Morpho‐metabotyping the oxidative stress response

Oxidative stress and reactive oxygen species (ROS) are central to many physiological and pathophysiological processes. However, due to multiple technical challenges, it is hard to capture a comprehensive readout of the cell, involving both biochemical and functional status. We addressed this problem by developing a fully parallelized workflow for metabolomics (providing absolute quantities for > 100 metabolites including TCA cycle, pentose phosphate pathway, purine metabolism, glutathione metabolism, cysteine and methionine metabolism, glycolysis and gluconeogenesis) and live cell imaging microscopy. The correlative imaging strategy was applied to study morphological and metabolic adaptation of cancer cells upon short-term hydrogen peroxide (H2O2) exposure in vitro. The combination provided rich metabolic information at the endpoint of exposure together with imaging of mitochondrial effects. As a response, superoxide concentrations were elevated with a strong mitochondrial localization, and multi-parametric image analysis revealed a shift towards fragmentation. In line with this, metabolism reflected both the impaired mitochondrial function and shifts to support the first-line cellular defense and compensate for energy loss. The presented workflow combining high-end technologies demonstrates the applicability for the study of short-term oxidative stress, but it can be suitable for the in-depth study of various short-term oxidative and other cellular stress-related phenomena.


Results
The correlative metabolomics/imaging method comprised parallel in vitro experiments followed by tailored sample preparation for metabolomics and staining for microscopy. To achieve the highest degree of parallelization, cells were seeded in the same surface density and ratio of cell number to total medium volume, utilizing live cell imaging compatible multiwell plates and media.
In a proof-of-principle experiment, HCT116 colon cancer cells were exposed to hydrogen peroxide for a short period (2h) in live cell imaging medium and this perturbation could be investigated on both morphological and molecular levels. Microscopic analysis ensured the responsivity of our cell type to the stimulation protocol, albeit maintaining cell integrity and spatial resolution.
A concentration-dependent increase of the mitochondrial superoxide (expressed as MitoSox/MitoTracker signal ratio, Fig. 1) was found. Since structural integrity is tightly related to functional status [31][32][33] , a multiparametric image analysis of the mitochondrial network was also implemented according to the protocol of Valente and colleagues 34 . Intriguingly, we observed a rather consistent effect of H 2 O 2 stimulation on the average mitochondrial length and to a certain extent also on the mitochondrial network (ramification and junctions, Fig. 2). These responses seemed more stable at a concentration of 500 µM than at 1000 µM. Indeed, incubation with the highest concentration of H 2 O 2 was also accompanied by cell morphological changes, possibly accounting for the loss of specificity for some structural parameters at the onset of toxicity and increased mitochondrial density and www.nature.com/scientificreports/ clustering in the peri-nuclear region 31 . To support this interpretation, a significant decrease in the average and maximal branch length of the mitochondrial network could be measured after incubation with 1000 µM H 2 O 2 . The parallel metabolomics experiments were performed under an exposure to 500 µM H 2 O 2 , as a molecular readout for the endpoint of exposure, since the interpretation would be challenged by the observed loss of morphologic specificity at higher concentrations. Sample preparation for metabolomics comprised internal standardization by fully 13 C-enriched yeast and a derivatization step targeting all thiol-containing metabolites. In our experimental model, metabolome adaptation was substantial despite the relatively short stimulation protocol (2 h). Absolute concentrations were determined for 100 + metabolites covering multiple relevant pathways of the primary metabolome. Among others, energy metabolism, DNA synthesis, TCA cycle, glycolysis, pentose phosphate pathway, amino acid metabolism, and glutathione synthesis were covered in-depth with this method. Absolute concentrations were obtained using fully labelled internal standards in combination with external calibration. As a novelty, the applied HILIC-HRMS covered metabolites as published elsewhere 35 together with metabolites along the sulfur pathway, which are key factors for the redox balance of a cell. Primary thiols (glutathione, cysteine, homocysteine and glutamyl-cysteine) were assessed as N-ethylmaleimide (NEM) derivatives to prevent oxidation during sample preparation and storage and to prolong stability 36 . Internal standardization and normalization to the protein content resulted in excellent analytical figures of merit and enabled absolute quantification of > 100 metabolites. Moreover, the presence of fully labeled 13 C biomass monitored unwanted oxidation artifacts during sample preparation 37 . This material, beyond enabling absolute quantification on LC-MS platforms and compensating for losses during sample preparation, was shown to form GSSG with typical isotopologue patterns ( 13 C: 12 C 50:50) as a result of oxidation of fully labelled 13 C GSH and unlabeled GSH. The final derivatization protocol was quantitative, as free thiols remained < LOD in all samples, and it prevented measuring biased GSH/GSSG ratios, as no mixed isotopologues were measured. Consequently, all observed glutathione oxidation could be attributed to the hydrogen peroxide treatment and was not related to abiotic artefacts from www.nature.com/scientificreports/ the sample preparation and handling. While oxidized glutathione (GSSG) was below LOD (< 10 pmol) in all eight control samples, the hydrogen peroxide treated samples contained approximately 1 nmol GSSG each. Eight biological replicates were investigated following exposure to 500 µM hydrogen peroxide for 2 h. The data were evaluated in a targeted manner. Exploratory statistical analysis of the absolute amounts of 100 unique compounds using principal component analysis and hierarchical clustering clearly distinguishes within the sample groups (Fig. 3). The first principal component (39.2% explained variance) separates according to treatment, while in the heatmap all the samples are assigned to their respective cluster. Despite the short incubation time, hydrogen peroxide treatment led to strong changes, as a t-test with an adjusted p-value cutoff below 0.05 resulted in 51 unique significantly altered metabolites (Table 1), from which 30 increased and 21 decreased in their absolute concentrations, including several amino acids, nucleotides, organic acids and carbohydrates or related compounds.
In order to further investigate these changes, we constructed a metabolic network with MetExplore 38,39 , mapped the significantly changed metabolites on it and extracted the subnetwork based on the mapping to create a concise, global view of the underlying phenotype upon the short redox stress induced by the hydrogen peroxide treatment. The resulting metabolic network (Fig. 4) displayed strong and consistent changes throughout the citric acid cycle (TCA), glycolysis/gluconeogenesis, glutathione metabolism, cysteine and methionine metabolism, nucleotide synthesis (purine metabolism) and pentose phosphate pathway. Here we show that many of the changes are biochemically close to each other, and only a few possible enzymatic reactions apart. Indeed, the distinct perturbed pathways exhibit a globally coordinated metabolic effort to maintain redox and energy homeostasis and restore cellular damage, as we will discuss below.

Discussion
Currently, correlative multimodal imaging -omics methods are emerging [40][41][42][43] . These include scientific questions for which untargeted analysis can take advantage of high spatial resolution, or when fast kinetic responses are involved, which can be described uniquely by live cell imaging 44,45 In this work, correlative imaging was established, integrating mass spectrometry-based metabolomics as an endpoint in live cell imaging experiments on cancer cells. Live cell imaging enables superb spatial resolution and even more importantly, to follow responses under live physiological conditions, which perfectly match the functional readout provided by metabolome analysis. The workflow was applied to study the impact of ROS on metabolism, accompanied by changes in cellular morphology and mitochondrial networks. Oxidative stress, the disturbance of oxidant-antioxidant balance favoring oxidizing environment is largely mediated by a few representative radical molecules. Mitochondria are the main ROS-active cellular sites, mainly due to electron leakage from complexes I or III, resulting in superoxide production, using a large fraction of cellular antioxidant capacity 46 . Under physiological conditions, low levels of ROS are used for cellular signaling and for regulating mitochondrial homeostasis and cellular respiration. We used hallmarks and established notions about oxidative stress response 3,6 upon hydrogen peroxide exposure with the aim of proving recurring patterns and phenomena in our data generated by the combined imaging -omics method. Perturbation of cancer cells by hydrogen peroxide was selected as a poster case. Although the applied hydrogen peroxide is moderately reactive, it can diffuse into cells or mitochondria through membranes via aquaporins with peroxiporin activity 47 and generate the highly reactive and toxic hydroxyl radical and superoxide, thereby leading to pathological effects and mutation accumulation.
Pioneering reports on the role of mitochondria in oxidative stress response date back to the last century. Mitochondria are known to handle the extramitochondrial oxidants to a large extent 48, thereby tuning the mitochondrial metabolism 49 , in particular glycolysis and oxidative phosphorylation 50 . Furthermore, DNA repair mechanisms 51 and apoptosis are induced by exogenous 52,53 and endogenous 54 hydrogen peroxide and the interplay of calcium fluxes. Oxidative stress and different forms of cellular death [55][56][57] have been extensively studied and summarized in exquisite reviews 58,59 . Regarding adaptive metabolism, seminal studies reported on the relevance of NADPH 60,61, glutathione 62,63 and amino acid metabolism 64 . Nrf2 and phase II detoxification were found to have a major role in antioxidant response. Transcriptional regulation 65 such as NF-kB 66 and antioxidant enzyme levels 67 were reported.
In our study, already a short-term exposure to hydrogen peroxide resulted in significantly changed morphology and a perturbed metabolism. More specifically, the external hydrogen peroxide treatment did not uniformly affect all organelles within the cells but primarily affected the mitochondria. Although we introduced H 2 O 2 to the cell culture medium and only applied a short 2 h incubation, hydrogen peroxide was able to diffuse into the cells and exert its effects, as clearly indicated by both the imaging and metabolomics. A striking result from our investigations is the high and strongly localized MitoSOX to MitoTracker ratio, indicating specific activation of the organelle. We see this despite the exogenous ROS source and the need for diffusion through the cytosol to the mitochondria. A possible explanation for this is that although the exogenous ROS affect the whole cell systematically, the primary effect is the exhaustion of antioxidant capacity, which manifests itself at the mitochondrial site, where the cell is unable to compensate for electron leakage from the ETC. This possible explanation is further supported by the morphological and network analysis of mitochondria. Multi-parametric image analysis revealed a significant reduction of the average and maximal mitochondrial length (Fig. 2), suggesting an imbalance of the fusion/fission equilibrium from mitochondrial network toward single organelles. Mitochondrial fragmentation (fission) represents the first step towards mitophagy and mitochondrial turnover, which is particularly relevant in the stress response to increased oxidative stress 68 . The finding of rapid adaptation to redox stress, as reflected in significant changes at different levels already after short-term exposure, is in accordance with previous studies. Intestinal cells were found to be able to translocate Nrf2 into the nucleus upon oxidative stress or mechanical stimulation within one hour 18  www.nature.com/scientificreports/ challenge within one hour from the stimulation and are able to up-regulate the pentose phosphate pathway (PPP), as well as to modulate glycolysis and the TCA cycle 69 . Furthermore, human skin cells also activate PPP and recycle glycolytic intermediates to maximize NADPH production even in response to ultra-short exposure to redox stress within seconds 70 . Also in this work, the structure of mitochondria was in agreement with the metabolic readout, with increased fragmentation of the mitochondrial network (Fig. 2). A significant decrease in GTP concentrations was observed, which aligns well with changes in the mitochondrial network, considering the dependence on local GTP concentrations, as three GTPases facilitate the fusion and division of mitochondrial membranes. Indeed, it was previously described that the fusion-fission equilibrium mirrors metabolic status: the ramified network (fused form) being prevalent during cell respiration and ATP production 71 , and increased clustering was related to a decrease in OXPHOS efficiency 33 . Indeed, several lines of experimental evidence imply mutual regulation between cellular energetic status and mitochondrial fusion/fission equilibrium 68 , making it even more important to monitor metabolic and morphologic adaptation in parallel.
Cancer cells exhibit lower levels of respiration in favor of aerobic glycolysis as well as remarkable metabolic plasticity in comparison with normal tissue. Despite this, it was previously reported, that the investigated colon carcinoma cell line HCT116 generates about 60% of its total ATP via OXPHOS 72,73 . Therefore, there is still a substantial margin for OXPHOS disruption upon mitochondrial impairment. Correspondingly, the cellular energy charge decreased upon treatment by 60% in our findings. Interestingly, phosphocreatine concentrations www.nature.com/scientificreports/ were also substantially lowered, which can help to regenerate ATP during buffer fluctuations and maintain cellular energy homeostasis 74 . This seems plausible, as the synthesis of reduced glutathione is energy expensive and requires two ATP-dependent reactions. Apart from adenosine nucleotides, multiple other purine and pyrimidine nucleotides have altered concentrations. Damage upon ROS exposure can induce DNA repair mechanisms and de novo nucleotide synthesis 75,76 , and redox stress primarily damages the mitochondrial DNA over nuclear DNA 77 .
The metabolic signature of antioxidant defense was accurately measured. It is also important to note that, in case of such a short-term response, cells primarily rely on the basal expression of ROS scavenging enzymes (catalases, superoxide dismutase, glutathione peroxidase and reductase, etc.) and small molecule antioxidants like reduced glutathione, until transcriptional regulation can be adapted 78 . Not only altered GSH/GSSG ratios (reduced glutathione pools due to H 2 O 2 treatment, and an increase of GSSG), but also multiple changes throughout the glutathione-synthesis pathway indicate a systematic response. Beyond the first-line response of building the GSSG dimer from its reduced monomer and using its reducing power, cells started to replenish the depleted GSH pools. Changes could be observed in several differentially regulated GSH precursors such as cysteine, glutamyl-cysteine, homocysteine, glutamate. These findings were also in perfect agreement with the live cell imaging analysis of the mitochondrial superoxide (Fig. 1). Additionally, the closely related cysteine and methionine metabolism was widely perturbed.
Regarding NADPH/NADP and NAD/NAD ratios, key indicators of the cellular energy state were not considered in this work, as tailored sample preparation is needed for their accurate analysis 79 . However, the indirect effects on pathways were monitored. Upon further redox stress, there is a high NADPH demand for restoration of the reducing capacity of several enzymes in antioxidant defense (co-factor of glutathione reductase, catalase). To prioritize NADPH synthesis, glycolytic flux is diverted into the oxidative branch of the pentose phosphate pathway 75,76 , and glycolytic intermediates from the non-oxidative phase of PPP can be recycled via gluconeogenesis [80][81][82] . Our results are in agreement with such a coordinated interplay as we see strong, consistent changes throughout these pathways. Increased amounts of glycolytic metabolites could also support the energy needs for glutathione synthesis. The other major NADPH producer apart from PPP is malic enzyme 83,84 , which seems to be also involved, as its reaction partners malate and pyruvate are significantly altered as well.
Diminishing NAD+ concentration can introduce a bottleneck in the conversion of isocitrate to alpha-ketoglutaric acid, thereby stalling upstream metabolites in the TCA cycle and increasing citrate, cis-aconitate and isocitrate concentrations, which is exactly what we observed in the experiment. Furthermore, the results correspond well regarding oxidative inhibition and regulation of TCA enzymes due to high mitochondrial redox stress levels 49,85 . The reduced rate of oxidative phosphorylation (which manifests itself by the increased AMP to ATP ratio) is a response to minimize the endogenous ROS generation, as has been presented and discussed in previous works 70,86,87 . Mitochondria play a pivotal role in cellular fate. Preserving cellular physiology demands tight control of mitochondrial fusion and fission, which are key regulating processes of morphology. There is broad evidence that the morphology and bioenergetic status of mitochondria are linked 88 . The cellular energy status is given by the intracellular nucleotide pool and ATP production. The energy metabolism is in turn connected to the cellular reactive oxygen species (ROS) homeostasis, working on a delicate equilibrium between ROS and the cellular antioxidant system. The advancements in understanding reciprocal, responsive processes of morphological regulation and cellular bioenergetics status emphasized the need for methods allowing to dissect these intertwined processes also with regard to the redox status.
Here we demonstrate that precise metabolomics data relying on internal standardization and absolute quantities combined with live cell imaging are suitable for in-depth investigation of complex biological processes like short-term oxidative stress. As all the molecular signature, spatial information and morphology highlight the implication of mitochondria, we hold that this approach is a viable alternative to workflows with selective mitochondrial extraction. Cellular compartment-specific metabolomics is challenging and so far it only rarely solved problems. Conventional organelle fractionation techniques followed by extraction and LC-MS based measurements are known to introduce significant biases, as lengthy times between isolation and extraction hamper rapid quenching of metabolism. Chen et al. 25 introduced a superb solution for rapid mitochondrial isolation. Their workflow allowed to obtain metabolic profiles of mitochondria, however, as a drawback it relied on the use of genetically modified in vitro models. While our approach cannot pinpoint the cellular compartment for metabolome perturbations, we mutually validated the functional and quantitative information about the cellular state by the orthogonal information of live cell imaging and accurate absolute metabolite quantities. Overall, we were able to recapitulate established notions previously described in regard to oxidative stress in the metabolism, cellular state and morphology, as well as demonstrate the correlation between these effects. Furthermore, without the need to use an epitope-tag, the methodology can easily be extended to any adherent cell line, avoiding at the same time possible negative implications (e.g. interference with the folding of the target protein and biological activity) introduced by the tag.

Materials and methods
Methanol and water were LC-MS-grade from Fisher Scientific or Sigma Aldrich; ammonium formate, ammonium bicarbonate eluent additives for LC-MS, hydrogen peroxide (H 2 O 2 ) and N-ethylmaleimide (NEM) were purchased from Sigma Aldrich. www.nature.com/scientificreports/ (FCS) (BioWest) without antibiotics at 37 °C under a humidified atmosphere containing 5% CO 2 . Cell culture media and reagents were obtained from Sigma-Aldrich, and all plastic dishes, plates and flasks were from Star-Lab unless stated otherwise.

Material and methods metabolomics experiment.
Experiment: seeding, treatment and extraction. Conditions in the metabolomics and imaging cultivation plates were matched in regard to seeded cell density and available growth medium. For metabolomics 250 × 10 3 HCT116 cells per well were seeded (N = 8) in a 12-wellplate format (12-well CytoOne, TC-Treated) in 1 mL of McCoy's 5a medium (Sigma-Aldrich) supplemented with 10% fetal calf serum (FCS) (BioWest) and 4 mM l-glutamine without antibiotics and incubated at 37 °C (StarLab) under a humidified atmosphere containing 5% CO 2 . At the same time for live cell imaging 70 × 10 3 cells and 291 µL medium per well were seeded into the imaging wells (Ibidi), in order to match the surface area ratio of the metabolomics well and imaging well (3.5 cm 2 vs 1 cm 2 ). This way we intended to achieve a similar degree of confluence at the day of the experiment. Two days following seeding, cells were still subconfluent. First, the medium was removed, wells were washed with pre-warmed PBS (37 °C) and 500 µM hydrogen peroxide in live cell imaging solution were added, while for controls live cell imaging solution only (also 37 °C pre-warmed) was added. After incubation for 2 h, imaging and metabolomics wells were treated and prepared for measurement simultaneously..

Extraction of metabolites.
We applied a combination of cold methanol extraction with NEM-derivatization: the extraction solvent consisted of 80% methanol (Fisher Scientific) and 20% aqueous 10 mM ammonium formate adjusted to pH 7 containing 25 mM N-ethylmaleimide (NEM) in order to form NEM-adducts of primary thiols 89 . The extraction solvent was prepared freshly prior to the experiment. Weighed in from NEM powder, dissolved in 10 mM ammonium formate with pH adjusted to 7.0 The solution was mixed with methanol and cooled down at − 20 °C.
Cells were washed three times with phosphate-buffered saline (PBS) (Sigma-Aldrich) (37 °C) and snap-frozen with liquid nitrogen. 20 μL fully 13 C labeled (U 13 C) internal standard from ISOtopic solutions e.U. (dissolved in 5 mL 10 mM NH 4 FA at pH 7) was added as well as 180 μL extraction solvent consisting of 20% 25 mM NEM in 10 mM NH 4 FA and 80% methanol). Cells were scraped off in the extraction solution and transferred to Eppendorf tubes as described elsewhere 90 . During extraction, the samples were kept on ice. Subsequently, samples were vortexed and centrifuged (14,000 rcf, 10 min, 4 °C) and from each sample 100 µL of supernatant was transferred into a corresponding MS-vial and 50 µL extract was used to collect a pooled quality control sample. Samples were measured directly without evaporation.
The applied LC-HRMS method was adopted from 35 as described in 91 . In short, a SeQuant ZIC-pHILIC column (150 × 2.1 mm, 5 µm, polymer, Merck Millipore) with a 15-min long gradient under alkaline conditions with eluents 10 mM ammonium bicarbonate, pH 9.2/10% acetonitrile and 100% acetonitrile were used. The measurement sequence was randomized, blank and pooled quality control samples were injected at regular intervals. The high-resolution mass spectrometer Thermo Scientific™ Q Exactive HF™ quadrupole-high field Orbitrap mass spectrometer was being operated in positive/negative ion-switching mode. External calibration with 133 compounds involving U 13 C internal standardization was carried out.  93,94 . From the 51 significantly changed metabolites (adjusted p-value cut-off 0.05, group variance: equal), 48 metabolites were mapped on KEGG, as mannitol, erythrol and N-acetyl-serine could not be retrieved. Following this, a subnetwork was extracted based on the mapping of significant metabolites.

Material and methods imaging experiments. Evaluation of mitochondrial superoxide production and
morphology. In parallel to metabolome analysis, cells were incubated for live cell imaging experiments. Reference measurements of mitochondrial superoxide production and morphometric analysis were performed via imaging workflows to accurately monitor cell status with an independent experimental set-up.
Mitochondrial morphology was visualized with MitoTracker dye and mitochondrial superoxide production with MitoSox dye (both from Thermo Fisher Scientific), as previously described 18,31 . For the imaging, we removed the cell culture medium and incubated the cells with staining solutions (1:1000 dilution in Live Cell Imaging solution, Thermo Fisher Scientific) for 15 min. At the end of the staining, cells were rinsed twice with pre-warmed PBS and immediately imaged in Live Cell Imaging solution.
For microscopy acquisition, we used a confocal Zeiss microscope 710\ELYRA system PS.1 equipped with Plan-Apochromat 63×/1.2 water objective for live cell imaging and an Andor iXon 897 (EMCCD) camera. To this aim, ROI (regions of interest) were randomly selected from the MitoTracker images. During this step, the MitoSox channel was temporarily disabled to avoid selection bias. Multiparametric morphological evaluation of the mitochondrial network was performed according to the method of Valente et al. 34 . Data resulted from the quantification of at least three different optical fields/30 ROIs for every experimental condition.

Data availability
Metabolomics data (LC high-resolution mass spectrometry-based metabolomics dataset in rawdata have been deposited to the EMBL-EBI MetaboLights database (Haug et al., 2020)  www.nature.com/scientificreports/ complete dataset can be accessed here http:// www. ebi. ac. uk/ metab oligh ts/ MTBLS 2672. The processed metabolomics data with absolute concentration, the morphological analysis and constructed subnetwork is provided in the supplementary material. Further information about the datasets generated during and/or analyzed during the study are available on request from the corresponding author G.K. (gunda.koellensperger@univie.ac.at).