Single-cell heterogeneity in the chloroplast redox state mediates acclimation to stress in a marine diatom

Diatoms are photosynthetic microorganisms of great ecological and biogeochemical importance, forming vast blooms in diverse aquatic ecosystems. Current understanding of phytoplankton acclimation to stress is based on population-level analysis, masking cell-to-cell variability. Here we investigated heterogeneity within Phaeodactylum tricornutum populations in response to oxidative stress, which is induced by environmental stress conditions. We combined flow cytometry and a microfluidics system for live imaging to measure redox dynamics at the single-cell level using the roGFP sensor. Chloroplast-targeted roGFP exhibited a light-dependent, bi-stable oxidation pattern in response to H2O2, revealing distinct subpopulations of sensitive oxidized cells and resilient reduced cells. Subpopulation proportions depended on growth phase, linking the bi-stable phenotype to proliferation. Oxidation of chloroplast-targeted roGFP preceded commitment to cell death and was used as a novel cell fate predictor. We propose that light-dependent metabolic heterogeneity results in differential stress responses that regulate cell fate within diatom populations.

Introduction membrane properties) and through aquaporins channels 32 . Combined properties as lower toxicity, diffusibility and selective reactivity make H2O2 suitable for studying signaling in various biological systems 23,28 . Since many environmental stressors induce ROS generation, application of H2O2 can reproduce the downstream cellular response. H2O2 application in marine diatoms led to oxidation patterns similar to other environmental stressors 16,17 . It also led to the induction of cell death, in a dose-dependent manner, with characteristics of PCD that included externalization of phosphatidylserine, DNA laddering, and compromised cell membrane 16 .
In addition, early oxidation of the mitochondrial GSH pool preceded subsequent cell death at the population level following exposure to H2O2 and diatom-derived infochemicals in the model diatom Phaeodactylum tricornutum 16,33 .
In the current work, we investigated phenotypic variability within diatom populations in response to oxidative stress. Until recently, research of phytoplankton's responses to environmental stress was carried out primarily at the population level, masking heterogeneity at the single-cell level. Cell-to-cell variability could result in different cellular strategies employed by the population. We established single-cell approaches to measure in vivo oxidation dynamics in the model diatom P. tricornutum using flow cytometry and microfluidics liveimaging. To measure oxidation dynamics of specific organelles, we used P. tricornutum strains expressing redox-sensitive GFP (roGFP) targeted to different sub-cellular compartments 17 . The oxidation of roGFP is reversible, and can be quantified using ratiometric fluorescence measurements 34 . The roGFP oxidation degree (OxD) reports the redox potential of the GSH pool (EGSH), which represents the balance between GSH and its oxidized form 34 . Therefore, roGFP OxD provides a metabolic input of the redox state of the cell, and represents the oxidation state of native proteins in the monitored organelle 17,34 . By measuring redox dynamics at singlecell resolution, we uncovered a previously uncharacterized phenotypic heterogeneity in the response of a marine diatom to oxidative stress. We revealed a bi-stable response in the chloroplast EGSH following H2O2 treatment, exposing distinct subpopulations with different cell fates. Our results revealed a specific link between oxidation patterns in the chloroplast and cell fate regulation.

Bi-stable chloroplast roGFP oxidation in response to oxidative stress reveals distinct subpopulations.
To assess roGFP oxidation in an organelle-specific manner, we measured OxD in P. tricornutum strains expressing roGFP targeted to the chloroplast, nucleus and mitochondria using flow cytometry. At steady-state conditions without perturbations, the OxD distribution in the population had a single distinct peak, representing a reduced state at all examined compartments ( Fig. 1 A, E, I and Figs. S1, S2). Application of H2O2 led to organelle-specific and dose-dependent oxidation patterns in these organelles (Fig. 1).
The chloroplast-targeted roGFP (chl-roGFP) exhibited a distinct bi-stable response following treatments of 50-100 µM H2O2, revealing two distinct subpopulations of "oxidized" and "reduced" cells ( Fig. 1 B-D and Fig. S3 A-C). These subpopulations emerged within the first few minutes after H2O2 addition ( Fig. 1 B-D). The existence of these subpopulation is masked in bulk analysis, demonstrating the importance of single-cell measurements ( Fig. S1). In the "oxidized" subpopulation, roGFP completely oxidized (~100%) in response to H2O2, reaching a similar distribution of the fully oxidized positive control (200 µM H2O2) ( Fig. 1 B-D  intermediate oxidation, suggesting that these subpopulations represent discrete redox states. Interestingly, a larger fraction of cells was within the "oxidized" subpopulation at 20-25 minutes post treatment compared to later time points, indicating that some cells were able to recover during this time (Fig. 1 M). The proportion between these subpopulations stabilized after 46-51 min post treatment, and was H2O2-dose dependent, as more cells were within the "oxidized" subpopulation at higher H2O2 concentrations ( Fig. 1 A-D, M). The quick emergence of stable co-existing "oxidized" and "reduced" subpopulations exposed underlying heterogeneity within the diatom population, resulting in a differential response to oxidative stress. This clear bi-stable pattern was unique to the chloroplast-targeted roGFP. Nuclear roGFP displayed a continuous distribution in response to H2O2 treatments, and no distinct subpopulations could be observed ( Fig. 1 F-H). Within minutes post treatment, nuclear roGFP exhibited fast oxidation even in response to a low H2O2 concentration of 50 µM, which had only a mild effect on the chloroplast (Fig. 1 B, F). At that concentration, nuclear roGFP oxidation was followed by a gradual and much slower recovery, which lasted >5 hours post treatment (Fig. 1 F). At higher concentrations, the entire population was oxidized within 3 minutes post treatment, and most cells remained stably oxidized >5 hours post treatment ( Fig. 1 G-H). The mitochondrial roGFP exhibited a heterogeneous redox response, as seen in the 80 µM and 100 µM H2O2 treatments starting at ~24 minutes post treatment (  S4). Therefore, we chose to focus on the chl-roGFP strain, which revealed two discrete subpopulations. respectively are shown as a reference. The "oxidized" and "reduced" subpopulations are marked by red and blue dashed boxes respectively (C, G, K). The experiment was done in triplicates, for visualization the first replica is shown except for the first 4 min in which all replicates are shown for higher temporal resolution. Each histogram consists of >8000 (A-D), 50

Oxidation of chl-roGFP precedes the induction of cell death.
Next, we examined the possible link between early chloroplast EGSH oxidation and subsequent mortality. We quantified cell death 24 hours post H2O2 treatment using flow cytometry measurements of Sytox green staining, which selectively stains nuclei of dying cells. The fraction of "oxidized" cells 1-2 h post treatment was correlated with the fraction of dead cells at 24 h ( Fig. 2 A, Spearman's correlation coefficient rS=0.98, P=4.6·10 -8 ), suggesting that early oxidation in the chloroplast in distinct subpopulations may predict cell fate at much later stages.
To investigate directly the link between early chl-roGFP oxidation and subsequent cell death, we used fluorescence-activated cell sorting (FACS) to sort cells based on chl-roGFP oxidation and measured their survival. Single cells of the "oxidized" and "reduced" subpopulations were sorted into fresh media at different time-points following the addition of 80 µM H2O2, and colony forming units were counted to assess survival ( Fig. 2 Fig. 2 D). These results suggest that after a distinct exposure time, cell death is induced in an irreversible manner in the "oxidized" subpopulation. In contrast, the "reduced" subpopulation from the same culture and treatment exhibited a high survival rate similar to the control at all time-points examined, demonstrating its resilience to the stress (Fig. 2 D, P≥0.86 for all comparisons, Dunnett test). In agreement with these findings, cell death measurements using Sytox staining of sorted subpopulations also showed higher mortality in the "oxidized" cells compared to the "reduced" and the control cells, which remained viable ( Fig. S5). Taken together, these results demonstrate that the "oxidized" subpopulation was sensitive to the oxidative stress applied, which led to induction of cell death in those cells, while the "reduced" subpopulation was able to survive. In addition, we detected a distinct phase of "pre-commitment" to cell death, ranging approximately 30-100 min in most cells, during which the fate of the "oxidized" subpopulation is still reversible upon removal of the stress.

Early oxidation of chloroplast EGSH predicts cell fate at the single-cell level.
In order to track oxidation dynamics and subsequent cell fate of individual cells, we established a microfluidics platform for in-vivo long-term epifluorescence imaging, under controlled flow, light and temperature conditions adapted for diatom cells (Fig. 3, Fig. S6 and movies S1, S2). We introduced cells expressing chl-roGFP into a custom-made microfluidics device, let the cells settle, and introduced treatments of either 80 µM H2O2 or fresh media (control) continuously for 2.5-3 hours, after which the treatment was washed by fresh media (see methods). In addition, the use of microfluidics enabled imaging of the basal OxD state of single cells prior to treatment, as well as the introduction of Sytox green at the end of the experiment to visualize cell death. We detected the distinct "oxidized" and "reduced" subpopulations following 80 µM H2O2 treatment, similar to the flow cytometry experiments (Fig. 3 C, F, and movie S1). However, no clear differences were observed in their OxD prior to treatment ( Fig. S7 and movie S1). The separation between the subpopulations emerged within 20 minutes of exposure to 80 µM H2O2, and remained stable over the course of the experiment with the "oxidized" subpopulation maintaining a high OxD above 80% (Fig. 3 F, Fig. S7 B and movie S1). The "reduced" subpopulation exhibited an immediate response to H2O2 comparable with flow cytometry measurements, from 25-45% OxD before treatment to 30-65% OxD during the first 20 minutes post 80 µM H2O2 treatment ( Fig. 3 F, Fig. S7 B, and movie S1). Following this initial oxidation, the "reduced" cells recovered gradually over the next hours, reducing to 5-25% oxidation 8 hours post treatment, below the initial basal state (Fig. 3 F, movie S1). A gradual slow reduction was also observed in the control cells over the course of the experiment (Fig. 3 E, movie S2), which may represent acclimation to the experimental setup or a diurnal redox alteration. Control cells did not oxidize in response to addition of fresh media (Fig. 3 E and Fig. S7 A), excluding the possibility that the oxidation observed in 80 µM H2O2 treated cells was due to shear stress during treatment.
We detected a clear correlation between initial oxidation in the chloroplast in response to oxidative stress and subsequent cell fate (Fig. 3 A-G). Cells that exhibited high chl-roGFP oxidation within the first 40 minutes also died at a much later stage, while cells that maintained a lower OxD were able to recover (Fig. 3 G). Logistic regression modelling of cell death as a function of chl-roGFP OxD at this time-point revealed a threshold of ~74% OxD, that separated between cells that were dead or alive at the end of the experiment (Fig S8). Using this threshold, the OxD of chl-roGFP 40 minutes post H2O2 treatment predicted subsequent death of individual cells as measured ~23 hours post treatment with a high accuracy of 98.8% (0.8% false positive, 1.7% false negative; Fig. 3 G and Fig. S8). These results corroborate the flow-cytometry analysis and demonstrate that under these conditions early chloroplast EGSH response accurately predicts cell fate at the single-cell level.  The distinct subpopulations derive from phenotypic variability and not from variable genetic backgrounds.
The differential chloroplast oxidation of the observed subpopulations could be due to genetic variability or due to phenotypic plasticity within the population. To differentiate between the two scenarios, we sorted chl-roGFP individual cells of the "oxidized" and "reduced" subpopulations 30 and 100 minutes post 80 µM H2O2 treatment as well as untreated control cells, and regrew them to generate clonal populations derived from cells exhibiting specific phenotypes. The clonal progeny cultures were subsequently exposed to 80 µM H2O2 and their chl-roGFP oxidation was measured. The two distinct subpopulations were detected in all the clones measured, and the fraction of the "oxidized" subpopulation was again correlated with cell death ( Fig. 4 and Fig. S9). Therefore, the different subpopulations observed did not originate from genetic differences, but rather represent phenotypic variability within isogenic populations. The "oxidized" subpopulation is enriched with cells at G1 phase.
One possible source for phenotypic variability in genetically homogenous populations can be explained by differences in the cell cycle phase, as the cell cycle is linked to metabolic changes including redox oscilations [35][36][37][38] . Therefore, we sorted the "oxidized" and "reduced" subpopulations 30 min following 80 µM H2O2 treatment into a fixation solution, and stained the fixed cells with 4′,6-diamidino-2-phenylindole (DAPI) to quantify DNA content for cell cycle analysis. Considering P. tricornutum divides about once a day 39 , we assumed that at this early time point the DNA content is likely to reflect the status of the cell prior to treatment, and is not yet affected by it. The sorted "oxidized" subpopulation had a higher fraction of cells at G1 (86.9 ± 1.8%) compared to control untreated cells (76.8 ± 0.7%, P=0.0024, Tukey test). The "reduced" subpopulation on the other hand had a smaller fraction of G1 cells (68.7 ± 2.2%) compared to both control (P=0.011, Tukey test) and "oxidized" cells (P=0.0001, paired t-test), and exhibited a larger fraction of G2/M cells (Fig. 5). These results demonstrate that although cell cycle phase alone cannot explain the differences between the subpopulations, it is linked to the chloroplast EGSH response to oxidative stress and may represent an important factor that affects H2O2 sensitivity in the population.

The fraction of "oxidized" cells is age dependent.
To further explore the physiological factors that may affect the bi-stable distribution, we investigated the effects of different growth phases on oxidation of chl-roGFP in response to H2O2. Since aging has been associated with changes in redox states in several model systems [40][41][42] , we hypothesized that culture growth phase may affect the susceptibility of the population to oxidative stress. We challenged cultures originating from the same batch at different growth phases with 0-100 µM H2O2 after adjusting cell concentrations between growth phases (see methods, Fig

The bi-stable chloroplast redox response is light dependent.
Photosynthesis is the major source for reductive power as well as ROS in algal cells, and exposure to dark was shown to increase sensitivity to oxidative stress in another marine diatom 18 . Therefore, we hypothesized that light regime will affect the bi-stable pattern of chl-roGFP following oxidative stress, and investigated the effects of short exposure to darkness during daytime. Cells were treated with 0-100 µM H2O2 and were immediately moved to the dark for 90 minutes, after which they were moved back to the light (dark treated, Fig. 7 and Fig. S13). These cells were compared to cells that were kept in the light during this time (light treated). The transition to the dark caused an immediate oxidation of the basal chl-roGFP OxD (without H2O2 treatment), reaching a peak within 15 minutes (Fig. 7 A, D). Then, while still under dark, chl-roGFP gradually reduced while maintaining a continuous distribution (Fig. 7 A, D). Upon shifting back to the light, chl-roGFP reduced within 2 minutes back to its basal state prior to dark exposure ( Fig. 7 A, D). The dark mediated oxidation was specific to the chloroplast and was not detected in the nucleus (Fig. S14), demonstrating the organelle specificity of these redox fluctuations.  co-existence of distinct subpopulations that employ diverse cellular strategies to improve the survival of this globally important phytoplankton group, as was shown in other microorganisms [43][44][45] . In this study, we established a novel system for studying phenotypic variability in the marine diatom P. tricornutum using flow cytometry and a microfluidics system for live imaging microscopy. Using organelle-specific measurements of EGSH dynamics we assessed the in vivo redox state of individual diatom cells, and were able to detect a differential response to oxidative stress within the population. Based on this metabolic parameter, we identified two distinct subpopulations that emerged as an early response to oxidative stress, demonstrating the importance of phenotypic variability in cell fate regulation in diatoms. We propose that this variability leads to a differential response to environmental stressors. Exposure to specific stress conditions leads to ROS accumulation at different subcellular compartments, including the chloroplast, which is used to sense the stress cue and regulate cell fate. Cells at a more susceptible metabolic state will accumulate high ROS levels and will subsequently die, as observed in the "oxidized" subpopulation. More resilient cells will exhibit milder oxidation and will be able to acclimate, as observed in the "reduced" subpopulation. Chloroplast EGSH oxidation is an early stage in this stress response and precedes the commitment to cell fate.
We propose that in diatoms, the chloroplast EGSH is involved in sensing specific environmental stress cues that induce oxidative stress, and in cell fate regulation (Fig. 8). Depending on the specific stress, ROS can accumulate at various sub-cellular compartments 16 "death threshold" based on early chl-roGFP oxidation (Fig. 3 G). This early response provided accurate cell fate predictions at the single-cell level. We propose that the balance between the prior metabolic state, antioxidant capacity, and the magnitude of the applied stress determines whether a cell will cross the "death threshold", leading to a differential response within the population. Harsher stress conditions will have a stronger effect on the population, leading to more cells crossing the threshold and exhibiting early oxidation and subsequent cell death, as shown with increasing H2O2 doses (Fig. 1 M). Similarly, different basal metabolic states can increase the sensitivity of the entire population, leading to a larger fraction of "oxidized" cells even when they are exposed to lower H2O2 doses, as observed during culture aging (Fig. 6). The increased sensitivity in the dark ( Fig. 7) further supports an involvement of the chloroplast in cell fate determination, suggesting that fluctuating light conditions experienced by diatoms in natural environments may greatly affect their susceptibility to additional stressors.
Redox fluctuations in the chloroplast can serve as a rapid mechanism to perceive specific environmental cues, by regulating key metabolic pathways on the post-translational level, prior to gene expression. Analysis of the redox-sensitive proteome in P. tricornutum revealed over-representation of chloroplast-targeted proteins, that were also oxidized to a greater degree under H2O2 treatment as compared to other subcellular compartments 17,46 . Chloroplast EGSH oxidation preceded the "point of no return" after which cell death was irreversibly activated (Fig. 2 D), supporting it has a role in sensing the stress cue. This "pre-commitment" phase provides an opportunity for cells to recover if conditions change during a narrow time frame of ~30-100 min following oxidative stress (Fig. 2 D), before the cell has accumulated damage beyond repair or a PCD cascade was fully activated. This "pre-commitment" phase was shown previously in two diatom species, as exogenous application of the antioxidant GSH rescued diatom cells from otherwise lethal treatments of infochemicals or H2O2 only during the first hour 16,18 . These findings underscore the importance of redox regulation in chloroplast metabolic reactions, and its involvement in stress sensing and cell fate determination.
The role of the chloroplast in mediating PCD remains elusive, although mitochondria-generated ROS are known to play a key role in PCD in plants and animals 47,48 . This knowledge gap is even greater in unicellular marine algae, for which the molecular basis for the PCD machinery is largely unknown 8 . The chloroplast is a major source for generation of both ROS and reductive power to generate and recycle NADPH, thioredoxin and GSH 23 . Chloroplast-generated ROS were demonstrated to be involved in plants in retrograde signaling from the chloroplast and in hypersensitive response cell death 23,31,47,49 . A recent model suggested possible mitochondria-chloroplast cooperative interactions in the execution of ROS-mediated PCD 47 . In diatoms, mitochondrial ROS were linked to cell death in response to diatom-derived infochemicals 16 , and chloroplast EGSH was shown to mediate changes in oxidative stress sensitivity upon light-dark transitions 18 . We propose that redox dynamics of both the mitochondria and the chloroplast are involved in cell fate regulation in diatoms.
The source of the single-cell variability observed in our system is yet to be further explored, but the results provide insights into factors that may drive it. Since clonal populations originating from single-cell isolates maintained the bi-stable chloroplast response, the variability does not result from genetic differences but rather from phenotypic plasticity (Fig. 4). It remains to be investigated whether the emergence of the subpopulations represents heterogeneity that occurs following exposure to stress, or rather a pre-existing variability within the population. Nevertheless, the difference in cell-cycle phase distribution between the subpopulations supports the latter. The combination of factors such as life history 33,50 , cell cycle phase 35,37 , cell age 51,52 , metabolic activity 53,54 , heterogeneous microenvironment 55 and biological noise 43,56 results in a distribution of different metabolic states within the population 57,58 . The metabolic state of the cell can affect its antioxidant capacity, therefore resulting in variability in sensitivity to oxidative stress as observed here.
Importantly, the mechanism that generates variability in our system is light-dependent, as the bi-stable chl-roGFP pattern was abolished when the cells were under darkness and the entire population became more sensitive to oxidative stress (Fig. 7). The antioxidant capacity of a diatom cell depends on photosynthesis-generated NADPH, which is also used for GSH recycling. The transition to the dark may have compromised the biosynthesis and recycling of GSH, therefore enhancing sensitivity to oxidative stress 18 . Furthermore, stationary cultures were shown to have lower photosynthetic activity 59 , which results in a lower flux of NADPH generation and may lead to the increased sensitivity observed in stationary cultures (Fig. 6).
Taken together, the source for heterogeneity could be variability in the flux of photosynthesis-derived reductive power.
Phenotypic variability can provide an important strategy to cope with fluctuating environments in microbial populations 57 . Future studies are required to investigate the possible tradeoff involved in maintaining high antioxidant capacity. For example, resilience to oxidative stress may come with a cost in the ability to sense environmental cues with high precision, as high ROS buffering capacity may mask milder ROS cues 46 .
Co-existence of subpopulations with different susceptibilities to specific stressors can be viewed as a "bethedging" strategy of the population, enabling at least a portion of the population to survive unpredicted stress events and subsequently leads to a growth benefit at the population level 44,45,51,57 . A recent study demonstrated the benefit of phenotypic variability when NH4 + limitation increased the cell-to-cell variability in N2 fixation in the bacterium Klebsiella oxytoca, leading to improved growth under fluctuating conditions at the population level 44 . Phenotypic plasticity in response to stress was also shown in the toxic phytoplankton Heterosigma akashiwo, that improved its chances of evading deleterious turbulent conditions by diversifying its swimming directions 45 . Furthermore, phenotypic variability enables individual cells within isogenic populations to exhibit various cell fates, including PCD, despite possible detrimental effects at the single-cell level. In diatoms, phenotypic variability in cell size, shape and susceptibility to stress conditions were suggested 1,16,18,60 , but until now the experimental setups were not designed to study individuality in stress response. Redox-based phenotypic variability may provide a rapid and adjustable strategy to cope with unpredicted stress conditions as compared to relying only on genetic diversity.
The novel approaches developed here provide new insights into individuality in marine microbes, and enable studying dynamic processes at the single-cell level in diatoms and possibly other ecologically relevant microorganisms. The ecological importance of variability in the chloroplast redox state and the mechanisms that underlie differential sensitivity to oxidative stress are yet to be explored.

Methods
Culture growth: P. tricornutum accession Pt1 8.6 (CCMP2561 in the Provasoli-Guillard National Center for Culture of Marine Phytoplankton) was purchased from the National Center of Marine Algae and Microbiota (NCMA, formerly known as CCMP). Cultures were grown in filtered sea water (FSW) supplemented with F/2 media 61 at 18°C with 16:8 hours light:dark cycle and 80 μmol photons m -2 sec -1 light intensity supplied by cool-white LED lights (Edison, New Taipei, Taiwan). Strains expressing roGFP were obtained as described previously 16,17 . Cultures were kept in exponential phase, experiments were performed in ~0.5-1·10 6 cells·ml -1 . roGFP measurements: roGFP oxidation was measured over time following the addition of H2O2 or in untreated control using the ratio between two fluorescence channels, i405 and i488, by fluorescence microscopy (described below) and by flow cytometry using BD LSRII analyzer, BD FACSAria II and BD FACSAria III. The roGFP ratio (i405/i488) increases upon oxidation of the probe 62  and i488red are the i488 of the maximum oxidized and maximum reduced forms respectively. For sorting purposes, roGFP ratio was used, as exact OxD cannot be calculated prior to sorting, both parameters give similar partition between the subpopulations (data not shown). In flow cytometry measurements, i405 was measured using excitation (ex) 407 nm, emission (em) 530/30 nm or 525/25 nm, and i488 was measured using ex 488 nm, em 530/30 nm. Relative expression level of roGFP was measured by multiplication of i405 and i488, and was used to gate roGFP+ cells during analysis (Fig.   S2 D). In sorting experiments, i405 and i488 were used instead to gate roGFP+ cells since relative roGFP expression could be calculated only post acquisition. Dynamic range of roGFP was calculated by ratio of Rox/Rred (tables S1, S2). Flow cytometry measurements were done under ambient light and temperature conditions, except for dark treatment during which samples were covered with aluminum foil. At time 0, H2O2 treatment was added from a freshly prepared 20 mM stock to P. tricornutum cultures to a final concentration of 5-200 µM. There were differences in roGFP fluorescence levels between the strains (Fig. S2 D). The mit-roGFP strain was less informative due to the lower roGFP fluorescence and therefore lower signal to noise ratio (SNR) and smaller dynamic range (Figs. S2 D and S4, table S2). It is important to note that the average leakage of chlorophyll auto-fluorescence (AF), especially into the i405 channel, increased over time starting ~116 minutes post 80 µM and 100 µM H2O2 treatments in all strains, but was most prominent in the mit-roGFP strain (Figs. S16 and S17). AF leakage was not linearly correlated with AF in the chlorophyll channel, and therefore could not be corrected at the single cell level. It may introduce a bias towards oxidation in roGFP oxidation calculations at later time points and can explain values above 100% OxD, but at early time points these effects were negligible at least in the nuc-roGFP and chl-roGFP strains (Figs. S16 and S17). (1.5% agarose + FSW/2 + F/2 + antibiotics) or "liquid" (FSW+F/20) fresh media. Cells grown in liquid were further diluted and then spotted on agar plates. 5-9 weeks post sorting the plates were scanned and colonies were counted manually, assuming each colony originates from a single surviving cell. Since survival was highly similar in liquid and in agar the results of these two methods were combined together. Each method was done in biological triplicates per medium type per experiment, data is shown for two independent experiments for time points 30 and 100 min and one experiment for the 60 min time point. For generation of clonal populations, single cells sorted into liquid medium were used. Clones were exposed to 80 µM and 100 µM H2O2 ~3-6 weeks post sorting, and their chl-roGFP oxidation was measured using flow cytometry. A total of 18 "control", 29 "oxidized" and 32 "reduced" clones were examined in two independent experiments.

Microfluidics chip preparation:
Microfluidics chip design was based on Shapiro et al. 63 , and was modified for P.
tricornutum cells. Each chip contained 4 channels of 2 cm length X 0.2 cm width X 150 µM height with one circular widening of 0.4 cm diameter, with a total volume of ~12.7 µl per channel (Fig. 3 H and Fig. S6 B). Each chip contained 4 chambers that were imaged sequentially: chl-roGFP control, chl-roGFP 80 µM H2O2 treated, WT 80 µM H2O2 treated and WT control (WT strain can be used to monitor auto-fluorescence changes and leakage during experiments). In each chamber, 5-6 different fields were imaged every 20 minutes over the course of >24 hours to avoid photo-toxicity. Ambient light was provided during light period using the microscope's BF illumination without a condenser, light intensity ranging between 34 (at the very edge, outside the imaging region) to 80 (center) μmol photons m -2 sec -1 . No images were obtained during the night to avoid disturbance to the diurnal cycle. After imaging the basal state of the cells, treatments of either 80 µM H2O2 dissolved in fresh media (FSW+F/2) or fresh media control were introduced to the system continuously for ~2.5-3 hours, after which they were gradually washed away by fresh media.
To quantify cell death, Sytox green was introduced into the system at 21.5-23 hours post treatment (see above) and was imaged using the roGFP i488 channel with shorter exposure time. The Sytox signal was stronger than the roGFP i488 and was localized mainly to the nucleus, enabling a clear separation between the two signals ( Fig. 3 B, D). Only a small fraction of cells within the control treatment were Sytox positive (0.0054%), indicating that cells remained viable in this experimental setup. Furthermore, cells of the "reduced" subpopulation and of control treatment were able to proliferate, further demonstrating their viability under these conditions (movies S1, S2).
Image analysis: Image analysis was performed using a designated MATLAB based script (see overview in Fig. S15) that is available on GitHub: https://github.com/aviamiz/ITRIA. Images were imported using bio-formats 64 . Then, image registration for XY drift correction was done using the Image Stabilizer plugin 65 for FIJI (Fiji Is Just ImageJ) and using MIJI 66 to access FIJI from MATLAB. Then images were normalized by bit-depth. Background subtraction was done based on mean value of a user-defined region of interest (ROI) that did not include cells. All fluorescence channels (i405, i488 and chlorophyll) were thresholded by a user-defined value to generate masks of positive expression. The roGFP relative expression level was calculated pixel-by-pixel by multiplication of the i405 and i488, only at pixels that were co-localized in the i405 and i488 masks. Then, roGFP relative expression (i405 * i488) was thresholded in order to include only pixels with high enough signal, based on a user-defined threshold. The roGFP ratio and OxD were calculated pixel-by-pixel as described above, pixels that were not included in the roGFP expression mask were excluded and set to NaN (not a number). For values of maximum oxidation and reduction of roGFP, cells were imaged in the same microfluidics imaging setup following treatments of 200 µM H2O2 and 2 mM DTT respectively (see "roGFP calculations"). Cell segmentation was based on i405 (chl-roGFP strain) or chlorophyll (WT strain) masks and fluorescence intensity using watershed transformation. Cells were filtered based on area, major and minor axis length, and eccentricity in order to exclude clumps of cells and doublets. Cell tracking was adapted and modified from a MATLAB code kindly provided by Vicente I.
Fernandez and Roman Stocker 67,68 . In short, particles were tracked based on minimizing the distance between particle centroids in adjacent frames within a distance limit. Sytox analysis was based on a user defined threshold and colocalization of the Sytox with the extended cell region within the cell segmentation mask. Images from the same experiment were analyzed using the same values for all thresholds and parameters, except for Sytox analysis in which the threshold was adjusted manually to validate correct assignment of cell-fate and to avoid effects of focus differences.
Cells that were not detected in the frame used for Sytox analysis or were not tracked for at least 6 consecutive frames were excluded from further analysis. The 74% OxD threshold used for early discrimination between the subpopulations and for cell fate prediction (Fig. 3  used as a reference to validate the gates for G1 and G2/M phases (data not shown). S phase was not clearly detected in this analysis, and therefore both gates likely include also S phase cells.

Growth phase:
The same culture batches of WT and chl-roGFP strains were diluted sequentially starting at day 0, and cell concentration was measured using Multisizer™ 4 COULTER COUNTER (Beckman Coulter) (Fig. 6 A and Fig. S12 A). On day 16, cultures at 4 different growth phases (early exponential, late exponential, early stationary and mid stationary, Fig. 6 A) were treated with 0-80 µM H2O2 and their chl-roGFP oxidation response and subsequent mortality were measured as described above. To exclude effects of different cell concentrations, 1-2 h before treatment cell concentrations were adjusted to 6.7·10 5 cells·ml -1 (concentration of early exponential culture) by centrifugation and resuspension in fresh media. In stationary cultures a small subpopulation (2-3%) with high AF was observed using ex 405 nm em 450/50 nm and exhibited OxD values >110%, and was therefore excluded from the analysis (table S1).
Statistics: All statistical analyses were done in R. ANOVA was used for multiple comparisons, and then Dunnett test or Tukey test were performed were applicable. For comparisons of two samples, t-test was used, and paired t-test was used where applicable. Values are represented as mean ± SEM unless specified otherwise. Box-plot was generated using the web tool BoxPlotR http://shiny.chemgrid.org/boxplotr/ 69 using Tukey whiskers, which extend to data points that are less than 1.5 x Interquartile range away from 1st/3rd quartile.
Data availability: All relevant data supporting the findings of the study are available in this article and its Supplementary   Information, or from the corresponding author upon request. Custom-made code for image analysis is available from the corresponding author upon request.  E). The ratio i405/i488 represents the roGFP oxidation state, and it increases upon oxidation. The "oxidized" and "reduced" chl-roGFP subpopulations are marked in red and blue respectively in B. (G) Flow cytometry measurements of roGFP OxD of the "oxidized" (dashed, empty symbols) and "reduced" (solid, full symbols) subpopulations over time post 0-200 µM H2O2 treatments. Maximum reduction following 2 mM DTT (black) is shown for reference. Results are shown as mean ± SEM, n=3 biological repeats. The ratio i405/i488 represents roGFP oxidation state (see methods). WT is shown for auto-fluorescence (AF) leakage reference, demonstrating the effect of lower expression level in mit-roGFP strain. A differential response was observed in exp 3 and in the exp shown in Fig. 1 K, L, but in exp 1 and 2 no distinct subpopulations were observed. This differential response was clearly observed only at later time points, which were not measured in exp 1 and 2, and may result from autofluorescecne leakage (see Fig. S16 and S17). Measurements were done in triplicates, one repeat is shown for visualization.  Chamber preparation Microfluidiuc chambers washed, cooled to 18 C Cell loading chl-roGFP and WT cells were introduced into parallel chambers at a changing flow rate (1-100 µl/min) with occasional stops. Following cell settlement flow rate was kept at 1µl/min. Ambient light was provided during light period using the microscope's bright field illumination.

Imaging
Images were obtained every 20 min during light period. 5-6 fields were imaged per chamber, using 4 channels: roGFP i405, roGFP i488, chlorophyll AF, and bright field. No images were obtained during the night to avoid disturbance to the diurnal cycle. Treatment Treatment was applied by introducing fresh media (control) or media containing 80 µM H 2 O 2 through the chamber inlet for ~2.5-3 hours. Inlet was then changed back to fresh media for the duration of the experiment.
Sytox staining 21.5-24 h post treatment Sytox green was introduced into the chamber, and was imaged to detect dead cells.    roGFP fluorescence (roGFP+, black) and the "oxidized" (red) and "reduced" (blue) subpopulations are marked. The experiment was done in biological triplicates, one representative repeat is shown for visualization. ≥9,000 cells per sample.      S13. Redox response of chl-roGFP to transition to the dark. Flow cytometry measurements of chl-roGFP oxidation in the population over time. Cells were treated with 0 µM (basal, A, same as in Fig. 2 A), 10 µM (B), and 30 µM H2O2 (C), and were then transitioned to the dark at time 0 (within 10 minutes post treatment). Cells were kept in the dark for 90 minutes (green) and were then transferred back to the light (cyan). For reference, chl-roGFP OxD following the same H2O2 treatment but without transition to the dark (black) and following maximum oxidation (200 µM H2O2, red) and maximum reduction (2 mM DTT, blue) are shown. The experiment was done in triplicates that were highly similar, for visualization purposes the first replica is shown. Each histogram consists of >8000 cells. . Cells were kept in the dark for 120 minutes (green) and were then transferred back to the light (dark cyan). For reference, roGFP OxD following the same H2O2 treatment but without transition to the dark (black, ~130 min post treatment) and following maximum oxidation (200 µM H2O2, red) and maximum reduction (2 mM DTT, blue) are shown. (E) The fraction of dead cells 24 h post H2O2 treatment with or without transition to the dark ("dark" and "light" respectively) as measured by positive Sytox staining in chl-roGFP strain. Data is shown as mean ± SEM, n=3 biological repeats.      Fig. S17. AF leakage into i488 channel in roGFP strains following H2O2 treatment. Flow cytometry measurements of i488 AF leakage in P. tricornutum strains expressing roGFP targeted to the chloroplast (chl-roGFP, A), nucleus (nuc-roGFP, B) and mitochondria (mit-roGFP, C) over time following treatments of 0-200 µM H2O2 or 2 mM DTT. Leakage into i488 channel was calculated by the mean i488 of the WT strain (n=3 biological repeats) divided by the mean i488 of the roGFP expressing strain (n=3 biological repeats) at the same time-point following the same treatment.  Table S1. Growth phase effects on roGFP fluorescence intensity. Flow cytometry measurements of chl-roGFP and WT (no roGFP) strains at different growth phases (see Fig. 6 A). Cell concentration (Cells·ml -1 ) is as measured before cell concentration adjustments. Dynamic range (Rox/Rred) was calculated by dividing the mean roGFP ratio under maximum oxidation conditions (200 µM H2O2) by that under maximum reduction conditions (2 mM DTT). All other measurements were performed on untreated cultures. Auto-fluorescence (AF) leakage into i488 and i405 channels were calculated by the mean intensity of WT strain divided by the mean intensity of chl-roGFP strain in the i488 and i405 channels