knowledge on the metabolism and its dynamics of the mammalian brain (and of other organs and organisms; Deuschle et al., 2006;Tsuyama et al., 2013;Yaginuma et al., 2014;Masia et al., 2018;Volkenhoff et al., 2018;Nguyen et al., 2019;Takaine et al., 2019;Arce-Molina et al., 2020;Kioka et al., 2020). As these sensors are proteins which can be genetically encoded, they allow cell type specific expression using specific promoters as well as subcellular targeting using appropriate targeting sequences. Combined with different stateof-the art microscopy technologies, the dynamics of metabolites can be followed in cultured cells, in tissue preparations like brain slices or the isolated optic nerve, but also in vivo in living and even awake animals (Bittner et al., 2011;Ruminot et al., 2011;Mächler et al., 2016;Díaz-García et al., 2017, 2019Trevisiol et al., 2017;Köhler et al., 2018;Baeza-Lehnert et al., 2019;Gerkau et al., 2019;Lerchundi et al., 2019;Arce-Molina et al., 2020;Zuend et al., 2020). However, while these nanosensors readily allow for monitoring relative changes of the metabolite concentration, deduction of absolute concentrations and absolute concentration changes (i.e., in mol/l) during treatments has remained challenging as calibration of the signal of the nanosensors to the actual concentration of the metabolite is hampered by both, theoretical and practical problems (Barros et al., 2018a). We here introduce the dual nanosensor approach, a novel strategy to determine [ATP] 0 using two nanosensors for ATP with different binding equilibrium constants k D .

Properties, Assumptions and Limitations of the Dual Nanosensor Approach
The dual nanosensor approach requires the availability of two nanosensors sensitive to the concentration of the metabolite of interest (here ATP), which differ in their binding equilibrium constants k D ( Table 1). Parallel experiments with cells expressing one or the other sensor reveal the signal change dR of both sensors (Figure 2). If both sensors would differ in their spectral properties, which is unfortunately not the case for AT and ATY (but has been reported for ATP sensors using a single fluorophore; Arai et al., 2018), both sensors could even be recorded within a single cell. Values for k D and n H need to be obtained from other experimental systems, e.g., from measurements of the purified sensor protein (Imamura et al., 2009), assuming that these values can be applied to the indicators in the cytosolic environment of the cells of interest (Table 1). While such transfer of parameters is often used also for other calibration strategies (e.g., Sotelo-Hitschfeld et al., 2012;Fernández-Moncada and Barros, 2014;Arce-Molina et al., 2020), several studies indicate that parameters of various genetically encoded nanosensors can be affected by the cellular environment (Hires et al., 2008;Pérez Koldenkova and Nagai, 2013;Yaginuma et al., 2014;Lerchundi et al., 2019Lerchundi et al., , 2020. Moreover, care has to be taken that these values are determined by appropriate methods including similar or -ideally-identical excitation and emission wavelengths (Pomorski et al., 2013). In addition, R min needs to be measured for each cell during the experiments (Table 1), thereby requiring a method to deplete the cells from ATP. Application of azide, which blocks oxidative phosphorylation at complex IV of the respiratory chain, and iodoacetate, which blocks glycolysis at glycerine-aldehyde-3phosphate dehydrogenase (GAPDH), resulted in a decrease of the AT-and ATY-sensor signal reaching the signal of the ATP-binding deficient AT1.03 R122K/R126K (AT MUT ; Figure 3A), suggesting that the fluorescence signal observed under these conditions is at least close to R min and reflects a concentration of cytosolic ATP of nominally zero. This observation is in line with previous observations using the same treatment to deplete ATP (Trevisiol et al., 2017;Winkler et al., 2017), but also with a different approach blocking glycolysis using 2-deoxyglucose and oxidative phosphorylation by oligomycin (Shulman et al., 2015). Finally, it has to be assumed that a given treatment of a cell results in the same change of [ATP] when either AT or ATY is expressed. This might be of special importance if the concentration of the analyte of interest is within the same range (or lower) as the concentration of the sensor, e.g., in the case of Ca 2+ -sensors. However, [ATP] 0 is at least one to two orders of magnitude higher than the typical concentration of genetically expressed sensor proteins (Barros et al., 2018a), suggesting that the difference in k D of the two sensors does not have a major influence on the dynamics of [ATP].
The dual nanosensors approach adds another option to the list of strategies for quantifying metabolite concentrations from fluorescent signals. The approach eliminates the necessity to determine R max (or dR max ) of the nanosensor ( Table 1) and is applicable to complex tissues like, e.g., the highly myelinated axons in the optic nerve (Trevisiol et al., 2017) or the brain in vivo, because it does not require experimental control of the metabolite concentration within the cell compartment of interest. Furthermore, any stimulation, which results in a reproducible change of the concentration of the metabolite, can be used to calculate [ATP] 0 without prior knowledge of the actual concentration change d [ATP]. Finally, once the mean [ATP] 0 of the observed cells was determined, [ATP] 0 of each single cell can be calculated (Table 1 and Figure 5A). Therefore, this approach will be an interesting option for determining basal metabolite concentrations based on fluorescence imaging.
For calculating the metabolite concentration using the single point calibration method, k D and n H obtained in other experimental systems are needed and R min has to be determined in each experiment ( Table 1) similar to the dual nanosensor approach. While only one nanosensor is needed, additionally an estimate of R max is required. However, R max of, for example, the lactate sensor Laconic and the pyruvate sensor Pyronic differs by a factor of two between purified protein and when expressed in cells (San Martín et al., 2013. Therefore, a determination of R max within the cell of interest is preferable, but requires experimental access to increase the concentration of the metabolite to saturating levels. Such a saturation of the sensor is feasible for metabolites for which endogenous transporters with favorable kinetic properties are expressed allowing equilibration of the extra-and intracellular concentration of the metabolite under appropriate experimental conditions, as exemplified, e.g., for glucose, lactate and pyruvate (Bittner et al., 2010;San Martín et al., 2013Mächler et al., 2016;Arce-Molina et al., 2020). However, cells do not express transporters for ATP. Therefore, one option to increase [ATP] would be to inhibit all ATP consuming processes and to rely on cellular ATP production. However, inhibition of all ATP consuming enzymes is not feasible, and the total amount of adenine nucleotides, which can be phosphorylated to ATP, might be insufficient to achieve sensor saturation. Alternatively, cell membranes are permeabilized without leakage of the sensor protein to allow access of ATP from the extracellular environment (Gerkau et al., 2019;Lerchundi et al., 2020). However, the permeabilizing agent itself or changes of the intracellular ion composition might also affect the sensor signal. Taken together, reliable determination of R max for the ATP sensor in the cell of interest is a difficult task; therefore, one advantage of the dual nanosensor approach is to avoid the need of estimating R max . Reevaluation of our imaging data for the AT sensor by the single point calibration method revealed mean [ATP] 0 = 7 mM (Figure 5B), i.e., higher as calculated by the dual nanosensor approach as well as higher than previously reported for glial cells in culture (1.4 mM; Ainscow et al., 2002), suggesting that the R max value obtained for the purified protein (2.3 × R min ; Imamura et al., 2009) does not reflect R max in cells. Taken together, compared to the dual nanosensor approach, the single point calibration method (Bittner et al., 2010;Mollajew et al., 2013;San Martín et al., 2013;Takaine, 2019;Arce-Molina et al., 2020) provides direct single cell information on the metabolite concentration with only a single nanosensor, but requires an additional estimate of R max (Table 1).
Ideally, nanosensors are calibrated within the cell of interest at the subcellular location of interest by applying the metabolite of interest at (numerous) different defined concentrations without interfering with the cellular environment (full calibration; Table 1). Unfortunately, at present this is feasible only in a very limited set of experimental systems. First, controlling the metabolite concentration within the cell requires good accessibility to the cell of interest which is limited in more complex systems like in axons of a highly myelinated nerve or in the brain in vivo. In addition, access of the metabolite to the inside of the cell requires either endogenous transporters or permeabilization. Such an approach has been used, e.g., for calibration of glucose sensors (Bittner et al., 2010), the lactate sensor Laconic (San Martín et al., 2013), or Peredox reporting the NADH/NAD + ratio (Hung et al., 2011;Hung and Yellen, 2014;Köhler et al., 2018). These studies took advantage of the inherent permeability of the cells to either glucose, or lactate and pyruvate due to the expression of appropriate transporters. In contrast, cell membranes are not permeable for ATP and a plethora of reactions consume ATP within the cell. Nevertheless, also ATP sensors have recently been calibrated in organotypic brain slice cultures using permeabilization of cell membranes (Gerkau et al., 2019;Lerchundi et al., 2020). On the other hand, full calibration of nanosensors in readily accessible systems like purified proteins relies on the assumption that these calibration curves are valid also in cellular and/or more complex systems, but such approaches have successfully been used for obtaining quantitative information in various settings (e.g., San Martín et al., 2013; Fernández-Moncada and Barros, 2014;Mongeon et al., 2016). Furthermore, full calibration of metabolic nanosensors in combination with fluorescence life time imaging (FLIM) allows quantifying metabolite concentrations as, e.g., shown for Peredox, a sensor for the NADH/NAD + -redox ratio, or the glucose sensor SweetieTS (Mongeon et al., 2016;Díaz-García et al., 2017, 2019. In summary, the dual nanosensor approach is a novel method for quantifying [ATP] from fluorescence data with its own advantages and limitations when compared to other methods (Table 1). Nevertheless, for certain applications like, e.g., complex and difficult to access tissues, it will provide an alternative/additional way for obtaining quantitative data.

ATP in Astrocytes
The dual nanosensor approach was employed to determine the basal concentration of ATP in the cytosol of astrocytes both in primary cultures and acutely isolated brain slices.
[ATP] 0 was determined at around 1.5 mM for cultured cells (Figure 2) as well as between 0.7 mM and 1.3 mM for cortical astrocytes in brain slices (Figure 6). These values are well within the range of [ATP] reported for glial cells in culture (1.4 mM; Ainscow et al., 2002) as well as other cells like, e.g., neurons (1 mM to 4 mM; Fukuda et al., 1983;Ainscow et al., 2002;Mollajew et al., 2013;Rangaraju et al., 2014;Toloe et al., 2014;Pathak et al., 2015). However, the values are lower than data obtained from biochemical assays (3-7.5 mM; Schousboe et al., 1975;Silver and Erecinska, 1997; 10 mM calculated from ATP content: 40 nmol/mg protein, Winkler et al., 2017; and cytosolic volume of cultured astrocytes: 4.1 µl/mg protein, Dringen and Hamprecht, 1998). This difference is most likely due to the fact that nanosensors measure the concentration of free cytosolic ATP, while biochemical assays measure the total amount of ATP within a cell including organelles as well as ATP bound to proteins.
Based on the dual nanosensor approach, changes of [ATP] induced by the three different incubation conditions were calculated. Importantly, the resulting d[ATP] did not differ between the different methods of calculation (Figure 4). Furthermore, the values obtained using the calculation from each paired experiment, but also from the pooled data of all experiments yielded very similar results suggesting that this approach is robust against these experimental variables. In cultured cortical astrocytes, glutamate application induced a d[ATP] of −0.16 mM, amounting to about 11% of [ATP] 0 (Figure 4), consistent with previous reports showing a (non-quantified) decrease in [ATP] (Magistretti and Chatton, 2005;Langer et al., 2017;Winkler et al., 2017). Uptake of glutamate released from synapses during neurotransmission is a major task for gray matter astrocytes (van den Berg et al., 1978;Bak et al., 2006) and the associated decrease in [ATP] has been implicated in the stimulation of astrocytic metabolism to support neighboring neurons (astrocyte-neuronlactate shuttle hypothesis; Pellerin and Magistretti, 1994;Voutsinos-Porche et al., 2003).
Another sign of neuronal activity relevant for regulation of astrocytic metabolism is an increase of the concentration of extracellular K + ([K + ] e ; Rash, 2010;Bittner et al., 2011;Ruminot et al., 2019). K + is released from all neurons during repolarization and is, therefore, not restricted to glutamatergic neurons (Rash, 2010;MacVicar and Choi, 2017). An increase in [K + ] e induces glycogenolysis and activates astrocytic metabolism (Hof et al., 1988;Ruminot et al., 2011;Choi et al., 2012;MacVicar and Choi, 2017;Fernandez-Moncada et al., 2018;Köhler et al., 2018). Previously, an increase in the ATY signal induced by increasing [K + ] e from 3 mM to 8 mM or 12 mM was reported for astrocytes in culture and acutely isolated brain slices (Fernandez-Moncada et al., 2018;Lerchundi et al., 2019). Using the dual nanosensor approach and increasing [K + ] e from 5.4 mM to 12 mM, d[ATP] was quantified here as 0.07 mM or about 5% of [ATP] 0 in cultured cortical astrocytes (Figure 4). The K + induced increase of [ATP] is likely limited by the availability of free ADP within the cell and, therefore, the rather small increase in [ATP] might underestimate the level of activation of metabolism.
Finally, when blocking mitochondrial ATP production using azide, a decrease of [ATP] of about −0.5 mM (i.e., −32% of [ATP] 0 ; Figure 4) with an initial rate of d [ATP] of −0.23 mM/min ( Figure 5C) was observed, consistent with our previous biochemical measurements on the same culture preparations . Astrocytes express all enzymes of glycolysis at rather high levels, but are also equipped with the enzymes necessary for fully oxidizing pyruvate and generating ATP in mitochondria (Lovatt et al., 2007). Nevertheless, astrocytes tolerate inhibition of mitochondrial ATP production well by upregulating glycolysis resulting in an increased NADH/NAD + -redox ratio and increased lactate production (Dringen et al., 1993;Bittner et al., 2010;Wilhelm and Hirrlinger, 2011;San Martín et al., 2013;Supplie et al., 2017;Westhaus et al., 2017), consistent with the surprising finding that mice lacking functional mitochondria in astrocytes survive for more than a year without any phenotype (Supplie et al., 2017). However, because numerous mechanisms can contribute to an increase of the glycolytic rate including, e.g., upregulation of glucose transport, upregulation of glycolytic enzymes or activation by allosteric regulators like, e.g., fructose-2,6-bisphosphate, the precise mechanism remains to be elucidated.
In summary, the dual nanosensor approach allows estimating the basal concentration of a metabolite of interest (here ATP) based on the signal changes of two nanosensors with different equilibrium constants. It adds an option to the toolbox for quantifying changes in cellular metabolite concentrations from fluorescence changes, which is particularly useful if determination of R max is difficult or impeded since the dual sensor approach is independent of this parameter. Finally, this method is not only applicable to metabolic nanosensors, but also for other sensors reporting, e.g., pH or the concentration of ions or second messengers (San Martin et al., 2014;Zhang et al., 2018;Bischof et al., 2019;Depaoli et al., 2019).

DATA AVAILABILITY STATEMENT
All datasets presented in this study are included in the article/Supplementary Material.

ETHICS STATEMENT
The animal study was reviewed and approved by Animal Welfare Office of the Faculty of Medicine, University of Leipzig and Landesdirektion Sachsen.

AUTHOR CONTRIBUTIONS
JH conceptualized and designed the study. SK, PF, and UW acquired data. SK, PF, HS, JH, and UW analyzed and interpreted the data. HS performed analysis using Mathematica. JH and UW supervised the project. JH acquired funding. SK, JH, and UW wrote the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

FUNDING
This work was supported by grants from the Deutsche Forschungsgemeinschaft (DFG; HI1414/6-1, HI1414/7-1). The funding sources were not involved in study design, data collection and interpretation, or the decision to submit the work for publication.

ACKNOWLEDGMENTS
JH would like to thank Klaus-Armin Nave, Göttingen, for longstanding collaboration and ongoing support. We thank Eva Kendzia, Christopher Wolff and Marit Sicker for help with cell culture, Grit Marx for help with mouse management, the Medizinisch-Experimentelles Zentrum of the Medical Faculty, University of Leipzig for excellent mouse husbandry as well as the mechanics workshop of the Max-Planck-Institute for Experimental Medicine for developing the silicon devices and flow-chambers used for cell culture. We acknowledge support from the German Research Foundation (DFG) and Universität Leipzig within the program of Open Access Publishing.