The Interplay of Light and Oxygen in the Reactive Oxygen Stress Response of Chlamydomonas reinhardtii Dissected by Quantitative Mass Spectrometry*

Light and oxygen are factors that are very much entangled in the reactive oxygen species (ROS) stress response network in plants, algae and cyanobacteria. The first obligatory step in understanding the ROS network is to separate these responses. In this study, a LC-MS/MS based quantitative proteomic approach was used to dissect the responses of Chlamydomonas reinhardtii to ROS, light and oxygen employing an interlinked experimental setup. Application of novel bioinformatics tools allow high quality retention time alignment to be performed on all LC-MS/MS runs increasing confidence in protein quantification, overall sequence coverage and coverage of all treatments measured. Finally advanced hierarchical clustering yielded 30 communities of co-regulated proteins permitting separation of ROS related effects from pure light effects (induction and repression). A community termed redoxII was identified that shows additive effects of light and oxygen with light as the first obligatory step. Another community termed 4-down was identified that shows repression as an effect of light but only in the absence of oxygen indicating ROS regulation, for example, possibly via product feedback inhibition because no ROS damage is occurring. In summary the data demonstrate the importance of separating light, O2 and ROS responses to define marker genes for ROS responses. As revealed in this study, an excellent candidate is DHAR with strong ROS dependent induction profiles.

(HO • ) are not. Nevertheless, molecular O 2 itself oxidizes biomolecules, for example, thiol groups, albeit at a much slower rate. The fundamental change in environment and the appearance of O 2 and ROS triggered the biggest mass extinction ever seen on Earth (2,3). Soon after, the much more efficient O 2 based metabolism (compared with fermentation) lead to an evolutionary explosion (4). Today, cells obtain energy from reduced organic molecules through O 2 based respiration.
In the past ROS were associated with cellular stress but strong evidence points toward a cellular ROS network that keeps ROS production and ROS scavenging in tight balance to ensure the maintenance of the cellular redox homeostasis and protection against ROS stress (5,6). An imbalance in this network has been associated with a wide array of human diseases such as cancer (7), neurodegeneration (8), Keshan disease (9), and many others (see also review (6)), although arguments have been brought forward that the origin of some diseases is not directly linked to ROS and that ROS are more likely to be the result of deteriorating cells (10). In any case, the cellular ROS network response to ROS stress is implicated in the progress of these diseases and understanding the network dynamics will have a significant impact in medicine.
Equally important, reduced ROS capacity or imbalance in the ROS network results in decreased crop yields and simple attempts to increase production yields by increasing ROS scavenging capacities in plants failed because those plants lost their ability to mount a defense against pathogens efficiently by the hypersensitive reaction (11), which implicates intended localized high yield ROS production. On the other hand Chang et al. could show that the knock-out of glutathione peroxidase 7 (gpx7), i.e., reducing ROS scavenging capacity, leads to an increased pathogen resistance but, unfortunately, to an increased photosensitivity as well (12), thus resulting in reduced crop production. The quintessence is that plants require the ability to produce sufficient amounts of ROS as part of their defense mechanism yet require some ROS scavenging capacity because photosynthetic growth inevitably produces damaging ROS. In order to effectively mount a hypersensitivity defense reaction, the ROS scavenging capacities have to be suppressed. Thus understanding the ROS network is an important global issue in the light of hunger in some parts of the world and the need for biofuels. Elucidating the key players of the ROS network will allow high production crop plants to be designed.
It seems clear that the ROS network, its dynamics and homeostasis are poorly understood. Understanding how to evaluate the ROS balance and how to restore ROS balance within a cell would have a strong impact on a medical and agricultural level. To put it in the words of Barry Halliwell: "the likely clinical value of 'antioxidant therapy' will depend on how well the exact role of reactive oxygen species," i.e., the ROS network, "is known" (13). ROS can be divided into two classes, i.e., H 2 O 2 and 1 O 2 based ones. Especially in plants, algae, and cyanobacteria, it is now widely accepted that the signaling pathways of H 2 O 2 (14) and 1 O 2 (15) are complex and entangled (16,17) simply through the nature of their production, i.e., via an active photosynthetic electron transport chain. However, there have been reports that clearly show the independence of H 2 O 2 and 1 O 2 mediated responses (see e.g. (18,19)). In Arabidopsis thaliana the ROS network, in particular the 1 O 2 aspect has been widely studied, but comprehensive proteomic studies are still required. The A. thaliana flu mutant was used to reveal 1 O 2 related retrograde signaling. The flu mutant accumulates protochlorophyllide when grown in the dark, and seedlings bleach and die whereas mature plants stop growing when transferred into light (20). 1 O 2 production yielded an induction of distinct genes and these differed significantly from genes induced by H 2 O 2 (15). Apel and co-workers identified the chloroplast localized EXECUTER1/2 proteins as key players in 1 O 2 retrograde signaling (18,21), highlighting that specific 1 O 2 induced signals trigger programmed cell death (PCD) rather than ROS induced damage. A flu-like gene (flp) was identified in Chlamydomonas reinhardtii, and its gene product FLP in its two splicing variants was shown to be involved in the chlorophyll biosynthesis (22). Regulation of FLPs were suggested to occur via light and retrograde plastid signals (22). The specific 1 O 2 signaling mechanism in A. thaliana was further extended by Ramel et al. (23). The authors could show that 1 O 2 induced damage to ␤-carotene, a major component in a ROS defense strategy, yields ␤-cyclocitral, which when produced and applied exogenously triggers a selective 1 O 2 response, similar to the one reported by Apel and co-workers when describing the effects of the flu mutant (15,18,21). However, the signaling pathways involving EXECUTER and ␤-cyclocitral show more and more independent features (see e.g. Lundquist et al. (24)).
ROS production is an inevitable part of the oxygenic photosynthesis and thus can be controlled noninvasively by light intensities. This is why plants, algae, and cyanobacteria offer a unique opportunity to investigate the ROS network. However, in plants the majority of ROS is produced in the chloroplast requiring O 2 as educt and the presence of light. Therefore, careful separation of the light, O 2 , and ROS responses is required. As a consequence, simple high light/low light comparisons are overshadowed by additional ROS production, and vice versa. A classical example is HSP70A in C. reinhardtii, which was originally reported to be light regulated (25) and later proven to be regulated by ROS (19), via two promoters that react specifically on H 2 O 2 and 1 O 2 , to be precise.
We have devised an experimental setup, which allows the ROS, high light/low light (HL/LL) and aerobic/anaerobic (AR/ AN) responses to be dissected on a proteome level using metabolic labeling and quantitative proteomics. We used an interlinked experimental setup that connects all four possible treatments in such a way that each treatment is compared with two other treatments. This offers a strong internal control because the changes in protein levels comparing two not directly connected treatments can be measured by two independent estimates. MS data was analyzed employing high quality retention time alignment to increase overall confidence in protein quantification, increase protein sequence coverage and increase coverage of all conditions. PyGCluster, a novel hierarchical clustering approach (26) was used to identify communities of proteins that are coregulated. Five communities/expression profiles are discussed: a) light and O 2 dependent induction, i.e., potential ROS related regulations, b) a novel regulation type, which shows induction of protein expression influenced additively by light and O 2 , but with light as the obligatory first step, c) light related induction (O 2 independent), d) light dependent repression (O 2 independent), and e) light dependent repression in the absence of O 2 , which might be a regulation linked to feedback inhibition by for example, molecules that are normally damaged by ROS.

Strains, Growth Conditions, and Biological
Repetitions-The wallless CW15 strain of Chlamydomonas reinhardtii cell was used in all experiments, hereafter referred to as the wild type. Cells were grown heterotrophically in tris-acetate-phosphate (TAP) medium (27) at a light intensity of 20 E/m 2 /s and at 22°C on a rotary shaker at 120 rpm or on TAP-agar plates containing 1.5% agar at a light intensity of 40 -50 E/m 2 /s. Cells were metabolically labeled by growing on TAP medium with 15 NH 4 Cl (Cambridge Isotope Laboratories, Inc., Tewksbury, MA, USA). Cells were grown constantly on 15 N containing agar plates to assure complete metabolic labeling. Overall labeling efficiency was analyzed using pyQms (supplemental Fig. S1, Niehues et al., manuscript in prep.).
The experiment was conducted three times, additionally technical repetitions were analyzed for two of the repetitions and repetition #2 was conducted as a complete label swap experiment (supplemental Table S1).
Cell Treatment-Cells were grown to a density of 2-6 ϫ 10 6 cells/ml and kept in exponential growth phase for 5 days at a light intensity of 20 E/m 2 /s. For each experiment, chlorophyll concentrations of cells were determined according to Porra et al. (28). Prior to treatments, cells were diluted to a chlorophyll concentration of 2.5 g chlorophyll/ml and subsequently shifted to photoautotrophical growth conditions in minimal medium (HSM, (27)). Cell densities were between 0.5-0.9 ϫ 10 6 cells/ml. Cells were incubated for 1 h at a light intensity of 20 E/m 2 /s for recovery prior to treatments. The unlabeled ( 14 N) and metabolically labeled ( 15 N) cultures were then divided in two, yielding four cultures that were treated as illustrated in Fig. 1. Labeled and unlabeled cultures were transferred to high light (HL; 180 -200 E/m 2 /s) or low light (LL; 15-20 E/m 2 /s) and either constantly purged with an aerobic atmosphere (AR; N 2 , 20% O 2 , 500 ppm CO 2 ) or with an anaerobic atmosphere (AN; N 2 , 500 ppm CO 2 ). Gases were mixed using Wö sthoff pumps (Wö sthoff, Rietberg, Germany). O 2 concentrations were measured regularly during the experiment using a "Dissolved Oxygen Meter" oxi3310 (WTW, Weilheim, Germany).
Cell Harvest-Cells were pelleted at 2500 ϫ g (Beckmann Coulter J 20 XP), resuspended in H6 buffer (5 mM HEPES, pH 7.5, 10 mM EDTA) with inhibitors of protein biosynthesis (1 mM lincomycin and 28 M cycloheximide), a proton gradient disruptor (10 M nigericin), and DNase. If required, samples were frozen in liquid nitrogen and stored at Ϫ80°C for further analysis.
Protein Concentration Determination, SDS-PAGE, and Immunodetection-Protein concentrations were determined using the bicinchoninic acid (BCA) based Thermo Pierce® Protein assay kit (Thermo Scientific, Perbio Science GmbH, Bonn, Germany). SDS-PAGE was performed using a modified protocol of Laemmli (29) using a total of 40 g of protein for each lane (denaturation for 15 min at 65°C). Immunoblots were carried out by semidry electroblotting of proteins to nitrocellulose membranes (GE-Healthcare, Little Chalfont, United Kingdom). Antibodies directed against LHCSR3 (1:1000) (30), CF1 (1:10000) (31) were used for the detection of proteins with enhanced chemiluminescence (ECL). Immunoblots were quantified using Im-ageJ (32). Ratios between the signals of the treatments were averaged for the three biological replicates.
Fluorescence Yield Measurements-A Maxi-Imaging PAM chlorophyll fluorometer (Heinz Walz, Effeltrich, Germany) was used to measure chlorophyll fluorescence yields. After treatment, cells were dark adapted for 30 min before measurement. Cell densities were between 0.6 and 0.9 ϫ 10 6 cells/ml. Maximal fluorescence (F Max ) was measured during 40 s dark incubation prior to measurement, i.e., before the illumination phase of 4 min at 800 E/m 2 /s followed by a final dark phase of 5.5 min. Nonphotochemical quenching (NPQ, (33)) was calculated as (F max -F m Ј)/F m Ј. F m Ј was determined during the illumination phase. All values measured were normalized by the ground fluorescence F o Ј. Nomenclatures of parameters were used as described in Van Kooten and Snel (34).
Proteomics Sample Preparation-Cells from 5 h treatments 1-4 ( Fig. 1) were mixed on equal protein content (200 g protein) and lysed using a 100 mM Tris/HCl buffer containing 2 or 4% SDS and 100 mM DTT. Protein samples were tryptically digested using a modified filter-aided sample preparation (FASP) protocol (35,36) as described in the following section. Cell lysates were concentrated in centrifugal filter devices (Millipore, Amicon Ultra-0.5, 30 kDa molecular weight cutoff (EMD Millipore Corporation, Billerica, MA) and diluted using a buffer containing 8 M urea. This had no visible consequence on the size of the proteins identified (supplemental Fig. S2). Washing steps with 8 M urea were repeated 5-6 times. Carbamidomethylation of cysteines was omitted as it introduces a reduction in 15 N labeling efficiency and thus an inaccurate estimation of 15 N labeled peptide amounts. After several 50 mM ammonium bicarbonate (NH 4 HCO 3 ) washing steps, digestion of proteins was performed using Trypsin (Promega sequencing grade Trypsin, 0.05 g/l in 50 mM NH 4 HCO 3 , enzyme to protein sample ratio 1:100) for a duration of 4 to 18 h (overnight) and peptides were segregated through a molecular sieve via centrifugation. The Post-FASP (35,36) method was used for peptide mixture fractionation yielding six pH fractions (pH 3, 4, 5, 6, 8, 11, and flow through). Post-FASP was performed with self-packed anion exchange tips. For this purpose 200 l pipette tips were packed with six layers of 3 M Empore disks anion exchange material (3 M, St. Paul, MN, USA). Peptide samples were suspended or diluted in Britton and Robinson universal buffer (BRUB) at pH11 (37), loaded onto the equilibrated anion exchange tips (methanol, 1 M NaOH, two times BRUB at pH 11) and subsequently eluted with BRUB of pH 11-3 (pH 3 containing additionally 0.25 M NaCl). Finally peptides were dried in a vacuum centrifuge and stored at Ϫ20°C.
Database Generation-A target-decoy database was generated using the C. reinhardtii genome annotation with protein model prediction performed by the Joint Genome Institute (JGI) version 4.3 and Augustus 10.2 (38,39). This database was merged with the translated C. reinhardtii NCBI databases of the complete chloroplast genome (BK000554.2, (40)) and the complete mitochondrial genome (NC_001638.1), yielding a database with 17195 nonredundant protein entries. Additionally, a set of common contaminants including keratin sequences were included from the cRAP database V1.0 (http:// www.thegpm.org/crap/) and other sources. Decoy sequences were generated by tryptic peptide amino acid shuffling (length over 8 amino acids) or all possible permutations (length less or equal to eight amino acids). Arginine and lysine positions were fixed to ensure a) an equal length distribution of the target and decoy peptides and b) the same number of target and decoy peptides. Immutable tryptic peptides (e.g. AAR) were not altered, because they are generally not identified using standard LC-MS/MS shotgun proteomics approaches.
Peptide Identification-Peptides were identified using the algorithms OMSSA (version 2.19, (41)), X! Tandem (version 2011.12.01, (42)), and SEQUEST (Bioworks software package version 3.3.1 SP1; Thermo Scientific, Bremen, Deutschland, (43)). OMSSA and X! Tandem were executed using Proteomatic (44). Default values were used for most of the search parameters. Precursor mass accuracy was set to 5 ppm, CID fragmentation ion accuracy was set to 0.5 Da, whereas HCD fragmentation ion accuracy was set to 0.04 Da for OMSSA and X! Tandem and 0.05 Da for SEQUEST, because 0.04 Da was not available. Further search parameters were variable modifications: oxidation of methionine, histidine, and tryptophan (ϩ15.9949 Da) as well as double oxidation of tryptophan (ϩ32.9898 Da). These modifications are inferred from ROS induced damage to synthetic peptides (Weber and Fufezan, unpublished results). Two missed cleavage sites were allowed. Phosphorylation as variable modification was allowed for serine, threonine, and tyrosine using default values. Additionally, X! Tandem considers acetylation of the N terminus (ϩ42.0106 Da), loss of water (-18.0106 Da), and desamidation (-17.0265 Da). Peptidespectrum-matches (PSMs) were equally conducted for 15 N-labeled peptides using default 15 N settings and a customized, high accuracy 15 N amino acid masses XML for X! Tandem (supplemental Table S2). Posterior error probability (PEP) and q-value were determined for each PSM using qvality (45). The application of PEP and q-Value allowed a qualitative comparison of the PSMs between the different search engines. All PSMs with PEPs greater than 0.05 were discarded. LC-MS/MS data in RAW format (Thermo scientific) was converted to the open mzML format (46) using msconvert (part of Proteowizard, version 2.0.1905, (47)). Using pymzML (48), the mzML files were subsequently converted to the mascot generic format (mgf) as input for OMSSA and X! Tandem. Alternatively RAW files were directly converted to the dta format required as input for SEQUEST using the Bioworks package. All mzML files are accessible at peptideatlas.org with the database tag "Barth_2013".
Retention Time Alignment-PSMs and quantification data were stored in piqDB, a laboratory internal database based on mongoDB (http://www.mongodb.org/), which is similar to SuperHirn (49). All further analyses, LC-MS/MS alignments and enhancements were performed using piqDB and the Python scripting language. The alignment of all LC-MS/MS runs was computed based on the cardinal spline interpolation technique. After all runs were aligned to a golden master representing a merge of all runs, specific peptide retention time (RT) windows were defined. This is similar to what is known as an accurate mass time tag used by SuperHirn (49), except piqDB uses a combination of peptide specific identifications (of multiple search engines) and quantifications (isotopologue matches) qualities. LC-MS/MS alignment quality can thus be judged by the length of the RT windows for all peptides in all LC-MS/MS runs. A RT window is defined by the difference between the latest (in time) and the earliest observation/identification (PEP Ͻ ϭ 0.05) for a given peptide over all LC-MS/MS runs. Supplemental Fig. S3 shows a density plot for all RT windows before (red line) and after alignment (blue line) for all 6037 peptide windows in all 99 LC-MS/MS runs. The alignment quality is evidenced in the overall shorting of RT window lengths (supplemental Fig. S3).
Protein Ratios-Quantifications were conducted using pyQms with default values (Niehues et al., manuscript in prep.) in combination with pymzML, similar to Hö hner et al. (50). Isotopologues were calculated for each peptide in both unlabeled and 15 N labeled molecular compositions and in charge states 2-5. These isotopologues were then matched onto all MS1 spectra yielding matching scores (M-scores) between 0 and 20. A M-Score of 20 describes a perfect match of measured MS1 features to the calculated isotopologues of a given peptide in a given charge state and label efficiency with respect to its predicted m/z and intensity values. The amount was then calculated by the sum of all intensities of the scaled theoretical isotopologue. Scaling was based on the matched measured peaks, weighted by their intensity. MS1 spectra, in which a pair of the light ( 14 N) and heavy ( 15 N) sister peptides could be matched were taken into account for the peptide ratio calculation. Finally, the peptide ratio was determined by the sum of all light amounts divided by the sum of all heavy amounts, whereas the individual MS1 scans were weighted by their combined (light and heavy) M-Scores. This ensured that MS1 spectra at the beginning of the elution peak are weighted less than MS1 spectra at the height of the elution peak. Protein ratios were based on a) proteotypic peptides only or b) on nonproteotypic peptides, which were combined into protein groups, indicated by "_" in their Fasta annotation lines. Ratios for proteins and protein groups were calculated based on the peptide ratios. Each peptide was weighted by the number of MS1 scans that contributed originally to its ratio, hence abundant peptides with multiple MS1 quantification events contribute more to the protein ratio than single events. Protein ratios and standard deviations of all proteins in all experiments (log 2 ) are shown in supplemental Table S3.
Overall the experiment was conducted three times including one label swap yielding three biological replicates. Correlation plots between the protein ratios obtained in the different biological replicates show very good reproducibility (supplemental Fig. S4).
Hierarchical Clustering of Proteomics Data-The combined dataset was subjected to a clustering analysis using pyGCluster (26) similar to Hö hner et al. (50). Briefly, pyGCluster injects noise into the dataset and uses those large number of modified datasets to perform a large number of agglomerative hierarchical clustering. Noise is injected into the datasets by re-sampling each protein ratio from the normal distribution given by its mean and standard deviation value. This allows highly reproducible clusters to be identified. The clusters and their identity are inspected and counted for each iteration. Accordingly, clusters with high counts are highly reproducible even when considering the noise during the clustering. pyGCluster depends on the Python module fastcluster (51), SciPy (http://www.scipy.org/) and NumPy (http://www.numpy.org/). For more information please refer to Jaeger et al. (26). Eleven distance-linkage combinations were evaluated: correlation distance with average, complete, single and weighted linkage, as well as Euclidean distance with average, complete, centroid, median, single, ward, and weighted linkage.
After 1E6 resampling iterations, clusters with a minimum count of four and frequency of 1% in any of the 11 assays were combined into communities. A community is an association of related clusters, i.e., overlapping clusters. Communities are determined in pyGCluster by a metaclustering approach, i.e., iteratively applying agglomerative hierarchical clustering using a customized similarity measurement and complete linkage. Briefly, the overlap and the frequency of clusters are taken into account to describe their distance, i.e., the similarity. During each step, clusters (in the first step, or communities in the other steps) sharing a minimum overlap among each other are combined into a community.

RESULTS
In oxygenic photosynthetic organisms, ROS production requires O 2 and is mainly driven by light. Several experiments have shown that reduced levels of O 2 rescue light sensitive mutants (61), thus light intensity and O 2 concentration are interwoven parameters in the ROS stress response. Separating those parameters is not trivial, thus an experimental setup was devised which allows the ROS, oxygen (O 2 ) and light responses to be dissected using quantitative proteomics, high quality retention time alignment, and extensive clustering analysis.
Design of the Experimental Setup-Chlamydomonas reinhardtii CW15 cells were kept in photoheterotrophic growth conditions for 1 week in exponential phase. Cells were suspended in minimal medium, i.e., switched to photoautotrophic growth (HSM) prior to treatment, and kept for 1 h under low light (LL) to ease the stress induced by the medium change. Metabolically labeled ( 15 N) and unlabeled cell cultures of the WT were each divided in two cultures. The four cultures were then treated for 5 h as shown in Fig. 1 with high light (HL, 200 E/m Ϫ2 /s Ϫ1 ) or low light (LL, 20 -30 E/m Ϫ2 /s Ϫ1 ) in combination with normal aerobic atmosphere (AR, 80% N 2 , 20% O 2 , 500 ppm CO 2 ) or anaerobic atmosphere (AN, 99.9995% N 2 , 500 ppm CO 2 ). The combinations yielded four treatments: treatment 1 (LL, AN), treatment 2 (HL, AR), treatment 3 (LL, AR) and treatment 4 (HL, AN). Although O 2 is constantly produced as a side product of Photosystem II (PSII), constant purging reduced the O 2 partial pressure to 25 M and 16 M for treatments 4 and 1, respectively.
By metabolically labeling two opposite samples with 15 N (red stars), the samples could be mixed in four combinations, i.e., conditions 2/3, 2/4, 3/1 and 4/1 on equal protein amounts. Mixing the samples on equal total protein amounts allows relative protein amounts between all four treatments to be quantified. In this setup, each treatment is compared with two other treatments and thereby creating a closed circle (supplemental Fig. S5A). We name such a setup an interlinked setup. An interlinked experimental setup has three advantages: i) Quality Assurance-Two treatments that have not been compared directly can be compared indirectly via two independent measurements. In the experimental setup presented here, comparing 2 and 1 can be performed via 2/3 ϩ 3/1 (path A) and 2/4 ϩ 4/1 (path B), thus ratio log2 2/3 ϩ ratio log2 3/1 ϭ ratio log2 2/4 ϩ ratio log2 4/1. A path difference (path A -path B) close to zero is therefore a criteria for high quality data (supplemental Fig. S5).
ii) Missing Ratios-The path difference criteria allows one missing ratio and its standard deviation to be estimated conservatively, i.e., including the maximal possible standard deviation for worst-case scenario (see experimental procedures).
iii) Factor Separation-Regulations induced purely by light or O 2 can be separated from the overlapping combined effects, here for example, ROS. Comparing the light response (i.e. HL versus LL treated cells) in an AR atmosphere (2/3) with the light response in an AN atmosphere (4/1) allows the closely related HL and ROS stress responses to be dissected. Briefly, proteins that show the same regulation in 2/3 as in 4/1 are truly light regulated and not triggered by any kind of ROS. Differences between 2/3 and 4/1 would indicate a regulation related to ROS or O 2 . Equally, comparing the AR/AN response under HL (2/4) with the AR/AN response under LL (3/1) will allow the closely related AR/AN and the ROS stress responses to be dissected.
NPQ Increase Correlates With Stress Induced LHCSR3 Protein Expression- Fig. 2A shows the fluorescence yields obtained from treatments 1-4 after 5 h. The samples exposed to HL (2 and 4) show distinct fluorescence yield transients compared with the samples treated under LL (3 and 1). The highest fluorescence yield is observed in treatments 3 and 1, whereas treatments 2 and 4 show very strong nonphotochemical quenching (NPQ, indicated as double arrow), i.e., (F m -F m ')/F m ', where F m ' is the maximum fluorescence measured (illustrated as vertical dotted lines in Fig. 2A) during the light phase (white bars in Fig. 2A). The traces of treatments 2 and 4 differ slightly in that the fluorescence yield decreases slowly over the 4 min illumination period after treatment 2, whereas the fluorescence yield is more or less stable after treatment 4. The traces of treatments 3 and 1 show very similar features and very low amounts of NPQ, i.e., high fluorescence yields.
The major key player in the NPQ mechanism in C. reinhardtii has been shown to be the light harvesting complex stress related protein 3 (LHCSR3) (62) and as such it was used to estimate the impact of the treatments and to verify the origin of the difference in NPQ. Fig. 2B shows the immunoblot analysis using specific anti-LHCSR3 antibodies and anti-CF1 as loading control. Shown are the starting cultures (labeled and unlabeled) grown photoheterotrophically (t ϭ Ϫ1h), cultures after 1 h of photoautotrophic acclimation in low light (t ϭ 0) and the cultures after 5 h of treatment in 1-4 ( Fig. 1). ImageJ analysis of the immunoblots yielded log 2 ratios of 2/3 ϭ 1.99 Ϯ 0.61, 2/4 ϭ 0.44 Ϯ 0.55, 3/1 ϭ 0.29 Ϯ 0.16, and 4/1 ϭ 1.84 Ϯ 0.39. The higher the amount of induced LHCSR3, the stronger NPQ is, thus the protein amounts of LHCSR3 (2 Ͼ 4 Ͼ 3 Ͼ 1) correlate very well with the fluorescence yield measurements ( Fig. 2A).
Enhancement of Proteomics Data Enables Comprehensive Proteome Studies-The treated samples were further analyzed using quantitative proteomics. The samples were mixed on equal protein levels and digested using the FASP protocol (35,36). A total of three biological replicates were measured. Furthermore, for two of those, four technical replicates of at least one FASP fraction were measured. Label swap experiments were performed in order to detect isotope effects. This was performed by swapping the labels for the biological repetition number 2 (supplemental Table S1). Isotope effects were not detected as visible in the protein ratio correlation between the two labeled swapped biological repetitions (supplemental Fig. S4). Peptide spectrum matches (PSM) were conducted using the search engines OMSSA (41), X! Tandem (42) and SEQUEST (43). Posterior error probability (PEP) was calculated using qvality (45) and a PEP Ͻ ϭ 0.05 was chosen as cut off. Fig. 3 shows a Venn diagram representing the overlap of all uniquely identified peptides for all search engines used. In total 65.2% of all identified peptides were identified by at least two search engines (Fig. 3).
RT alignment of all 99 LC-MS/MS runs, data storage and data procession were performed using piqDB (protein identification and quantification database, AG Fufezan, unpublished), a laboratory internal database (see experimental procedures).
Quantification of peptides was performed using pyQms scans in all MS-runs. Only those quantification profiles that show one single isotopologue elution peak in the complete MS-run and these elution peaks were within the predefined RT windows (after RT alignment) were finally incorporated further into the quantitative analysis. Quantifications without identifications in the same LC-MS/MS runs can thus be confidently obtained. As a result, more peptides can be quantified for a given protein in all conditions and as such the overall quality of the protein quantification increases. It is most important to note that any large and/or wrongly assigned RT window, any low quality RT alignment would ultimately lead to an increase in the standard deviation of the protein quantification. Therefore, treating proteomic data in such a way, if not performed properly, causes poorly interpretable results. However, if the LC-MS/MS RT alignment is performed stringently, RT windows allow the quantification of peptides identified in one/several runs to be "transferred" to all runs, what we call "enhancing." Enhancing proteomic data could therefore yield to an overall increase in protein sequence coverage and as a result protein quantifications are based on more peptides, i.e., quantifications become more confident. Supplemental Fig. S6 shows a sequence coverage density plot for all proteins quantified in all runs before (red line) and after enhancement (blue line). Naturally, the unenhanced sequence coverage has its maximum at 0% because 4991 out of 9626 proteins cannot be quantified confidently in all MS-runs classically. The classical approach requires an identification (MS2) within 1 min of the quantification event (defined by the elution profile). This is most of the time not given because of the under sampling of the LC-MS/MS runs (49). The enhanced proteins mostly have a sequence coverage of around 5% and achieve a maximum sequence coverage of 70% (supplemental Fig. S6).
Supplemental Fig. S7 shows the correlation of the relative standard deviations of proteins before (x axis) and after the enhancement procedure (y axis) as a heatmap. Each standard deviation tuple is rounded and counted in a given bin, ultimately defining its heat. Bins with x ϭ Ϫ2 are those proteins that could only be quantified after enhancing the data. This plot highlights that a) the majority of proteins have not changed their standard deviation by the enhancement procedure as illustrated by the bins on the diagonal b) a few proteins show a slightly increased standard deviation after enhancement as illustrated by bins left of the diagonal, and c) proteins that are quantifiable because of the enhancement procedure, i.e., that have not been quantifiable without enhancement, show the same distribution in relative standard deviation as seen in the unenhanced proteins (bins with x ϭ Ϫ2). Thus, the enhancement procedure allowed protein quantifications/ratios to be obtained for an increased number of conditions without a substantial increased in standard deviations. For example, although LHCSR3 was only identified in three out of four MS-samples/conditions in the first biological replicate, the enhancement procedure helped to overcome this problem and the ratios calculated for all four conditions correlate with the protein amounts obtained by the immunoblot (Fig. 2B) (supplemental Table S3).
Enhancing proteomic data to obtain more relative protein ratios, especially over all conditions, is a prerequisite for clustering approaches. The change in condition coverage comparing unenhanced and enhanced data for all proteins is very prominent (supplemental Fig. S8). On the level of the three independent experiments, the unenhanced data shows that the majority of proteins, i.e., 35.7% have a ratio for only one condition (e.g. 2/3) and only 9% of all ratios are calculable for all four conditions. Enhancing the data yields a maximum of 32.4% in three and 28.1% in all four conditions. Most importantly the total number of proteins with at least one calculable ratio is increased from 1667 to 3246. Furthermore, the advantage of the interlinked experimental setup allows one missing ratio to be calculated conservatively via the path difference (see experimental procedures, supplemental Fig. S5). This advantage increases the number of proteins that can be clustered, i.e., having ratios for all four conditions, from 151 (9%) to 600 (36%) in the case of the unenhanced data set and from 912 (28.1%) to 1964 (60.5%) in the case of the enhanced data set. Table I summarizes the results of the enhancement procedure. The enhancement procedure was also successfully applied in Hö hner et al. (50). The summary of all ratios determined over all experiments, accompanied by their sequence coverage and the number of peptides used for the calculation of the protein ratio can be found in supplemental Table S3.
Protein ratios that are based on one peptide are normally discarded for further analysis, thereby introducing a bias toward large proteins, eliminating smaller proteins. Supplemen- tal Fig. S9 shows two heat maps for all proteins with respect to the number of peptides versus sequence coverage before (supplemental Fig. S9A) and after (supplemental Fig. S9B) enhancement. It becomes evident that simply discarding all protein ratios that are based on only one peptide would yield the loss of more than 200 proteins that have a sequence coverage higher than 10%. Thus, in this study another approach was conducted, by which the sequence coverage and number of peptides that define a protein ratio are evaluated together, i.e., protein ratios are only considered for further evaluation if the sequence coverage * number of peptides is larger than four. It is important to note that all peptides included in the analysis have a PEP Ͻ ϭ 0.05, thus following the argumentation of Nesvizhskii and coworkers (63), those proteins have a probability of at least 95%.
A Novel Clustering Approach Reveals Distinct Regulons-In this study 888 proteins that represent the merge of all three biological replicates were clustered. Merged protein ratios calculation was based on the same calculation rules as protein ratios for a single repetition. Co-regulated proteins were identified using pyGCluster (26). In total 11 distance-linkage combinations (DLC) have been evaluated and 1E6 resampling iterations were performed. Only clusters with a frequency higher than 1% in at least one DLC were considered and grouped into communities. Further information on clustering can be found in the experimental procedures section and detailed information on all clusters found is summarized in supplemental Table S5. For the sake of clarity only five communities are discussed. The heat maps for all communities can be found in supplemental Fig. S11. The color code for all heat maps range from black to green to yellow indicating an up-regulation and from black to red to purple indicating a down-regulation. The uncertainty of the protein ratio in each condition is illustrated by the size of the box in the heat map, whereby the smaller the box the higher the standard deviation (see legend in Fig. 4). Protein names are shown as JGI4 Augustus 10.2 tags, followed by the protein community frequency (obCoFreq), the path difference and a manually annotated protein tag. Additionally, a superscript number after the path difference indicates if and which ratio was added via the path difference criteria.
All communities described above highlight the impact of resolving the effects of light and oxygen separately. It is note worthy that communities can share elements because the metaclustering approach of pyGCluster merges most frequent clusters into communities. Nevertheless, communities shown here were chosen because they showed the highest obCoFreqs. DISCUSSION Oxidative photosynthetic organisms offer the unique opportunity to study the ROS network because they allow ROS production to be triggered noninvasively by light. Here, we used an interlinked experimental setup to elucidate parts of the light, O 2 and ROS networks using quantitative proteomics and bioinformatics. High quality retention time alignment allowed protein sequence and condition coverage to be increased significantly (supplemental Figs. S3, S6 -S9). The expression maps of five communities are discussed, which show distinct oxygen and/or light dependent responses: ROS and light communities are compared with different studies on ROS and light responses in C. reinhardtii (53,55,56) and A. thaliana (15,23,24). (3860 and 13558, Fig. 4) show an increase in protein levels under HL in the presence of O 2 , i.e., treatment 2, which lead to the conclusion that these proteins are regulated by ROS and/or O 2 and are thus directly or indirectly related to the ROS network. The C. reinhardtii annotation for most of the proteins found in those two communities are unknown and their functions can only be deduced from closest BLAST (52) hits to A. thaliana or domain annotations, for example, Pfam (66).

Combined Light and Oxygen Induction-ROS Communities-The ROS communities
Proteins related to ROS defense, i.e., proteins that protect or enhance the assembly of putative ROS production sites or ROS targets are expected to be found in these communities. The well-known major ROS production sites in cells of oxygenic photosynthetic organisms are photosystem II (PSII) and the light harvesting complexes (LHC) (67,68). Thus it comes with no surprise that HCF136 (Cre06.g273700.t1.1), one assembly factor of PSII (69) is part of the ROS community (13558). Damaged LHCs require their cofactors, i.e., chlorophylls and carotenoids to be stored during turnover, i.e., degradation and resynthesis. Such a carotenoid storage function was associated with plastid lipid-associated proteins FIG. 4. Node map and communities obtained by pyGCluster. Selected communities of the clustered proteins are shown (Nodemap illustrated as backdrop; full map can be found in supplemental Fig. S13). The heat map represents the ratios of the proteins ranges from black to green to yellow indicating an up-regulation and from black to red to purple indicating a down-regulation. The standard deviation is represented by the size of the box, the smaller the box, the higher the standard deviation of the protein ratio (see legend). For each protein in the heat maps, the JGI4 Augustus 10.2 tag, community protein frequencies (obCoFreq), the path difference (superscript indicate which ratio was added via path difference calculation) and a manually annotated name for protein description/classification are listed. Only clusters with a frequency higher than 1% in at least one DLC were considered and grouped into communities.
(PAPs) (70) and a PAP like protein (Cre03.g197650.t1.1) was identified in the ROS community. Both proteins support the notion of enhanced LHC/PSII damage and therefore an increased ROS induced protein turnover under treatment 2. Equally, a knowledge-based heat map for both photosystems and their LHCs reveal an overall small induction (ϳ20 to ϳ50%) for all proteins under treatment 2 compared with treatment 4 (supplemental Fig. S12A), i.e., indicating ROS induced damage.
Another member of the ROS community is a homolog to fibrillin 4 from A. thaliana. Fibrillins take part in the formation of plastoglobules, which were shown to be induced under oxidative stress (71,72). Lundquist et al. (24), investigating two plastoglobules located kinases (AtABC1K1/3) could identify AtFIB4 to be up-regulated as a result of light stress, similar to the work of Giacomelli et al. (73). Our proteomics data narrow light stress down to ROS stress in C. reinhardtii, which is in line with the proposed plastoglobule function of prenyl-lipid metabolism, for example, carotenoid recycling and ␣-tocopherol synthesis (74). The induction of GDP-D-mannose pyrophosphorylase (GMP1, VTC1, Cre16.g672800.t1.1) is part of another protective mechanism against ROS. VTC1 was shown to be a member of the ascorbate biosynthesis (75), a major cellular antioxidant (76,77). Thus the induction of this protein observed is in line with reports from Urzica et al. (75), in which the authors showed that ascorbate biosynthesis and ascorbate recycling genes are induced under oxidative stress conditions. The ascorbate recycling protein dehydroascorbate reductase (DHAR) (Cre10.g456750.t1.1) was also found in the ROS community (13558) further supporting the classical theme of ascorbate as a major protective agent against ROS stress.
An unexpected yet interesting aspect of the ROS community is the induction of two Calvin-Benson cycle enzymes, i.e., fructose 1-6 bisphosphate aldolases (FBA1, Cre01.g006950.t1.2 and FBA2, Cre02.g093450.t1.1). A knowledge based cluster containing all 16 Calvin-Benson cycle enzymes that have been observed shows an equal induction of ϳ30% under treatment 2 (supplemental Fig. S12B). Supplemental Fig. S12B shows two additional enzymes that represent branching points of the Calvin-Benson cycle which allow intermediates to be feed into the glycolysis (GAP1a) or the oxidative pentose phosphate pathway (TAL1). Both enzymes show a reduction under treatment 2. This down-regulation prevents the leakage of Calvin-Benson cycle intermediates into those pathways. Several Calvin-Benson cycle enzymes have been shown to be controlled via the ferredoxin/thioredoxin (FD/TRX) mechanism (reviewed in (78)). Activation or deactivation occur via the reduction of disulfide bridges to form two cysteine thiol groups. However, this switch can be deactivated in the presence of O 2 or ROS (79). Given the fact that light intensity and CO 2 concentration are equal under treatments 2 and 4, one could assume that the Calvin-Benson cycle activity, i.e., the flux should match under those two treatments. Finally, this would indicate that the induction of ϳ30% of all Calvin-Benson cycle enzymes under treatment 2 is because of a ϳ30% loss in flux, i.e., because of the inevitable inactivation of those enzymes by O 2 or ROS. Thus the induction observed might help to compensate for this loss. Such a large effect could be an interesting target for further optimization of crop production, if flux analysis would support equal CO 2 fixation rates. Clearly this is speculation and requires a more detailed analysis employing metabolomics.
ROS production is mainly attributed to the chloroplast and as such the finding of several mitochondrial proteins in the ROS community came as a surprise. These proteins are prohibitin (PHB1), proteins of the mitochondrial translocon machinery (TOM40 and TIM17), two manganese superoxide dismutases (MSD1 and MSD2), the 17kDa subunit of complex I (NUO17) and a putative mitochondrial ribosomal protein subunit (MRPS1).
Prohibitin (PHB) was proposed to have multiple functions, yet its least debatable function is as a mitochondrion membrane bound chaperone (80). PHB especially stabilizes membrane bound respiratory electron transport chain proteins and thereby helps to reduce ROS production and thus helps to avoid ROS induced programmed cell death (PCD) (81). Ahn et al. could show that the silencing of PHB results in higher ROS production in Nicotiana benthamina (82) and the plants were more susceptible to oxidative stress induced by H 2 O 2 or paraquat, a chemical that increases O 2 Ϫ• production. The authors proposed that PHB has a role in protection against oxidative stress. In turn, higher PHB levels may increase the tolerable level of ROS in the mitochondria and in the whole cell.
ROS detoxification enzymes, i.e., the manganese superoxide dismutases MSD1 and MSD2 were also found in the ROS community. Both are supposed to be localized in the mitochondrion (83). Surprisingly a chloroplast localized superoxide dismutase (FSD1, Cre10.g436050.t1.1), which was expected to be up-regulated under treatment 2 (83), did not show a ROS induced expression profile, but rather a redox II profile (see below).
PHB and MSDs indicate an increase in ROS production in the mitochondrion or generally increased ROS levels, for which a higher degree of protection is required. Increased ROS levels lead to a higher ROS induced damage to for example, proteins, thus leading to a higher protein turnover and possibly to a restructuring of the proteome in the mitochondrion. The finding of induced TIM17 and TOM40, both essential parts of the inner and outer mitochondrion membrane translocon, supports this theory. If mitochondrial ROS production is increased, a change in the amounts of the mitochondrial electron transport chain proteins is to be expected. However, the overall expression patterns shown by those proteins (supplemental Fig. S12C) do not reflect a ROS induced pattern. Overall, the major ROS production site in plants, algae or cyanobacteria is the chloroplast, thus the results presented here indicate that the induction of mito-chondrial ROS related proteins and compensatory mechanisms are triggered via retrograde signaling from the chloroplast. Alternatively an increased ROS production in the mitochondria as part of the photorespiration (84) or by other means that help to reduce the over reduction of the chloroplast or indirect ROS signaling via intra-organelle signaling could be envisioned. However, an increased mitochondrial based ROS production seems less likely based on the unaffected levels of respiratory enzymes. It still needs to be elucidated whether these effects on the mitochondrion are a) because of an increased ROS production in the mitochondrion itself, for example, because of increased efforts to balance the redox pressure in the chloroplast or b) because of an indirect effect caused by the overall increased ROS production.
Glutathione peroxidase (gpxh) and the glutathione S-transferase (gsts1) were identified as specific marker genes for 1 O 2 in C. reinhardtii through the application of photosensitizers Rose bengal (RB) and neutral red (87,88). Furthermore and most importantly Ledford et al. reported an acclimation of C. reinhardtii cells to 1 O 2 when pretreating them with a low dosage of RB, i.e., a low dosage of 1 O 2 (56). In this study, GSTS1 (Cre16.g688550.t1.1) could not been quantified in all four conditions, yet the ratio observed in 2/4 is around ϳ10% (supplemental Table S3). This is surprising because one would expect to see GSTS1 induced in 2/4, simply because of the difference in O 2 concentration. This discrepancy between transcriptome and proteome approaches could be explained by an increased protein turnover of GSTS1, i.e., strong induction would only be visible on transcript but not on protein level, or that our treatment was less strong than the application of photosensitizers. Similarly, GPX5 (Cre10.g458450.t1.2) was not covered in all conditions and a ratio could only be calculated for 4/1, which showed an induction of ϳ30% (Ϯ 15%) (supplemental Table S3). This would indicate that GPX5 is either already induced by very small amounts of ROS (overall O 2 concentration in treatments 4 and 1 are less than 25 M) or not induced by ROS under the experimental conditions of this study. Furthermore, two specific marker genes for 1 O 2 (AAA-ATPase, At3g28580) and H 2 O 2 (Ferritin, FER1, At5g01600) were described for A. thaliana (15,89,90). In the study presented here, FER1 (Cre09.g387800.t1.1) showed only a weak to no regulation, thus FER1 might not be induced under the HL treatments used in this study, i.e., H 2 O 2 levels may not have been high enough to trigger induction or transcript levels differ from protein levels, for example, because of a higher protein turnover. Several homologs of the A. thaliana AAA-ATPase could be identified in C. reinhardtii via BLASTP (52). The best (Cre24.g768900.t1.1, BCS1, putative Ubiquinol:cytochrome c oxidoreductase, e-value 1E-16) and second best hit (Cre07.g329700.t1.1, RPT2, 26S proteasome regulatory subunit, e-value 1E-9) showed no distinct regulation in the data set presented here. The third best hit (e-value 3E-09) was FTSH1 (AAA-metalloprotease, Cre12.g485800.t1.1). FTSH1 and FTSH2 were reported to be induced by neutral red treatment in C. reinhardtii (91). FTSH proteases are proposed to be involved in PSII repair and degradation (92,93) and in the general degradation of incomplete assembled iron-sulfur proteins in the chloroplasts (94). In the study presented here, FTSH1 (Cre12.g485800.t1.1) and FTSH2 (Cre17.g720050.t1.1) show a very similar regulation, i.e., a strong induction of ϳ60% in 2/3 and a lower induction of ϳ10 to ϳ30% in 4/1 and 2/4, respectively (supplemental Table S3). The highest protein levels are observed under treatment 2, supporting the concept of a participation of both FTSH proteins in the degradation and repair of oxidized proteins. However, the FTSH genes in A. thaliana were found to be induced by light (95), but the proteomics data presented here point to an extended regulation pattern in C. reinhardtii, i.e., a pattern i.e. better described by the redox II community (see below), where light and O 2 have additive effects.
An interaction between FTSH and PHB in the mitochondria of A. thaliana was reported by Piechota et al. (96) and such a functional co-regulation was also partially observed in the proteome study presented here.
Given the observations in the ROS communities, it seems clear that the choice of parameters used in this study induce physiological ROS stress and as a result an increased protein oxidation under treatment 2.
Proteins that are Regulated Additively by Light and Oxygen (redox II )-The general expression profile of members of communities 12016 and 13769 show a strong induction in condition 2/3, whereas the change in 3/1 is around zero and the inductions of 4/1 and 2/4 sum up to the same induction as seen in 2/3. Thus, the effects of light and O 2 are a) additive and b) light is the obligatory first step. It seems reasonable to speculate that the induction of those proteins is controlled via two different mechanisms. For example, the stronger induction in 2/3 compared with 4/1 could be because of an increased protein turnover, i.e., higher amounts of damaged proteins under treatment 2. This would result in protein levels that seem higher under treatment 2 compared with 4, al-though the difference would only reflect the amount of proteins damaged. Both kinds would indistinguishably be observed using a proteomics approach. Alternatively, instead of damaged proteins, simple inactivation because of the presence of O 2 , as discussed above for the Calvin-Benson cycle enzymes (supplemental Fig. S12B), could also lead to the same additive light and O 2 induction pattern.
Four out of 11 members of the redox II communities are proteins that take part in the ROS detoxification, redox regulation/balance and redox signaling (61,64,65), such as GRX1, TRX-x, PRXQ, PRX4, and TRX-o. These proteins and their expression profiles highlight the interwoven O 2 /ROS and redox signal pathways. The induction of those proteins is primarily linked to light and might be because of their involvement in a pre-emptive photo protective mechanism or because of their property as a redox buffer. On the other hand, a higher O 2 concentration under treatment 2 compared with 4 lead to the oxidation of thiol groups that are re-reduced via the TRX, GRX systems, thus O 2 directly modulates the ratio between the redox pairs of glutathione, GRX, TRX and others. It seems tempting to speculate that the stronger induction of those proteins is because of partial inactivation via O 2 or ROS, thus larger amounts of these proteins are required to maintain functionality. As for the Calvin-Benson cycle enzymes, this would be an attractive approach for crop plant optimization.
PRX4 and TRX-o are both localized in the mitochondria and the light dependent up-regulation of mitochondrial enzymes observed implies an active and redox mediated signaling. Furthermore, PRX4 and TRX-o were suggested to act in a complex as part of the detoxification and ROS signaling cascade (64,97), in which oxidized PRX4, if not re-reduced by TRX-o will ultimately lead to ROS signaling. Thus an O 2 independent induction of PRX4 and TRX-o as seen in 4/1, where none to very little mitochondrial respiration and ROS production is expected, points toward an alternative reason. One could speculate that the increase in PRX4 increases the sensitivity of the proposed PRX4/TRX-o mechanism, because it would allow oxidized PRX4 to reach critical amounts earlier and thereby induce a ROS signal. As such, a redox dependent induction of ROS detoxification enzymes suggests that the up-regulation is part of the mitochondrial preparation for ROS stress.
Rubisco activase (RCA, Cre04.g229300.t1.2) is also part of the redox II communities (12016). RCA is an essential protein in the catalytic cycle of Rubisco (Ribulose-1,5-bisphosphate carboxylase/oxygenase) because it lowers the affinity of inhibitors to Rubisco. Some Rubisco inhibitors are produced during normal Rubisco catalysis, i.e., CO 2 fixation. Such an inhibitor is xylulose-1,5-P 2 (XuBP), which can be formed during reprotonation of the activated 2,3-enediol form of ribulose-1.5-bisphosphate (RuBP) (98). Another inhibitor is d-glycero-2,3-diulose-1,5-bisphospahte (PePB), which was found to be the product of nonenzymatic RuBP oxidation in the presence of metals (99). Taken these observations together, one can divide Rubisco inhibitors into two classes, depending on the requirement/involvement of O 2 and thus the amount of Rubisco inhibitors should be higher under treatment 2 than under treatment 4. As a result the additive effect of light and O 2 on the expression levels of RCA as seen in community 12016 seems coherent. Furthermore this type of regulation could either indicate an independent signaling pathway for both classes of inhibitors or alternatively a regulation via inhibited Rubisco in general.
Given the strong exciton pressure and the resulting over reduction of the chloroplast, enriching the substrate for Rubisco, i.e., inducing proteins that increase the CO 2 partial pressure in the chloroplast/pyrenoid seems perceivable. The induction of LHCSR proteins show an alternative counter mechanism against the over reduction of the chloroplast. An increase in the protein amounts of the members of the LHCSR protein families (Cre08.g365900.t1.1, Cre08.g367400.t1.1, and Cre08. g367500.t1.1) as seen in 2/3 (log 2 ϭ 3.4) and 4/1 (log 2 ϭ 2.7 -4.5) increases NPQ (Fig. 2, (62)), i.e., the conversion of absorbed light energy into heat, thus reducing linear electron flow and production of NADPH ϩ H ϩ . Another counter mechanism offers the alcohol dehydrogenase (ADH1, Cre20.g758200.t1.1) in conjunction with the pyruvate decarboxylase (PDC1 2, not found), by converting pyruvate into alcohol, thereby releasing CO 2 and reoxidizing NADPH (103). This mechanism of modulating the redox poise of the chloroplast by reoxidizing NADH via ADH1 was also proposed by Magneschi et al. (103).
The ascorbate peroxidase (APX, Cre02.g087700.t1.1) is also part of the HL community. APX plays a central role in the water-water cycle, which is one of the major pillars of the system that prevents the over reduction of the chloroplast (104). The treatments used in the experimental setup reduce the O 2 concentration in the anaerobic (AN) treatments 4 and 1 to 25 and 16 M, respectively, and are thus above the apparent K m of 7-10 M for PSI mediated photoreduction of O 2 (104). A fully functional water-water cycle is therefore to be expected even under the AN treatments used here. As such it seems legitimate that the water-water cycle is truly controlled via the redox state, because its O 2 reduction effectiveness is very high and the electrons used to reduce O 2 are ultimately extracted from water releasing O 2 . However, it is worth noting that the apparent K m for the oxygenation reaction of Rubisco is around 500 M in Spinach (105,106) and thus, given the conservation of Rubisco between plants and C. reinhardtii, one can assume that photorespiration is barely active under the AN conditions used in this study.
Overall, the induction of CCM related proteins indicate a limitation of CO 2 in our experimental setup. This is surprising because all cultures were purged with 500 ppm CO 2 , which is 25% higher than the natural CO 2 concentration of 400 ppm. Therefore, it seems, that an over reduction of the chloroplast is inevitable and normal in the natural environment. Taken together, C. reinhardtii accepts the potentially higher ROS production that comes with an over reduced chloroplast in order to maintain the highest possible yield in light absorption to carbon fixation. This might reflect the evolutionary adaption to the natural habitat of C. reinhardtii in the soil, where light is limiting.
Light Dependent Down-regulation-Light Repressed Community-The members of the light-repressed community (7049) are proteins that are down regulated in HL, i.e., under treatments 2 and 4. Generally, these proteins show a minor impact of O 2 , but some proteins show stronger down-regulation under treatment 4 compared with treatment 2. This community is hereafter referred to as the light-repressed community.
Two proteins of the light-repressed community (7049) are members of the thiamin biosynthesis pathway (THI4, Cre04. g214150.t1.2 and THIC, Cre05.g240850.t1.1). Thiamin biosynthesis was shown to be regulated via a riboswitch, i.e., product feedback mechanism. The active form of thiamin pyrophosphate (TPP), binds to the mRNAs encoding for THIC and THIA and by reducing the mRNA stability the TPP biosynthesis pathway is down-regulated (107,108).
Conclusively, high thiamin/TPP levels are likely to be present under treatments 2 and 4 compared with treatments 3 and 1. Thiamin/TPP is an obligatory cofactor for different metabolic pathways like glycolysis, oxidative pentose phosphate pathway and tricarboxylic acid (TCA) cycle (reviewed in (109)). The shutdown of the latter two would be required to counteract a) the loss of Calvin-Benson cycle intermediate and b) the generation of additional redox equivalents, both described above as light dependent effects.
The THI4 gene was found to be down regulated twofold comparing C. reinhardtii (strain 4Aϩ) cells that have been exposed to nonlethal concentrations of 1 O 2 to cells without treatment, thus the regulation was described to be 1 O 2 mediated (56). The proteomic data presented here do not correlate with the observation by the transcriptome data of Ledford et al. because THI4 is generally down regulated under HL. Furthermore, THI4 shows a stronger down-regulation under AN treatments, i.e., treatment 4, in which less 1 O 2 is to be expected compared with treatment 2. This points to a more complex regulation scenario where ROS might play a role, yet the major regulatory mechanism involves light.
The variations in the expression profile of this community are shown in FAD7 (Cre01.g038600.t1.1) and GUN4 (Cre05. g246800.t1.1). Both show a more prominent down-regulation under treatment 4 compared with treatment 2 and as a result the 2/4 ratio shows a stronger induction. Such variations in the communities are because of the clustering algorithm pyG-Cluster, which forms communities as a function of overlapping proteins/objects. Nevertheless, both proteins are also part of the 4-down community 13612 (see below) with a higher obCoFreq.
Proteins that are Reduced in Anaerobic High Light Conditions-4-Down Community-The 4-down community (13612) shows a strong reduction of protein levels under treatment 4 (HL/AN treatment), where the cells experience high metabolic turnover/CO 2 fixation with very little ROS production/stress.
Chlorophyll is a major target of ROS and the 4-down community includes four proteins involved in chlorophyll biosynthesis. These are the two subunits of the magnesium chelatase (CHLI1, Cre06.g306300.t1.1 and CHLD, Cre05.g242000.t1.1), the light-dependent protochlorophyllide reductase (POR, Cre01. g015350.t1.1) and a regulatory or enhancing subunit of the Mg-chelatase GUN4 (Cre05.g246800.t1.1) (110). Classically, chlorophyll biosynthesis flux is regulated via negative feedback inhibition on the level of the Mg-chelatase (111). The control of the protein expression is however not fully understood. It requires retrograde signaling from the chloroplast to the nucleus, were those enzymes are encoded. The data presented here shows a down-regulation of chlorophyll biosynthesis enzymes under treatment 4, which is perceivable because chlorophyll is expected to be damaged less compared with treatment 2.
Currently, there is an ongoing debate whether Mg protoporphyrin IX (MgProtoIX) is directly involved in retrograde signaling (112,113) or whether MgProtoIX and its chlorophyll biosynthesis precursors are simply photosensitizers producing ROS (114,115). One marker used to identify MgProtoIX as a retrograde signaling molecule was the derepression of nuclear encoded genes, such as the small subunit of Rubisco (RBCS). Using the experimental setup described here, our data shows an induction of RBCS by ϳ40% under 2/3 and by ϳ15% under 2/4 and 4/1, thus pointing toward a mixed effect of ROS/O 2 dependent and independent factors, similar to community redox II or toward a very sensitive ROS dependence. However, the regulation of the chlorophyll biosynthesis enzymes described here point toward an involvement of ROS via feedback inhibition.
LHCA3 (Cre18.g749750.t1.1), a light harvesting chlorophyll a/b protein of PSI is also part of the 4-down community. Alboresi et al. showed in A. thaliana that LHCA4 (knock out) or LHCA3 (knock down) mutant strains exhibit an increased membrane peroxidation as evidenced by higher malondialdehyde content (116). The authors' conclusion was that LHCA3/4 protect PSI from photoinhibition by acting as a safety valve. This is based on the observation that LHCA3 has "far red" chlorophylls (117), i.e., act as exciton "sinks" in the PSI antenna complex because of an unusual asparagine chlorophyll ligand. This particular asparagine is however absence in the LHCA3 homolog of C. reinhardtii. In fact, Gibasiewicz et al. (118) showed that C. reinhardtii contains a special set of low-energy chlorophylls that are iso-energenic with the primary donor of PSI, yet retain some of the red chlorophyll characteristics. Furthermore the locations of the "far red like" chlorophylls still need to be identified. Thus, the proteomic data presented here show that LHCA3 is down-regulated under treatment 4 and this could be explained by LHCA3 acting as a suicide component of PSI similar to its counterpart in A. thaliana. Under treatment with reduced O 2 levels, such ROS production would be absent thus LHCA3 would not be required to be synthesized at regular high rates, thus this could explain why LHCA3 is down regulated under treatment 4. Additionally, LHCA3 was shown to participate in PSI and LHCI remodeling under iron deficiency (119). N-terminal processing of LHCA3 yields a conformational change, which in turn results in remodeling of LHCI and consecutively reduces excitation energy transfer to PSI (119).
The protein levels of a PSI assembly factor, ycf4 (NCBI 28269761) is also reduced under treatment 4. Low O 2 and ROS levels under treatment 4 lead to a reduced PSI ROS production and thus to a reduced ROS induced damage to PSI. The major ROS targets in PSI are the iron-sulfur clusters (120), thus reduced ROS production leads to a reduced demand for the FeS-assembly proteins. This can also be seen for ycf3 (NCBI 28269760), albeit to a minor extent with a higher standard deviation (supplemental Fig. S12A). Both proteins are required for the assembly of PSI (121). However, the amounts of photosystem subunits are not down regulated under treatment 4 (supplemental Fig. S12A). This reduced requirement of FeS assembly is further supported by the reduction of another iron-sulfur cluster assembly protein (SUFC, Cre07.g339700.t1.1). In Arabidopsis an involvement of SUFC in iron-sulfur cluster repair and preservation in plastids, upon oxidative damage, was proposed (122). Again, this is in line with the reduced requirement of SUFC under treatment 4, because of the reduced oxidative damage of the iron-sulfur cluster.
The 4-down community shows a novel regulation type involving high light and the absence of oxygen, which could only be identified separating oxygen and light as factors in the ROS response. Treatment 4 shows a reduced ROS production under normal high light metabolic turnover, i.e., CO 2 fixation. Such a reduced ROS environment allows proteins that are generally up-regulated to cope with ROS stress to become down regulated, because less ROS damage is experienced. This leads to a reduced requirement of de novo synthesis of, for example, chlorophyll or FeS clusters. Such a regulation, albeit very plausible, has not yet been associated with ROS.
Comparison with ROS and High Light Transcriptomics Studies-Identifying a Marker Gene/Protein for ROS Stress-A direct comparison between different studies is made more difficult by the fact that gene models of organisms are regularly updated and a direct mapping of the gene identifiers, even within one species and its different versions of gene models, is not necessarily straightforward. Here, an extensive BLAST analysis was performed to map the proteins of the ROS and the light induced communities onto other studies that used A. thaliana or C. reinhardtii and DNA array chip techniques, deep sequencing (RNASeq), and quantitative proteomics. All proteins that are part of the ROS or light-induced community described above were blasted against the reference genomes used in each study. Best BLAST hits (cutoff 1E-20) were used to compare the regulation of the proteins observed in the different studies. Overall such a mapping has to be taken with a pinch of salt. The complete list of BLAST hits, the mapping to the compared studies and the corresponding data can be found in supplemental Table S4. Fig. 5 shows the results of the comparison illustrated as a heat map, where columns 1-4 represent the changes in protein levels in conditions 2/3, 2/4, 3/1, and 4/1 (a). Furthermore the comparison is made with studies investigating the ROS network by mutants that increase internal ROS production or exogenously applied ROS and ROS producers.
First, the work of op den Camp et al. (15) comparing the flu mutant against the wild type of A. thaliana after a 0.5 h, 1 h, and 2 h illumination period (shifted from the dark) is shown in columns 6 -8 (b). The flu mutant accumulates precursors of the chlorophyll, i.e., MgProtoIX and protochlorophyllide, because of deregulation of the chlorophyll biosynthesis. As a result the flu mutant, when shifted from the dark into the light, produces large amounts of 1 O 2 (15).
Second, the work of Ramel et al. (23) is shown in column 9, in which A. thaliana was incubated with ␤-cyclocitral and compared with incubation with water (c). The authors identified ␤-cyclocitral as a 1 O 2 damage induced product of ␤-carotene. ␤-carotene properties make it the major photo-protective molecule in the light harvesting complexes of nearly all photosynthetic organisms by drastically reducing triplet chlorophyll exciton life (123) and direct scavenging of 1  Third, the work of Lundquist et al. (24) is shown in columns 10 and 11 (d). The authors investigated two plastoglobule localized kinases in A. thaliana. The wild type and knockdown mutants of those kinases were subjected to high light treatment (520 mol photons m Ϫ2 s Ϫ1 ) for 3 and 5 days. Their quantitative changes on the proteome level are estimated by spectra counts. Shown are the results of the wild type only.
Fourth, the work of Ledford et al. (56) is shown in column 12 (e). The authors demonstrated with a series of experiments that C. reinhardtii (strain 4Aϩ) acclimate to 1 O 2 using exogenously applied photosensitizer RB. The finding was that nonlethal pre-exposure to 1 O 2 triggers a memory effect, which allows C. reinhardtii to withstand higher and normally lethal concentrations of 1 O 2 . Their DNA chip experiments compared the pretreated sample, i.e., exposed to nonlethal 2 M RB to the mock-treated sample, in order to identify the genes/proteins that are involved in the memory effect, i.e., acclimation.
Fifth, the work of Fischer et al. (55) is shown in columns 13 and 14 (f). There, the authors reported the transcriptome analysis of C. reinhardtii cells (strain 4Aϩ) that have been treated with 3 M photosensitizer neutral red (NR) or 100 M tert-butylhydroperoxide (tBOOH).
Finally, the comparison is also made with a study investigating the light response, i.e., the work of Duanmu et al. (53), in which C. reinhardtii cells (strain 4Aϩ) grown on TAP in the dark were compared with cells that were shifted for 0.5 h into high light, i.e., 150 E m Ϫ2 s Ϫ1 (g, column 15).
This overview allows the effects of light and ROS in the different studies to be compared, offers an increased insight into the cellular response to ROS and/or light and shows the strength of the experimental setup used in this study.
The ROS communities (a) show a very coherent picture with respect to DHAR, when compared with the ROS studies performed in A. thaliana (b, c, d) and in C. reinhardtii (e, f, g). The induction of FBA1 as seen in the ROS communities (a) is the opposite in the A. thaliana studies (b, c, d) after 5 days, however similar to (d) after 3 days indicating that FBA1 could be part of the rapid ROS response, which is ultimately down regulated with higher ROS exposure. The C. reinhardtii ROS studies (e, f, g) show a very good correlation with proteins of the ROS communities.
The comparison of the light community (a, lower part) to the light study in C. reinhardtii (g) shows a very consistent picture for all proteins that could be mapped to both studies. The comparison of (a) to the high light study in A. thaliana (d) shows mixed effects. Although some proteins like APX1, CEP, and CLPP4 become strongly induced by the 5 day treatment, others like PRG5 are induced after 3 days, yet become down regulated after 5 days. Again, this might reflect two different scales of ROS adaptation.
Most interestingly is the comparison of the light induced community (a) to the ROS studies performed in the A. thaliana (b, c) and C. reinhardtii (e, f), which highlights for example, the interwoven light and ROS responses.
The study performed on the flu mutant is the obvious candidate for such a comparison because the flu manifests its phenotype only in the presence of light, thus the question arises which proteins are induced because of the 1 O 2 produced and which are just side effects of the illumination. This possibility still remains although op den Camp et al. (15) compared the changes in expressions levels to the wild type exposing it to the same treatment, because flu could perceive light differently. Proteins of the light induced community (a) that could be mapped onto the flu data set show mostly an up-regulation. These proteins are AtOST1, two proteins that contain the domains PF00168 or PF01145, CCP1 and RABF1. Thus their induction might either already be induced by very low amounts of 1 O 2 (as to be expected in treatment 4) or be a pure light effect in C. reinhardtii.
The comparison of the light induced community (a) with the work of Ramel et al. shows a very surprising results, because those authors did not have a light response that overlaid the effects of ␤-cyclocitral, yet their reported induction of AtOST1, three proteins with domains PF00112, PF00168 or PF01145, APX1, CCP1, CEP, GGH2, LCI19, RABF1, and YKT6 have a light induced pendant in C. reinhardtii (a). This could be a) a purely species-specific effect, b) because ␤-cyclocitral is not only a messenger for 1 O 2 stress but also a messenger for HL or c) because ␤-cyclocitral is already formed in the presence of very low O 2 concentrations.
Equally, the work of Ledford et al. does not suffer from an overlaid light effect and furthermore, species-specific effects can be excluded, yet it is important to note that a different strain was used. Nevertheless, the proteins of the light community (a), when mapped onto the data of Ledford et al. (e), show a broad range of expression profiles. A very prominent feature is the strong down-regulation of CAH4 and CCP1, both proteins that are part of the CCM, which indicates that less CO 2 is fixated and as a results the cells might have slowed down in growth. Using strain cw 15 arg 7 mt Ϫ Fischer et al. (91) reported a strong inhibition of cell growth when 1.0 M RB was applied or even cell death when 1.5 M RB was applied (91). It is possible that a slowed down cell division might in fact help to withstand higher concentrations of RB, because of the extend time in which ROS defense mechanisms can be established and DNA damage can be repaired.
The profiles of CEP (Cre09.g407700.t1.1) and a protein containing a PF00112 domain (Cre05.g247800.t1.1), which is most probably a cysteine protease, come as a surprise. Both are mapped to various A. thaliana gene models (AT2G27420, AT1G47128, AT5G43060), and were reported to be down regulated in the flu study (b, AT2G27420, (15)) and up-regulated in the ROS studies of Ramel et al. (c, AT1G47128, AT5G43060, (23)) and in the case of CEP by Ledford et al. (e,(56)), yet both proteins show a pure light regulation in this study of C. reinhardtii. The high light data by Lundquist et al. (d, AT1G47128, (24)) show a reduction of CEP within 3 days and an induction after 5 days, which is in principle showing an early flu like and a late ␤-cyclocitral like expression. This discrepancy, especially between the A. thaliana studies indicates other factors that are participating in the reported responses of flu and/or to ␤-cyclocitral treatment.
Clearly such comparisons are problematic (see above), yet the consistencies found within A. thaliana or C. reinhardtii and comparing A. thaliana and C. reinhardtii (Fig. 5) highlight the importance of dissecting the light, O 2 and ROS responses and help to define accurately marker genes for ROS responses. The best candidate so far is DHAR with strong ROS dependent induction profiles on the transcriptome (b, c, e-g) and proteome level (a, d).

CONCLUSIONS
This study elucidated the ROS response using quantitative proteomics by separating it from pure light and oxygen effects in C. reinhardtii. Moreover novel regulation patterns have been identified, such as the redox II community or the 4-down community. The former shows a regulation that is induced by light and oxygen in an additive manner (with light as the obligatory first step) and the latter shows light induced reduction but only in the absence of oxygen pointing toward a plausible yet novel ROS related regulation. Overall, the results reported in this study show new insights into the ROS network and highlight the advantages and importance of dissecting the ROS, light and oxygen responses in the field of ROS research in oxygenic photosynthetic organisms.