Oxidative pentose phosphate pathway and glucose anaplerosis support maintenance of mitochondrial NADPH pool under mitochondrial oxidative stress

Abstract Mitochondrial NADPH protects cells against mitochondrial oxidative stress by serving as an electron donor to antioxidant defense systems. However, due to technical challenges, it still remains unknown as to the pool size of mitochondrial NADPH, its dynamics, and NADPH/NADP+ ratio. Here, we have systemically modulated production rates of H2O2 in mitochondria and assessed mitochondrial NADPH metabolism using iNap sensors, 13C glucose isotopic tracers, and a mathematical model. Using sensors, we observed decreases in mitochondrial NADPH caused by excessive generation of mitochondrial H2O2, whereas the cytosolic NADPH was maintained upon perturbation. We further quantified the extent of mitochondrial NADPH/NADP+ based on the mathematical analysis. Utilizing 13C glucose isotopic tracers, we found increased activity in the pentose phosphate pathway (PPP) accompanied small decreases in the mitochondrial NADPH pool, whereas larger decreases induced both PPP activity and glucose anaplerosis. Thus, our integrative and quantitative approach provides insight into mitochondrial NADPH metabolism during mitochondrial oxidative stress.

tion rates of H 2 O 2 in mitochondria and assessed mitochondrial NADPH metabolism using iNap sensors, 13 C glucose isotopic tracers, and a mathematical model. Using sensors, we observed decreases in mitochondrial NADPH caused by excessive generation of mitochondrial H 2 O 2 , whereas the cytosolic NADPH was maintained upon perturbation. We further quantified the extent of mitochondrial NADPH/NADP + based on the mathematical analysis. Utilizing 13 C glucose isotopic tracers, we found increased activity in the pentose phosphate pathway (PPP) accompanied small decreases in the mitochondrial NADPH pool, whereas larger decreases induced both PPP activity and glucose anaplerosis. Thus, our integrative and quantitative approach provides insight into mitochondrial NADPH metabolism during mitochondrial oxidative stress. acts as a signaling molecule that can initiate expression of survivor genes such as antioxidant response elements (e.g., Nrf2), induce DNA repair mechanisms (e.g., p53 and ATM), or activate programmed cell death pathways (e.g., NF-ĸB). 4,8,9 As accumulation of H 2 O 2 can induce toxicity, cells maintain a defensive system to clear H 2 O 2 via redox reactions. 2,10,11 In the antioxidant network, NADPH plays a critical role by serving as a reductant during removal of H 2 O 2 to maintain redox homeostasis.
It donates two electrons to reduce oxidized cysteine residues of thioredoxin via thioredoxin reductase, or glutathione via glutathione reductase. [12][13][14][15] Thioredoxin with reduced cysteine residues reacts with peroxiredoxins, which have been known to be the major scavenger of H 2 O 2 based on their abundance and fast second order rate coefficient compared to glutathione peroxidase reaction at low levels of intracellular H 2 O 2 . 16,17 In mitochondria, peroxiredoxin 3 is known to scavenge 90% of H 2 O 2 , suggesting peroxiredoxin-thioredoxin-NADPH as the major clearance pathway for mitochondrial H 2 O 2 . 16 In parallel, glutathione reacts with glutareredoxin which serves as a reductase for oxidized proteins, and with glutathione peroxidase during direct reactions with H 2 O 2 . 18,19 Oxidative stress, a condition caused by an inadequate clearance or excessive production of H 2 O 2 , has been reported to decrease the total NADPH pool, NADPH/NADP + ratio, and modulate NADPdependent metabolic fluxes. 12,13,20 For instance, oxidative stress in fibroblast cells led to a shift of glycolytic flux toward the oxidative pentose phosphate pathway to regenerate NADPH. [20][21][22][23] In isolated cardiac myocytes under a pathological workload, the direction of mitochondrial nicotinamide transhydrogenase reaction was reversed, lowering the total NADPH pool and increasing the production of mitochondrial ROS. 24 Similarly, the availability of NADPH in mitochondria along with NADPH-producing enzymes such as isocitrate dehydrogenase 2 (IDH2) and nicotinamide nucleotide transhydrogenase (NNT) was shown to control the antioxidant network for clearance of H 2 O 2 . 25 To our knowledge, little has been known about the causal relationship between oxidative stress derived within mitochondria and mitochondrial or cytosolic NADPH pools. Previously, the total NADPH level or its ratio to NADP + has been reported to decrease by exogenous oxidative stress introduced extracellularly or to the exterior of isolated mitochondria, but no direct evidence of compartmentalized NADPH dynamics by mitochondria specific oxidative stress in living cells. Gas or liquid chromatography coupled to mass spectrometry and enzymatic cycling assays provide great sensitivity and specificity for measurement of NADPH, but they are based on measurements of the average of cell lysates, making it difficult to preserve spatial and temporal information of NADPH in living cells at single cell resolution. 26,27 Co-expressing compartment-specific NADPH sensors and D-amino acid oxidase (DAAO) which is used as H 2 O 2 generator, we evaluated mitochondrial and cytosolic NADPH dynamics upon localized H 2 O 2 stress first time. iNap sensors were geneticallyencoded probes for NADPH and provided a wide dynamic range with a ratiometric fluorescent readout, which could be simply recorded using a fluorescence microscope at single-cell resolution. 28 For mitochondrial NADPH experiments, we expressed the sensors using mitochondria using localization tags. Fluorescence ratio was defined as a ratio between fluorescence emissions at 515 nm excited at 415 and 488 nm (R = Fem:515,ex:415 Fem:515 nm,ex:488 ). Due to the influence of pH fluctuations to iNap fluorescence at 488 nm, the fluorescence ratio of iNap could be normalized to that of iNapC, a control iNap sensor engineered to lose its binding affinity to NADPH. As cytoplasmic pH was shown to be stable during oxidative stress, 28  is reduced to water by reacting with reactive cysteine residues of Prx, GPx, or proteins. Through subsequent redox cycling reactions, NADPH acts as an ultimate reductant by donating electrons to oxidized redox species. The mitochondrial NADPH pool was monitored by measuring a fluorescence ratio from a mitochondrial iNap sensor, a genetically encoded sensor for NADPH. Detailed mitochondrial redox reactions considered in our system can be found in Table S1 F I G U R E 2 Validation of a system that generates hydrogen peroxide in mitochondria with D-amino acid oxidase (DAAO) and measures NADPH with an iNap sensor. (a) Hela cells were transiently transfected with a mito-DAAO-FLAG and its localization to mitochondria was confirmed. Staining: Mitotracker (red), anti-FLAG (green), DAPI (blue). (b) The enzymatic activity of DAAO was measured via a horseradish peroxidase based Amplex UltraRed assay. Fluorescence intensity was measured after incubation of HeLa cell lysates with D-alanine for an hour. Fluorescence readings were converted to hydrogen peroxide concentrations using a standard curve constructed using known concentrations of hydrogen peroxide. Data represent two independent experiments ±SD. (c) Cell numbers were counted after 24 hr of incubation of D-alanine with Hela/mito-DAAO cells with error bars representing SEM of three independent experiments. (d) Pseudo-colored images represent the change of fluorescence intensity of Hela/mito-iNap cells in the presence or absence of NADPH, or diamide. For incubation of NADPH, 0.05 mg/ml digitonin was used to permeabilize the mitochondrial membrane. (e) Fluorescence readout (R 0 = RiNap−mito R iNapC − mito Þ was quantified before and after addition of 400 μM NADPH in digitonin treated Hela cells expressing iNap sensors. Data represents the mean of fluorescence ratio of individual cells from three independent experiments ±SEM. (n = 22 and 16 cells) F I G U R E 3 Generation of mitochondrial H 2 O 2 via mito-DAAO and measurement of compartmentalized NADPH via iNap-mito or iNap-cyto. (a) Schematics depicting a system that generates mitochondrial H 2 O 2 via expression of mito-DAAO and measures the mitochondrial NADPH pool using the iNap-mito sensor. (b) Single cell images representing a fluorescence ratio of mitochondrial iNap sensors in response to D-alanine treatment. Raw images were exported to MATLAB, processed to remove background signal, and the ratio of two images obtained from the 415 and 488 nm excitation channels was calculated. Individual pixel values were pseudocolored in range of 0 to 15, representing a dark-blue (low) to red (high), whose values were used only for graphical visualization of cells. (c) Fluorescence readout (R 0 ), normalized to that of initial value, was measured every 3 min after stimulation with D-alanine ranging from 0 to 50 mM. Normalized R 0 represent mean values of individual cells from at least two independent experiments ±SEM.   Figure S8). 29,40 The findings shown in Figure 2c are consistent with previous results that a high dose of D-alanine such as 25 mM killed cells through an apoptotic pathway. 29 Next, we tested the functionality of the mitochondrial iNap sensor before implementation of experiments with DAAO system. First, we introduced an artificial oxidative stress by stimulating cells with 500 μM diamide, which was previously shown to minimally influence the fluorescence of iNap control sensor. 28 We recorded an excitation spectrum with the emission wavelength centered at 515 nm, confirming a decrease of the ratio of 515 nm emission upon excitation with light centered at 415 and 488 nm as previously described (Figures 2d and S1E). 28 Afterwards, we obtained the maximum fluorescence ratio of the iNap-mito sensor. We stimulated Hela/iNapmito cells with 400 μM of NADPH with 0.05 mg/ml digitonin, and measured the change of fluorescence ratio every 20 s ( Figure S1F).
The effective fluorescence ratio was achieved by normalizing the fluorescence ratio of the iNap-mito sensor to that of iNapC-mito, which was designed to function as a control sensor that responds to pH ( Figure 2e). 28   We previously demonstrated that an excessive generation of H 2 O 2 via mito-DAAO system could increase oxidation states of peroxiredoxin as well as glutathionylation of proteins under high concentration of D-alanine, and suggested a threshold concentration of D-alanine that triggered cellular toxicity to be between 15 and 25 mM for short perturbation times in Hela-DAAO system. 29,41 Similarly, our data demonstrated that 25 mM D-alanine was the threshold concentration that allowed a significant decrease of NADPH pools in mitochondria. Additionally, we assessed whether the absence of carbon source such as glucose or glutamine could influence the mitochondrial NADPH pool during high production of mitochondrial H 2 O 2 .
In the absence of glucose, the normalized R 0 from iNap-mito decreased by nearly 50% compared with the samples that were treated with 25 mM D-alanine with glucose ( Figure S3A). In the presence of glutamine, the fluorescence readout was not statistically different, suggesting glucose metabolism as the primary source for maintenance of mitochondrial NADPH pool in 30 min.
Next, we explored whether the generation of mitochondria H 2 O 2 influenced the cytosolic NADPH pool. We generated mitochondrial   Figure S3C-E).

| Production of cytosolic H 2 O 2 decreases cytosolic NADPH first followed by mitochondrial NADPH pool
As the fluorescence ratio of the cytosolic sensor was maintained under mitochondrial production of H 2 O 2 , we investigated whether    Table S1). Reasoning that NADPH could be transferred between cytosol and mitochondria via indirect metabolite shuttle systems above a certain threshold hydrogen peroxide generation rate in mitochondria, 47,48 we introduced the stress-dependent NADPH flux, which was defined as a α × v total H2O2 . As the mitochondrial NADPH pool is decreased by high production rates of mitochondrial H 2 O 2 flux, we expect an additional NADPH flux introduced in mitochondria to maintain the mitochondrial NADPH level.
We used a weighted least-square minimization method based on , which is derived from the binding kinetic equation between the sensor and NADPH. The initial concentration of redox species involved in this system were calculated as previously described (Table S2). 33 The model data, including Y model , was obtained by solving the system of ordinary differential equations with 1,000 different sets of randomly-chosen initial parameter values along with the basal NADPH concentration obtained from experiment and initial concentration of redox species found in literature (Table S2, The stress-dependent NADPH flux coefficient (α) was found to be 80.9. In the absence of α, the model prediction failed to predict the experimental data under higher perturbation ( Figure S7D (Table S1).
Furthermore, the mitochondrial NADPH/NADP + ratio decreased by 67-fold when the generation of mitochondrial H 2 O 2 increased by 19 times higher compared with basal rate (Figure 6e). The NADPH/ NADP + ratio plays critical roles as multiple NADP dependent enzymes are reversible and a subtle change of ratio can alter the directionality of reactions, thereby switching the cellular metabolism. 12,50 Our model estimated the steady-state mitochondrial NADPH/ NADP + ratio at basal condition (0 mM D-alanine) to be 13.4 (Table 1). This value is approximately 100-fold lower than the whole NADPH/NADP ratio of live cells examined from classic literature, which reports that the ratio can reach as high as 1,000 under starved condition using a near equilibrium approximation. 26 Under different initial  Several genome-scale flux balance analysis (FBA) models have been recently adopted to predict NADPH flux at steady-state and assessed metabolic reactions that contributed NADPH pools with constraints obtained from experimental results such as metabolite intake and uptake rates or proteome bioinformatics data. 23,37 Although these models demonstrated feasibility of identifying metabolic reactions that contributed most to NADPH pool such as folate cycle pathway, PP pathway or IDH reaction, the model was limited by  Images were captured every 20 s, 1 or 3 min and exported to either ImageJ or MATLAB 2016a for post image processing.

| Image analysis of iNap sensors
Backgrounds of short (415 nm) and long (488 nm) wavelength images were subtracted using a rolling ball algorithm from ImageJ.
Long-wavelength images were converted to 32 bit and a threshold of one was applied to minimize artifact. The pixels values of the 415 nm filters were divided those of the 488 nm. Individual cells, neither too bright not too dim, were randomly selected and the mean fluorescence intensity of region of interest was calculated.
All the fluorescence emission ratios were recorded as the mean ± SEM. The images were created using the image processing algorithm in MATLAB 2016a with pseudo colors.

| Metabolite extractions and GC/MS analysis
Extraction and analysis methods were followed as previously described. 67  In regards to the regeneration rate for NADPH in mitochondria, we set a first order kinetic equation that represents major enzymatic reactions for NADPH production such as IDH2, ME3, NAD + transhydrogenases (NNT), and methylene tetrahydrofolate dehydeogenase 2 (MTHFD2). 68,69 We included the transport and degradation rates of glutathione, removing the glutathione synthesis rate as it was exclusively formed in the cytoplasm. 70 The thioredoxin influx and degradation rates were removed as their rates were indicated to be three orders magnitude lower than other redox reactions, thereby its sensitivity to the system low. 32 The generation rate of mitochon- The initial concentration of oxidized redox species in mitochondria were calculated based on the steady state approximation with the molar balance equations as described before. 33 Unless noted, we assumed rate constants of redox reactions in mitochondria are within the same order of magnitude of those in cytosol and thus used accordingly as listed in Table 1. For the initial concentration of mitochondrial NADPH concentration, we converted the fluorescence ratio of the sensor to NADPH concentration based on the digitonin based calibration experiments ( Figure S1F) as described previously. 28 The average concentration of NADPH in mitochondria was determined to be 41.8 μM based on the fluorescence images of 243 single cells ( Figure S7A).

| Quantification of NADPH
We quantified the mitochondrial NADPH level by calculating the fraction of sensor readout (Y exp ) and equating it to the fraction of sensor-NADPH complex (Y model ), which are expressed as follows: 0 is an effective fluorescence signal obtained by R 415nm=488nm j iNap − mito R 415nm=488nm j iNapC − mito . 28 R 0 max was obtained by permeabilizing Hela cells expressing mito-iNaps with an optimized concentration of 0.05 mg/ml of digitonin and incubating with 400 μM NADPH ( Figure S1F). To minimize artifact effects such as leakage of sensor, we have used the control NADPH sensor with no binding affinity to NADPH in parallel and normalized the sensor readout to that of control sensor. 28 To control the permeabilization of mitochondrial NADPH, we varied the concentration of digitonin from 0 to 1 mg/ml in presence or absence of NADPH in context of our experiments ( Figure S1F). As NADPH sensor was localized to mitochondria, NADPH sensor fluorescence signal did not change unless we introduced a threshold digitonin concentration. We observed the rise of signal was dependent on time-scale as lower concentration (0.05 mg/ml) allowed increase of signal at later time-points while higher concentration (0.1 mg/ml) allowed permeabilization effect in earlier time-point such that we observed a sharp increase of signal followed by steep decrease of signal potentially due to the leakage of sensors.
R 0 min was determined by addition of 100 mM D-alanine to Hela cells expressing DAAO and mito-iNaps. Due to the presence of antioxidant network present in cells, the R 0 min could be underestimated. Thus, we compared the quantified NADPH concentration to that of reference 28 , where the estimation of intracellular NADPH level from different variants of iNap sensors was consistent with in vtiro evaluation, and our estimates of free NADPH value fell within 5% of that determined in the reference (S7B). For Y model , NADPH represents concentration of mitochondrial NADPH and K d is the dissociation constant of the iNap3 sensor, which is 25.2 μM taken from the literature. 28 Y model is derived based on the equation of binding interaction between the ligand and the sensor with one to one stoichiometry.

| Objective function for parameter evaluation
With time-course measurement of NADPH level in terms of concentration, we fitted model parameters by minimizing an objective function that calculates the sum of squared difference between predicted and observed values as follows: where y obs is the experimentally observed NADPH concentration, w i (t k ) is the 1/σ i (t k ) 2 , σ is the SEM, N k is the number of data points taken for duration of 60 min with time interval of 3 min, and N exp is the number of experiments with five different conditions.

| Sensitivity analysis
Using the finite approximation methods, we implemented a sensitivity analysis to the NADPH level by every parameters in the reaction model. 75 The equation of sensitivity analysis is as follows: where s i represents the sensitivity to the θ i model parameters and C NADPH is the concentration of NADPH. Parameters were varied by 10% and the time was evaluated at 3 min. As the parameter values vary by orders of magnitude, we normalized the sensitivity to C NADPH (t) and θ i . The final sensitivity equation is as follows: All the normalized sensitivities were evaluated at corresponding D-alanine perturbations and the top 5 most sensitizing parameters were reported (Table S5).

| Quantification and statistical analysis
All results are represented as mean ± SEM of at least three biological replicates, unless indicated.