Monitoring ATP dynamics in electrically active white matter tracts

1In the vertebrate nervous system, long axons have emerged as the 'bottle neck' of neuronal integrity. In neurodegenerative diseases, axons often reveal first signs of perturbation, e.g. spheroids, long before the pathological hallmarks of disease are seen in neuronal somata, such as large protein aggregates or neuronal cell death. In white matter tracts, Wallerian degeneration of axons is a carefully regulated program of self-destruction that upon reaching a threshold leads to rapid calcium-dependent proteolysis of axonal proteins. This degeneration can be experimentally triggered in different ways, including acute axonal dissection or by blocking mitochondrial energy metabolism. In disease situations, the mechanisms leading to axon loss are much less clear. For example, in spastic paraplegia (SPG), which comprises an entire family of inherited axon degeneration disorders, long myelinated tracts of the spinal cord are progressively lost. However, virtually none of the subforms shows a clear link between the primary defect and the later loss of myelinated axons, with the possible exception of mitochondrial disorders, in which the perturbation of axonal energy metabolism is assumed to underlie a loss of axonal ATP. However, a disturbed energy balance has never been shown. Similarly, in the equally heterogeneous group of Charcot-Marie-Tooth (CMT) neuropathies of the peripheral nervous system only a small subset is caused by mitochondrial dysfunction. In most patients (CMT1A) the primary defect resides in the axon-associated myelinating glia (Schwann cells), and it is unknown whether the axonal energy metabolism is perturbed before the onset of axon degeneration.

2Axonal ATP consumption in white matter tracts largely depends on the electric spiking activity that differs between regions, and it is virtually impossible to biochemically determine ATP selectively in the axonal compartment and under 'working conditions' with a high temporal and spatial resolution. To solve this problem, we have developed a novel approach to simultaneously study action potential propagation and ATP levels in axons of optic nerves of mice, a white matter tract suitable for studying axonal energy metabolism. We established the transgenic mouse line ThyAT, in which the fluorescent ATP-sensor ATeam1.03YEMK shows pan-neuronal expression in vivo. Stimulating electrically optic nerves acutely isolated from those mice, we combined recordings of compound action potentials (CAP) with confocal imaging of the ATP-sensor to evaluate and compare both axonal conductivity and ATP levels. We find that (1) axonal glycolysis is not sufficient to robustly sustain CAPs and physiological ATP levels, but mitochondrial function is needed to provide ATP; (2) during high-frequency propagation of action potentials, CAPs are a well-suited parameter to estimate axonal ATP levels; (3) lactate metabolism is essential for the maintenance of axonal ATP levels, and its inhibition severely affects energy homeostasis of myelinated axons.

3The genetically encoded ATP-sensor ATeam1.03YEMK allows FRET-based monitoring of cellular ATP levels close to real-time. We generated transgenic mice for pan-neuronal in vivo expression of this sensor driven by the murine Thy1.2 promoter. One mouse line with 21 copies of the transgene [B6-Tg(Thy1.2-ATeam1.03YEMK)AJhi, referred to as ThyAT] displayed widespread neuronal expression (Figure 1). In the retina, fluorescence marked ganglion cells and their axons (Figure 1G,H) and expression of the ATP-sensor in myelinated axons appeared robust in all recorded channels (Figure 1I–K).

4Optic nerves from adult ThyAT mice were studied ex vivo, assessing axonal ATP levels by confocal microscopy and simultaneously monitoring stimulus-evoked CAPs (Figure 2). To verify sensor function, nerves were subjected to ATP depletion by blocking mitochondrial respiration with sodium azide (MB) and glucose deprivation (GD). An immediate drop of the FRET signal (Figure 2C,D) and increase in CFP emission (Figure 2C,D) indicated that ATP levels in axons dropped. Changes in pH can modulate YFP-fluorescence. However, ATeam1.03YEMK is almost insensitive to pH within the physiological range and YFP emission upon direct YFP excitation was unchanged (Figure 2C,D; and data not shown), suggesting that pH changes are not the cause of altered FRET signals.

5The ratio of FRET/CFP fluorescence (F/C-ratio) was calculated as a measure for the cytosolic ATP concentration (; Figure 2E,F). The baseline F/C-ratio (control condition; 10 mM glucose) was similar and stable between different nerves analyzed (Figure 2F), while after 2 min of MB+GD the F/C-ratio was reduced by 35.1 ± 1.7% (p<0.001; n = 19 nerves; Figure 2F). In further experiments, F/C-ratios were normalized between one (10 mM glucose) and zero (MB+GD; Figure 2G). As expected, during MB+GD the decrease in axonal ATP was mirrored by a decay of CAPs, reflecting conduction blocks (Figure 2H). Strikingly, also the decline rates of axonal ATP and conductivity were similar (Figure 2I).

6Next, we compared ATP and CAP dynamics in response to different modes of energy deprivation (Figure 3; Figure 3—figure supplement 1), including glucose deprivation (GD), blockage of mitochondrial respiration by azide (MB), and a combination of both (MB+GD). During GD both ATP and CAPs decayed after 13 min (Figure 3A). Inhibition of respiration resulted in a drop after about 3 min with a steeper decline (Figure 3B). MB+GD induced an immediate and fast decline of ATP, followed by an equally fast decline in CAPs (Figure 3C–E). To reduce permanent damage and to study recovery, MB and MB+GD were limited to 5 min. The observed change of the ATP sensor signal is most likely not caused by major changes in pH as YFP fluorescence remained almost unchanged under all conditions of energy deprivation (Figure 2D and Figure 3—figure supplement 2). These data suggest that optic nerves continue ATP production under GD, presumably by metabolizing glycogen of astrocytes. However, axonal ATP production strongly depends on mitochondrial respiration, and axonal glycolysis alone is insufficient to maintain ATP levels, even in the virtual absence of electrical activity (baseline recording conditions with 0.033 Hz stimulation frequency).

7When nerves were reperfused (aCSF, 10 mM glucose) after GD, CAPs recovered with a delay of about 3 min, and the recovery of CAPs preceded the visible recovery of ATP levels by several minutes (Figure 3A,F). Interestingly, following MB+GD, CAPs recovered significantly sooner, concomitantly with axonal ATP, and faster (Figure 3F,G). We presume that, under MB+GD, lactate/pyruvate was not depleted from the optic nerve and is readily available for ATP generation by mitochondria after cessation of MB+GD.

8Maintaining axonal conductivity ex vivo depends on energy-rich substrates and high- frequency stimulation of axons causes CAPs to decrease. However, it has never been possible to study the reverse, i.e. the impact of electrical activity and spiking frequency on axonal ATP levels. In ThyAT optic nerves incubated in aCSF containing 10 mM glucose and subjected to electrical stimulation at 16 Hz, 50 Hz or 100 Hz for 2.5 min, CAPs dropped in a stimulation frequency dependent manner (16 Hz: 94.9 ± 1.5%; 50 Hz: 79.6 ± 0.4%; 100 Hz: 66.4 ± 1.0%; n = 5 nerves; Figure 4A,C, Figure 4—figure supplement 1). At the same time, axonal ATP levels decreased as stimulation frequencies increased (16 Hz: 91.6 ± 1.0%; 50 Hz: 82.3 ± 1.4%; 100 Hz: 68.9 ± 2.8%; n = 5 nerves; Figure 4B,D) indicating higher axonal ATP consumption. Even maximal ATP changes and maximal loss of conductivity correlated over stimulation frequencies (Figure 4E,F). Similar results were obtained with a different stimulation paradigm with continuously increasing frequencies (Figure 4—figure supplement 2); however, the causal relations of the correlation of CAP and ATP changes are most likely more complex (see discussion below). Finally, both the rates of ATP drop and recovery correlated with the amplitude of ATP changes (Figure 4—figure supplement 3).

9Next, we asked whether lower (physiological) glucose concentrations have an impact on nerve conduction and axonal ATP levels when nerves were challenged with different spiking frequencies (Figure 4C,D; Figure 4—figure supplement 4). In presence of 10 mM and 3.3 mM glucose, CAP performance and ATP levels dropped similarly with increasing stimulation frequencies. However, when only 2 mM glucose was applied both CAP and ATP levels were substantially decreased (Figure 4C,D) and the ATP/CAP ratio was more variable at 2 mM glucose (Figure 4F). Analysis of the fluorescence intensity of single channels indicated that the observed changes of the ATP sensor signal are indeed caused by ATP changes because YFP fluorescence remained rather unaffected (Figure 4—figure supplement 5). Taken together, these data suggest that 2 mM glucose is below the threshold required for maintaining the axonal energy balance of the optic nerve in the ex vivo experimental setting.

10At high-frequency stimulation and under aerobic conditions, optic nerve CAPs were maintained equally well with either 10 mM glucose, 10 mM lactate or 10 mM pyruvate as extracellular energy substrate (Figure 5A), in agreement with earlier findings presumably because absolute axonal ATP consumption is unaffected by the type of metabolic support. However, at 100 Hz the relative decrease in axonal ATP levels was larger with lactate (or pyruvate) than with glucose (glucose: 68.9 ± 2.8%; lactate: 56.3 ± 2.8%; pyruvate: 47.4 ± 2.8%; p=0.05 and p=0.01, respectively; n = 3 nerves; Figure 5B, Figure 5—figure supplement 1), showing that 10 mM glucose maintains axonal ATP production better than 10 mM lactate or pyruvate if applied exogenously.

11Do axons require pyruvate/lactate metabolism also in the presence of glucose as energy substrate? To address this question, we studied optic nerves conduction and ATP levels in 3.3 mM glucose plus 20 mM D-lactate, which competitively inhibits transport through MCTs as well as lactate dehydrogenase (LDH; D-lactate is a stereoisomer of L-lactate, which is the physiological molecule in energy metabolism). D-lactate did not affect CAPs (Figure 5C, Figure 5—figure supplement 2) but had a strong effect on axonal ATP (at 100 Hz: Glc: 59.7 ± 2.9%; Glc+D-Lac: 29.2 ± 2.2%; p=0.0001; n = 5 nerves; Figure 5D). Similarly, when MCT1/MCT2-mediated lactate transport was inhibited using the specific inhibitor AR-C155858, relative ATP levels were strongly reduced (at 100 Hz: 24.8 ± 4.1%; p=0.0004; n = 5 nerves), but with no effect on CAP performance (at 100 Hz: Glc: 69.5 ± 4.2%; Glc+AR-C155858: 62.8 ± 3.3%; p=0.37; n = 6 nerves; Figure 5C,D; Figure 5—figure supplement 2), indicating the presence of a ‘safety window’. When lactate metabolism was impaired, ATP/CAP ratios strongly deviated from the normal ‘1:1 ratio’ observed in the presence of glucose alone (Figure 5E,F). Collectively, these results strengthen the conclusion that pyruvate/lactate metabolism is required to maintain ATP levels in fast spiking axons.

12The axonal energy balance is determined by the equilibrium of ATP consumption and ATP generation, which is a function of axonal spiking frequency and metabolic support, the latter provided in part by glycolytic oligodendrocytes. Here, we have generated transgenic mice expressing an ATP-sensor in neurons and in the axonal compartment of myelinated nerves. These mice allowed us to study for the first time the interrelationship of electrical activity and axonal ATP homeostasis in real time by adapting an established optic nerve model of white matter electrophysiology. Specifically, we show that both CAPs and axonal ATP content decrease in a stimulation frequency-dependent manner, and that even in the presence of glucose as (exogenous) energy substrate, lactate contributes to ATP homeostasis. We also analyzed short-term adaptations of metabolism in a model of ischemic stroke when ATP demands are increased, i.e. the recovery phase from transient glucose deprivation and mitochondrial inhibition.

13While being a powerful experimental system, some caveats need to be taken into account to interpret the obtained results. First, the signal of the ATeam1.03YEMK sensor is not linearly related to ATP over an extended range of concentrations. However, in cultured neurons the cytosolic ATP concentration has been estimated at approximately 2 mM, i.e. within the linear range of ATeam1.03YEMK. We are thus confident that the observed changes in sensor signal reflect similar relative changes of axonal ATP. However, as calibration of the ATP-sensor signal to absolute cytosolic concentrations is difficult in the myelinated optic nerve, a precise estimation of the absolute concentration of axonal ATP awaits further technical advances. Nevertheless, changes of the ATP signal in the course of an experiment could be detected even after mild stimulations, which suggest that the basal axonal ATP level is within the dynamic range of the sensor with its KD of 1.2 mM.

14Secondly, with the spatial resolution of our approach, ATP signals reflect mean cytosolic concentrations in the axoplasm. It has been suggested that local differences of ATP consumption by Na+/K+-ATPase lead to ATP microdomains close to the cell membrane with low ATP concentrations. Contrarily, ATP diffusion is reported to be very fast not allowing formation of ATP microdomains. While such highly localized changes of ATP would be missed in our experiments, they are unlikely to affect the principle findings.

15Finally, CAPs recorded with supra-threshold stimulation reflect the summation of action potential propagation of virtually all axons within the optic nerve. In contrast, ATP imaging was performed on a single optical plane close to the surface of the nerve comprising typically around 30 axons. Therefore, the observed ATP signals are likely not representative for all axons, as the diffusion of substrates into the core of the nerve may be less efficient than to axons closer to the surface. Our attempts to address this issue by 2-photon laser scanning microscopy deeper within the nerve have failed so far, due to scattering occurring in a densely packed and fully myelinated optic nerve (ASS, unpublished observation). Taking such an inside-out gradient of increasing perfusion efficacy into account, our approach is presently the best possible approximation of the simultaneous monitoring of both parameters. Thus, we are not yet able to assign absolute ATP concentrations to physiological properties of spiking axons. However, we can readily (i) monitor ATP changes as a function of axonal spiking activity, (ii) compare metabolites, (iii) assess the role of disease-causing mutations on axonal conduction and energy metabolism, also under physiological challenge, and finally (iv) study selected drugs and nutrients in axonal energy metabolism to explore treatment strategies.

16Targeting the energy producing biochemical pathways by the withdrawal of glucose (GD), blocking mitochondrial respiration (MB), or both (MB+GD), rapidly reduced axonal ATP content, as expected. In GD, ATP fell after 13 min, consistent with the utilization of astroglial glycogen as a reserve fuel. In contrast, MB caused an immediate drop in ATP. In all conditions, ATP decreases preceded or coincided with CAP decay, confirming that energy depletion causes axonal failure (but see caveat above). Interestingly, after GD in the subsequent recovery phase with glucose, the restoration of ATP levels appeared delayed compared to the recovery of conduction. We therefore suggest that at the beginning of reperfusion most newly generated ATP is consumed by Na+/K+-ATPases for reestablishing ion gradients and conduction before it accumulates in the axon and binds to the ATP-sensor.

17Astrocytes can mobilize glycogen and generate lactate to support axonal energy homeostasis. However, it has been a long-standing debate whether neuronal compartments prefer glucose or lactate as energy-rich substrates. In the optic nerve, glucose and pyruvate/lactate can sustain CAP propagation. We found that axonal ATP levels were significantly lower during HFS when nerves were incubated with lactate or pyruvate as the sole energy substrate. This suggests that lactate/pyruvate alone does not support ATP production as well as glucose and confirms observations in cultured cerebellar granule cells with repetitive activation of NMDA-receptors.

18Nevertheless, lactate metabolism contributes to axonal energy homeostasis, even in the presence of glucose, as evidenced by blocking lactate metabolism with D-lactate or AR-C155858. In the presence of these inhibitors, ATP decreased more strongly than CAP (Figure 5E,F), suggesting the presence of a ‘safety window’. It also breaks the striking correlation of ATP and CAP observed under control conditions (Figures 4E,F and 5E,F). One possible explanation is that changes in CAP are related to ATP consumption, while inhibition of lactate transport affects ATP production. Since the axonal stimulation protocol is unchanged, also the consumption of ATP should be similar in the presence of inhibitors of lactate metabolism. Therefore, the inhibition of lactate metabolism impairs ATP production and causes a lower steady-state level of ATP, but the consumption rates during HFS remain the same.

19Furthermore, this finding suggests that ‘endogenous’ lactate (i.e. produced from glucose) is more efficient than lactate provided in the medium. One possibility is that glucose enters the nerve more rapidly than lactate, because glucose transporter expression is optimized in comparison to MCT1. Moreover, glucose and glycolysis intermediates are readily shuttled via gap junctional coupling between astrocytes and oligodendrocytes, with lactate being generated also close to the periaxonal space, whereas exogenous lactate has a more complex path of diffusion involving mostly extracellular space. We note that from an evolutionary point of view, such optimization of glucose utilization is of advantage for the brain, because the liver always aims at generating a constant blood glucose level, even under starvation conditions.

20Taken together, combining ATP imaging and electrophysiology, we demonstrate that metabolic imaging is a very sensitive analysis tool for real-time monitoring of axonal energy homeostasis and the underlying neuron-glia interactions in electrically active fiber tracts. Applied to models of diseases this technology will be able to detect subtle perturbations of axonal energy homeostasis and allow addressing the hypothesis that energy deficits are an early and most likely causal event for neurodegeneration.

21Animals were treated in accordance with the German Protection of Animals Act (TSchG §4 Abs. 3), with the guidelines for the welfare of experimental animals issued by the European Communities Council Directive 2010/63/EU as well as the regulation of the institutional ‘Tierschutzkommission’ and the local authorities (T04/13, T20/16; Landesdirektion Leipzig, LAVES Niedersachsen).

22Animals were bred in the animal facility of the Max-Planck-Institute for Experimental Medicine as well as the Medical Faculty of the University of Leipzig. Mice were housed in a 12 hr/12 hr light dark cycle with access to food and water ad libitum. The transgene construct was assembled in pTSC containing Thy1.2 promoter sequences. The plasmid pDR-GW AT1.03YEMK containing the open reading frame of the ATP-sensor ATeam1.03YEMK was obtained from Wolf Frommer (via Addgene; plasmid #28004). The open reading frame of ATeam1.03YEMK was subcloned into the XhoI restriction site of pTSC using PCR. The linearized transgene was injected into fertilized mouse oocytes of the C57BL/6J mouse strain. Transgenic founders were identified using PCR-based genotyping on genomic DNA isolated from tail tips (primer 1: 5’-CGCTGAACTTGTGGCCGTTTACG-3’; primer 2: 5’-TCTGAGTGGCAAAGGACCTTAGG-3’). The mouse line has been registered at the RRID Portal (https://scicrunch.org/resources; RRID:MGI:5882597).

23Mouse genomic DNA was isolated from tail biopsies of four C57BL/6J and six ThyAT-mice with Invisorb Spin Tissue Mini Kit according to the manufacturer’s instructions (Stratec Biomedical, Birkenfeld, Germany). Short FAM-labeled hydrolysis probes (UPL) were used for qPCR reactions (Roche Diagnostics GmbH, Mannheim, Germany). Primers and the UPL-probe were designed by ProbeFinder version 2.50 for Mouse (Roche Diagnostics; Primer Thy1s: TGCCGGTGTGTTGAGCTA; Thy1as: TGGTCCTGTGTTCATTGCTG; UPL 60; amplicon 73 bp). The genomic sequence of Nrg1 was used to calibrate for the amount of DNA (Primer Nrg1s: GGCTATAATGCTAACACAGTCCAA; Nrg1as: AGTGGATCGTAACAACACTGTCA; UPL 38; amplicon 61 bp) as described. 10 to 25 ng of genomic DNA was subjected to qPCR amplification to measure the amount of Thy1.2 on a Light Cycler 480 system (Roche Diagnostics) according to the manufacturer’s instructions. The copy number of the Thy1.2-ATeam1.03YEMK transgene was calculated using the ΔΔCt method compared to wild type mice carrying only the two endogenous alleles of the Thy1 gene.

24Adult mice were transcardially perfused with 4% formaldehyde solution (PFA, in phosphate buffered saline: 137 mM NaCl, 2.7 mM KCl, 8 mM Na2HPO4, 1.5 mM KH2PO4, pH 7.4) under deep anesthesia. The brain and the eyes were removed and post-fixed for 24 hr in the same fixative. 45 μm thick sections were cut on a vibratome (Leica VT1000 S, Nussloch, Germany) and slices were mounted directly after cutting with Vectashield embedding medium (Vectashield HardSet Mounting Medium, Vector Laboratories, Burlingame, CA, USA). From the eyes, retinal whole mounts were prepared.

25For imaging of fixed brain slices and retina, confocal images were acquired on a LSM Olympus IX71 inverted microscope using an UPlanFL 10x/0.3 objective (Olympus, Hamburg, Germany) for overview images (Figure 1A–C,G) or an UApo/340 40x/1.35 oil objective (Olympus) for detailed images (Figure 1D–F,H), respectively. Microscopic images were acquired and processed using Olympus Software Fluoview v5.0. ATeam1.03YEMK sensor fluorescence was excited with a 488 nm argon laser and detected through a BA 510–540 nm emission filter (AHF Analysentechnik AG, Tübingen, Germany). For all images, a Kalman filter of two was used for denoising and images were acquired with 1024 × 1024 resolution (pixel size for overview images: 0.9 µm; pixel size for detail images: 0.35 µm). Z-stacks comprise 12–38 singles z-planes for overview images and 30–85 z-planes for detailed images, respectively, with distances between each z-plane of 2 µm (overview), 0.5 µm (Figure 1D–F) and 0.35 µm (Figure 1H). Single z-stacks were converted to maximum intensity projections (by using Fiji macro ‘Flattening V2f.ijm’; generously provided by Jens Eilers; Jens-Karl.Eilers@medizin.uni-leipzig.de) and for overview images maximum intensity projections of different positions (210 for Figure 1A; 154 for Figure 1B; 160 for Figure 1C; 49 for Figure 1G) were stitched by using Fiji software and a Fiji stitching plugin.

26For optic nerve experiments, mice were used in an age of 8 to 12 weeks. Optic nerves were excised from decapitated mice, placed into an interface perfusion chamber (Harvard Apparatus, Holliston, MA) and continuously superfused with artificial cerebrospinal fluid (aCSF). The perfusion chamber was continuously aerated by a humidified gas mixture of 95% O2/5% CO2 and experiments were performed at 37°C. Custom-made suction electrodes back-filled with aCSF were used for stimulation and recording as described. The stimulating electrode, connected to a battery (Stimulus Isolator 385; WPI, Berlin, Germany) delivered a supramaximal stimulus of 0.75 mA to the nerve evoking compound action potentials (CAP). The recording electrode was connected to an EPC9 amplifier (Heka Elektronik, Lambrecht/Pfalz, Germany). The signal was amplified 500 times, filtered at 30 kHz, and acquired at 20 kHz or 100 kHz. Before recording, optic nerves were equilibrated for at least 30 min in the chamber.

27The CAP, elicited by the maximum stimulation of 0.75 mA, was recorded at baseline stimulation frequency at 0.1 Hz. During HFS a burst-stimulation was applied consisting of 100 stimuli at the given frequency (16, 50 or 100 Hz), separated by 460 ms, during which the CAP was recorded; HFS overall duration was 150 s, independent of the frequency.

28Live imaging of optic nerves was performed using an up-right confocal laser scanning microscope (Zeiss LSM 510 META/NLO, Zeiss, Oberkochen, Germany) equipped with an Argon laser and a 63x objective (Zeiss 63x IR-Achroplan 0.9 W). The objective was immersed into the aCSF superfusing the optic nerve. Theoretical optical sections of 1.7 μm over a total scanned area of 66.7 μm x 66.7 μm (512 × 512 pixels) of the optic nerve were obtained every 10.4 s in three channels, referred as CFP (excitation 458 nm; emission 470–500 nm), FRET (Ex 458 nm; Em long pass 530 nm) and YFP (Ex 514 nm; Em long pass 530 nm).

29Optic nerves were superfused by aCSF containing (in mM): 124 NaCl, 3 KCl, 2 CaCl2, 2 MgSO4, 1.25 NaH2PO4, and 23 NaHCO3, which was continuously bubbled with carbogen (95% O2/5% CO2). This solution was substituted with the appropriate energy substrates as indicated for which the solution containing 10 mM glucose served as the standard solution in respect to pH and osmolarity. For glucose deprivation (GD), glucose was removed from the aCSF and substitute by sucrose (Merck Millipore, Darmstadt, Germany) to maintain the correct osmolarity. For mitochondrial blockage (MB), aCSF containing 10 mM glucose was supplemented with 5 mM sodium azide (Merck Millipore). For the combination of MB+GD glucose was substituted by sucrose and 5 mM sodium azide was added.

30For HFS, aCSF was supplemented with different substrates. Glucose (Fluka BioChemika, Munich Germany) was used at three concentrations: 10 mM, 3.3 mM and 2 mM. L-lactate and pyruvate (Sigma-Adrich, Munich, Germany) were used at 10 mM. Inhibitors of lactate metabolism were added to a basal concentration of glucose of 3.3 mM: the MCT1 inhibitor AR-C155858 (Med Chem Express, Sollentuna, Sweden) was used at 10 μM; sodium D-lactate (Sigma-Aldrich) was used at 20 mM. All aCSF-based solutions were adjusted for the same pH, sodium concentration and osmolarity.

31Optic nerve function was monitored quantitatively as the area under the supramaximal CAP. The CAP area is proportional to the total number of excited axons and represents a convenient and reliable means of monitoring optic nerve axon function. CAP was expressed as area under the CAP wave form in a range of approximately 1.5 ms, defined by the beginning of the response (typically at 0.2 ms after the stimulus) and approximately the end of the second peak of the CAP wave form, thereby analyzing the first and second peak characteristic of the optic nerve evoked CAP. These two peaks are indicative for the large and medium-sized axons within the optic nerve, which are those that will mainly be resolved by confocal imaging of the ATP sensor expressed in axons. During high frequency action potential propagation, amplitudes of CAP peaks generally decrease, while peak latencies increase. The measured CAP area integrates these two functional changes of nerve conduction. It is, therefore, a parameter well suited to quantify overall functional changes of the optic nerve. Data were normalized to the mean obtained during the initial 15 min (defined as baseline, i.e. no experimental condition applied). Results from several nerves were pooled, averaged, and plotted against time.

32The relative amount of ATP in the optic nerve was calculated in two steps. Initially, the mean intensity of FRET and CFP channel were extracted from the imaging time-series (in Fiji) and the ratio F/C was calculated. All fibers visible in the field of view were included within the analysis without any prior selection. The F/C ratio was then normalized to zero using the F/C value obtained by application of MB+GD at the end of each experiment, i.e. by depleting all axons from ATP. Furthermore, F/C values obtained during baseline recordings in 10 mM glucose at the beginning of each experiment were set to one. In all experiments, YFP fluorescence upon direct excitation of YFP was analyzed in parallel. Importantly, YFP fluorescence intensity remained stable throughout the experiments suggesting that the observed changes of the F/C ratio are most likely not due to artificial modulation e.g. by changes in intra-axonal pH.

33For GD, MB and MB+GD conditions, m coefficients for the slope-intercepts at the point of maximal change of the signal were used to calculate the CAP and ATP rate per second, during the initial decay and the recovery following reperfusion with aCSF containing 10 mM glucose. The time of the start of decay/recovery of the ATP and CAP curves were defined as the first time-point at which the slope-intercept of the signal was above a threshold value based on the standard deviation calculated at the baseline. For HFS analysis three parameters were considered for both CAP and ATP: (1) the overall amplitude obtained by averaging the data at the time points within the last 15 s of the stimulation (2) the rate of initial decay defined within the 45 s that followed the beginning of the stimulation (3) the rate of recovery defined within the 60 s that followed the end of the stimulation. Rates were calculated from the m coefficients for the slope-intercepts in the defined time-range.

34All data are presented either as mean ± s.e.m. or as stripe plots showing all data points and the mean (line within the plot). The number of optic nerves analyzed for each condition is given as n. As for no condition both optic nerves of one animal were used, the number of nerves is equal to the number of animals analyzed for each condition. If not indicated otherwise in the figure legends, data were statistically evaluated using Welch’s t-test and assuming a normal distribution (*p<0.05; **p<0.01; ***p<0.001).