The rise and fall of the photoinhibition-related energy dissipation qI

Photosynthesis converts sunlight into chemical energy, sustaining the vast majority of the biosphere. Photosystem II (PSII), the oxygen-forming enzyme that initiates photosynthesis, is however particularly prone to light-induced damage in a process known as photoinhibition, which limits the productivity of both aquatic and land photosynthesis. Photoinhibition is associated with an energy dissipation process of unknown origin, termed qI. Here, we present a detailed biophysical and biochemical in vivo study of qI in model green alga Chlamydomonas reinhardtii. Time-resolved fluorescence measurements demonstrate the origin of qI, and indicate the PSII reaction centre as the site of the quencher. Oxygen-dependence of quenching site formation, but not photoinhibition itself, is shown, suggesting that two types of PSII damage – donor and acceptor-side impairment – can be separated. We then demonstrate that the quenching loss takes place in the absence of PSII repair, and is mediated by the degradation of photoinhibited PSII cores by the FtsH protease. Finally, we integrate data ranging from picoseconds to hours in the context of structure-function excitation energy-transferring membrane patches, revealing the extent of PSII heterogeneity from the onset of photoinhibition until the breakdown of damaged PSII. Graphical Abstract Highlights Upon photoinhibition, oxygen sensitization results in an irreversible formation of quenching (qI) and inactivation of Photosystem II qI takes place in the PSII reaction centre Photoinhibition-induced D1 cleavage is much slower than qI formation FtsH metalloprotease is required to degrade quenching PSII reaction centres A multiscale energy transfer model describes heterogeneity of PSII during photoinhibition

Graphical abstract: 30 Highlights: • Upon photoinhibition, oxygen sensitization results in an irreversible formation of quenching (q I ) and inactivation of Photosystem II • q I takes place in the PSII reaction centre 35 • Photoinhibition-induced D1 cleavage is much slower than q I formation • FtsH metalloprotease is required to degrade quenching PSII reaction centres • A multiscale energy transfer model describes heterogeneity of PSII during photoinhibition Introduction: Oxygenic photosynthesis supplies virtually the entire biosphere with chemical energy and oxygen. The first 40 steps of this process involve light harvesting and H 2 O oxidation by Photosystem II (PSII), the water:plastoquinone photooxidoreductase. PSII is arranged in supercomplexes in the appressed regions of the thylakoid membranes within plant and algal chloroplasts (Daum et al., 2010;Wietrzynski et al., 2020). In the model green alga Chlamydomonas reinhardtii these multimeric, pigment-protein supercomplexes consist of i) PSII core: composed of the reaction centre (RC) complex (the D1 and D2 proteins, 45 cytochrome b 559 and PsbI), the CP43 and CP47 antennae, the luminal proteins PsbO, P, and Q, stabilising the oxygen-evolving Mn 4 CaO 5 cluster (OEC), and 10 other small transmembrane subunits; ii) the minor antennae CP26 and CP29; and iii) the light-harvesting complex II (LHCII) (Croce and van Amerongen, 2020;Shen et al., 2019). The antennae greatly enhance the absorption cross-section of PSII allowing efficient light harvesting and excitation energy transfer to the reaction centre, where charge separation 50 takes place and electron transfer begins. The antenna complexes also link different RCs, allowing the excitation to travel in the membrane until it is used to drive stable photochemistry (Bennett et al., 2018;Joliot and Joliot, 1964). However, both the charge-separated states and the long-lived chlorophyll excitation are potentially detrimental to PSII due to oxygen sensitization. PSII has adapted to minimise reactive oxygen species (ROS) formation thanks to the heterodimeric RC complex design (Brinkert et al.,55 2016; Johnson et al., 1995;Rutherford et al., 2012) and through auxiliary mechanisms that dissipate a fraction of the excitation (Peers et al., 2009;Roach et al., 2020). The latter principally describes energy quenching, termed q E NPQ (Non-Photochemical Quenching), which in Chlamydomonas is mediated by the LHCSR3 antenna after protonation of its lumen-exposed residues (Bonente et al., 2011;Liguori et Surprisingly, photoinhibition itself is associated with a decrease in chlorophyll fluorescence yield (Krieger et al., 1992;Matsubara and Chow, 2004;Richter et al., 1999;Zavafer et al., 2019). This behaviour is unexpected because in other conditions where PSII is inactive or absent (e.g. in the absence of electron acceptors (Baker, 2008;Kautsky and Hirsch, 1931); in the presence of the herbicide DCMU blocking PSII 85 acceptor side (Lazár, 1999); in the PSII RC knock-out mutant (Wollman et al., 1980); and when only LHCs accumulate in the thylakoids (Dinc et al., 2016)), the fluorescence level is high (average Chl* lifetime of >1.2 ns, e.g. (Tian et al., 2019)), due to the lack of photochemical quenching from chargeseparating PSII. The site of q I , the quenching species, the mechanism, and the role of the energy dissipation related to photoinhibition, are at present unknown. 90 To study photoinhibition and the related quenching, we used an integrated approach in vivo. Multiscale analysis of fluorescence changes from picoseconds to hours provided information about q I and the extent of photodamage of the photosynthetic apparatus. The use of a range of mutants allowed us to pinpoint the location of q I and the mechanism of its formation and loss. Finally, we used membrane-scale modelling of energy excitation transfer to reveal the extent of heterogeneity in PSII populations upon photoinhibition. 95 Results:

Loss of fluorescence under high light treatment
The protocol used throughout the study to investigate the origin of the slowly-reversible chlorophyll fluorescence decrease upon high light (HL) treatment in Chlamydomonas is shown in Fig. 1A. To focus exclusively on the photoinhibition-related effect, the experiments were performed in the absence of q E 100 (using cells not previously exposed to HL), PSII repair (in the presence of lincomycin), and initially in the stt7-9 strain, which is unable to perform State transitions (Depège et al., 2003). Throughout the 90 minutes of HL treatment (1500 μmol photons / m 2 / s), the maximal fluorescence yield of the cells (F M, where the PSII RCs are in closed state) decreased to below 50% of its initial value (Fig. 1B,S1), while the dark-adapted fluorescence value (F 0 , PSII RCs in open state) increased (Fig. 1B). 105 The latter observation indicates that quenching coincides with the closing of PSII RCs upon photodamage since if only quenching occurred, F 0 would also decrease. The F V /F M parameter -often used to quantify the extent of PSII damage -strongly correlates with the decrease in oxygen evolution, confirming photoinhibition of PSII during the HL treatment. Photosystem I (PSI) was instead little affected by the treatment, as demonstrated by the small decrease of the amplitude of photooxidisable P 700 (Fig. S2). 110 Fluorescence quenching is a term that describes an increase in the overall rate of non-radiative excited state decay. It needs to be distinguished from a decrease of fluorescence caused by a decrease in absorption. The observed reduction in F M is on a timescale of < 2 h an energy quenching process, as indicated by the decrease in fluorescence lifetime of the cells (Figs 1C, S3), and the fact that the absorption capacity of the cells decreases only slightly during the HL treatment (Fig. S1). This quenching 115 process is thus hereafter termed q I . Figure 1. Decrease of chlorophyll fluorescence following photoinhibition in vivo is a quenching 120 (A) scheme of the treatment prior to the measurements. (B) quantification of the changes in relevant parameters throughout the HL treatment in the stt7-9 strain. The last timepoint (95 mins) represents the value in samples treated with lincomycin but not exposed to HL. n = 6 ± S.D. (C) Two-dimensional maps of time-resolved fluorescence data of the stt7-9 strain at t = 0, 20 and 90 min of HL treatment, detected with a Streak camera setup. False colouring depicts the intensity (number of photons) in each bin. A representative 125 dataset is shown. (D) Decay-associated spectra (DAS) of the data in (C). Top panel, the sums of spectra integrals at t = 20 and 90 min were normalised to the t = 0 sum. Lifetimes and spectra were free parameters of linked fitting of 3 biological replicas of the experiment (in total 9 images; see analysis of the other replicas in Fig. S15 and fitting quality in Figs S16-S18).

130
Site of the q I quenching There exist several possible quenching sites in the thylakoid membranes of Chlamydomonas. These include (1) PSI (which can act as a quencher of PSII when the two complexes are in close contact (Bag et al., 2020); (2) LHCII (as proposed for q E in vascular plants (Horton et al., 2005); (3) LHCSR (the site of the pH-dependent quenching (Tian et al., 2019); and (4) the PSII core itself (through an unknown 135 mechanism). To identify the q I site, a combination of genetic and spectroscopic approaches was employed. First, time-resolved fluorescence measurements were performed before-(0 min HL) and after 20-and 90 minutes of HL treatment to investigate photoinhibition-dependent changes in spectra and lifetimes of the cells (Fig. 1C). Three decay components were sufficient to describe the fluorescence kinetics at each timepoint. The Decay Associated Spectra (DAS) are shown in Fig 1D.  lifetimes τ 3 = 1.76 ns and τ 2 = 219 ps before photoinhibition are associated with PSII, while the shortest component had spectrum and lifetime (τ 1 = 54 ps) typical of PSI. After 90 min HL treatment, the amplitude and lifetime of the longer PSII component decreased, and those of the shorter PSII component increased. Crucially, no noticeable differences were observed in their spectra (Fig. 1D). The absence of new emitting species after HL treatment, in particular one with a red-shifted spectrum, suggests that LHC 145 aggregation is not the mechanism behind q I (Miloslavina et al., 2008). The PSI spectra and lifetimes were similar before and after HL treatment, indicating that energy spillover from PSII to PSI did not take place during photoinhibition. This conclusion is supported by the similarity of q I amplitude and kinetics in the stt7-9 and ΔPSI strains (Fig. S4).
To verify if q I involves the LHCs via an aggregation-independent mechanism, we examined three strains 150 with reduced antenna content (Fig. S5). In all mutants, the relation between q I and photodamage was similar to reference strains (Fig. S5), supporting the conclusion that the LHCs are not the site of q I . We observed that despite the initial absence of LHCSR3 in the cells, expression of this q E -inducing protein took place during HL treatment (Fig. S5). To verify whether it could contribute to q I , we analysed the stt7-9 npq4 double mutant, where LHCSR3 is knocked out (Peers et al., 2009). In this mutant, the q I 155 amplitude and kinetics remained comparable to the control, excluding the option of LHCSR3-dependent q I (Fig. S5). The results described above suggest that q I occurs within the PSII core. This is supported by the analysis of the ΔPSII mutant, which shows far less quenching and slower induction kinetics than the control and the ΔPSI mutant ( Fig S4). 160 Loss of q I is light-independent and does not require chloroplast translation Next, we investigated the dependency of the quenching from PSII repair by measuring the stability of q I in the long timescale in the absence of PSII repair. We also extended the study beyond the stt7-9 mutant and included three WT strains with distant genetic backgrounds (Gallaher et al., 2015):  1009, and CC-1690. The influence of State transitions on fluorescence signals during photoinhibitory treatment was accounted for (See Fig. S3 for details). Following photoinhibition, all tested strains developed q I . While the maximal amplitude varied between 1 and 1.4, in all strains q I reached a maximum and then strongly decreased (Fig. 2A;see Fig. S3E for the stt7-9 strain data). Surprisingly, this decrease proceeded in the presence of lincomycin, indicating that it 170 was not related to de novo PSII synthesis (Fig. 2B) or to a recovery of the PSII function, as also indicated by the fact that F V /F M did not change during the fluorescence recovery period (Fig. 2C). Furthermore, the loss of quenching occurred already during the HL period. Neither the duration of the HL treatment (after reaching the peak of q I amplitude), nor whether it was followed by a dark period or low-light treatment, had a significant influence on the loss of q I (Fig. S6). Together, these results indicate that a 175 slow, light-independent process which does not require active chloroplast translation governs the q I relaxation. 185 (C) PSII activity in WT (CC-124) during HL treatment followed by a LL period, measured using fluorescence (FV/FM) and O2 evolution capacity. n = 3.

Loss of q I relies on PSII proteolysis by FtsH
We hypothesized that the slow, light-independent q I loss is due to the degradation of PSII core subunits 190 within which q I occurs (Fig. 2B). To test that, we measured q I in ftsh1 mutants, where the major metalloprotease involved in PSII degradation (Kato et al., 2012;Malnoë et al., 2014;Wang et al., 2017) is inactive (ftsh1-1) or present in a low amount (ftsh1-3). The amplitude and kinetics of q I induction in the ftsh1 mutants closely followed those of the reference WT strain (Fig. 3A,S7). However, the mutants remained quenched for a longer time (Fig. 3A, S7) than the controls. This behaviour was ftsh1 -dependent, 195 as demonstrated by the fact that the complemented strains showed q I loss (Fig. 3A,S7).
To be able to correlate the protein degradation with fluorescence changes in the absence of State transitions influence on the emission, we constructed the stt7-9 ftsh1-1 double mutant (Fig. S8). As shown in Fig. 3B, both stt7-9 and stt7-9 ftsh1-1 strains exhibited a loss of full-length D1 protein upon HL treatment, with the initial cleavage being slower in the double mutant (Fig. S9). As observed before, 200 only the stt7-9 ftsh1-1 strain accumulated short fragments of D1 (Fig. 3B), due to the absence of ftshmediated exoproteolytic degradation (Malnoë et al., 2014). In line with these observations, Blue-native (BN) PAGE showed that the smaller PSII supercomplexes decreased far more in stt7-9 than in the double mutant during photoinhibition (Fig. S10). Together, these experiments demonstrate the impaired PSII RC complex proteolysis in the ftsh1 205 background. Overall, the loss of q I showed correlation with the PSII proteins degradation across stt9 and stt7-9 ftsh1-1 strains (as well as in the WT), although it exhibited a lag with regards to proteolysis . These observations indicate that loss of intact PSII core leads to the loss of quenching . Interestingly, the lag in observed quenching loss could potentially be explained in the context of energetic connectivity in the thylakoids as decrease of quenching due to PSII degradation being temporarily 210 limited by presence of quenching side in nearby PSIIs (See section Photoinhibited PSII reaction centres are heterogenous for details). The location of the q I site in the photoinhibited PSII core was finally confirmed by the ΔPSII mutant, in which no loss of q I was observed neither in HL nor in LL, indicating that the slow quenching observed in this strain (Fig. S4) has a different origin, and resembled what was before observed in lacking both 215 photosystems cells exposed to HL (Tian et al., 2015). 225 (B) D1 degradation timecourse in stt7-9 and stt7-9 ftsh1-1 strains, followed by immunoblotting. One representative biological replica is shown; cytochrome f accumulation is not affected by photoinhibition and was used as a loading control. (C) PSII RC proteins loss (average of relative signal from 3 antibodies [α-DE loop and α-C-ter. of D1; α-CP47], see Fig. S9 for raw quantification data), n = 1, and relative qI NPQ, n = 6, in WT (CC-124). (D) PSII RC proteins loss, n = 3, and relative qI NPQ, n = 3, in stt7-9 strain.

Formation of q I site requires oxygen
To understand the sequence of events upon photoinhibition we focused on the initial kinetics of q I and other functional-and biochemical obervables. As shown in Fig. 4A, loss of fluorescence (F M ) was by far 235 the fastest event upon HL treatment (t 1/2 of ~15 min). F V /F M decreased slower (t 1/2 of ~30 min), but as expected it showed good correlation with the decrease of oxygen evolution. Crucially, the degradation of PSII RC proteins was significantly slower than the changes in fluorescence and oxygen evolution (Fig. 4A), in line with previous reports (Aro et al., 1993). These results substantiate the above hypothesis that protein degradation is related to the loss of q I rather than its induction, and highlight that q I is one of the 240 earliest events upon photoinhibition.
The fact that the formation of the quenching site within the core of PSII precedes D1 cleavage leaves several options regarding the quenching mechanism, including non-photochemical energy dissipation and charge separation-based quenching. It was recently shown that protein oxidation events take place early 245 upon photoinhibition (Kale et al., 2017) and that they can induce quenching in pigment-protein complexes (Lingvay et al., 2020); furthermore, it was observed that oxygen influences the F M level during photoinhibition (Gong and Ohad, 1991). We thus investigated whether oxygen is necessary for (1) the quenching itself, or (2) for the formation of the q I site. To test the first hypothesis, we induced anoxia in photoinhibited cells. In this case, the quenching capacity remained unchanged, demonstrating that O 2 is 250 not necessary for energy dissipation (Fig. S11). To test the second hypothesis, photoinhibitory treatment was conducted in anoxia. Addition of DCMU was necessary to prevent O 2 evolution by PSII but it did not affect the loss of fluorescence (Fig. 4B). The q I amplitude was strongly reduced in anoxic conditions, (Fig. 4B) and the PSII closure was slower (Fig. S12), suggesting that ROS-mediated oxidation of a specific pigment within the PSII RC creates the quenching site (Fig. 4B). In conclusion, the oxygen-dependence 255 of q I site formation indicates that quenching arises by the inactivation of PSII RC, likely through an RC chlorophyll oxidation. These observations support an acceptor side photoinhibition-related formation of the q I site.
260 Figure 4. qI site formation is rapid and requires the presence of oxygen All experiments were performed in the presence of lincomycin, which inhibits chloroplast translation, in the stt7-9 strain. Red boxes -HL illumination (1500 μmol photons / m 2 / s); grey boxes -low light period (15 μmol photons / m 2 / s). (A) Changes in photosynthetic parameters and protein accumulation upon photoinhibition. FM, FV/FM, and oxygen evolution changes n = 11; PSII RC proteins accumulation, n = 3. FM was normalised between its maximal and minimal value, and 265 FV/FM to its maximum. (B) Fluorescence quenching in anoxic conditions. n = 3 (DCMU; anoxic, DCMU) and n = 11 (oxic conditions).

Photoinhibited PSII reaction centres are heterogenous
The induction and loss of photoinhibition-related quenching is a complex process that occurs on minutes 270 to hours timescale. During high light exposure, changes in fluorescence parameters, such as F M and F V /F M , occur faster than the degradation of the reaction centre proteins . Notably, the decrease of F M precedes the loss of PSII activity (F V /F M ). Moreover, during the dark period following high light exposure, there is a loss of q I correlating with the D1 preoteolysis after a lag phase and without a corresponding change in F V /F M (Fig. 2,. To understand these observations in the context of a structure-based 275 model of light harvesting , we hypothesized that there are two distinct types of photoinhibited RCs, both unable to perform photochemistry: (1) 'quenching' centres which exhibit q I capacity and (2) 'broken' centres which do not. All together our minimal model contains active RCs that are either 'open' (F 0 ) or 'closed' (F M ), and inactive RCs that are either 'quenching' or 'broken' (Fig. 5D). 280 We built a kinetic model to describe the slow (min-hour) processes that convert reaction centres between the four types ( Fig. 5E). We explain the decrease of F M following high light exposure by assuming there is a light-dependent reaction that converts active RCs to quenching RCs. We assume the rate of q I site formation is equivalent in both the stt7-9 and the double mutant stt7-9 ftsh1-1 (see supplementary discussion for other models). In the absence of PSII repair, the loss of fluorescence quenching suggested 285 a proteolytic degradation of quenching RC complexes (Fig. 3), which we propose forms 'broken' RCs. The ftsh1 mutation decreases the proteolytic activity and thereby reduces the conversion rate of quenching into broken complexes. Thus, our kinetic model has three parameters to describe the time evolution of reaction centre populations: the rate from active to quenching ('q I site formation') and the rate from quenching to broken ('degradation of quenching centres'), which has two possible values depending on 290 the presence/absence of FtsH. A B We connected the evolving reaction centre populations to the fluorescence observables using a structurebased model of excitation-energy transport on a 40000 nm 2 patch of the appressed thylakoid membrane (Amarnath et al., 2016;Bennett et al., 2013Bennett et al., , 2018 (Fig. 5A). Excitation-energy transfer between 295 chlorophyll domains was described using generalized Forster theory (Fig. 5B,C; see methods for details), (Scholes, 2003;Sumi, 1999) and charge separation at active reaction centres ( (Bennett et al., 2013;Caffarri et al., 2009); Fig. 5C) was described using previously established parameters (see Methods and (Amarnath et al., 2016;Bennett et al., 2013) for details). The rate of excitation dissipation (k qI ) at quenching RCs was treated as a free parameter due to the absence of any experimental bounds. For a 300 given population of open, closed, quenching, and broken centres we constructed the ensemble average fluorescence values using twenty realizations of different random assignments of PSII complexes to each group. The simulated ensemble average was then compared to the experimental fluorescence measurements. 305 We constructed the best fit model for photoinhibition by simultaneously minimizing the error compared to the steady-state fluorescence measurements for both the stt7-9 and stt7-9 ftsh1-1 mutants (Fig. 5E). For stt7-9 at 0 min HL, the excitation-energy transfer model successfully reproduces both steady-state F V /F M (sim.: 0.8, exp.: 0.82) and time-resolved fluorescence measurements (Fig. 5N). The stt7-9 ftsh1-1 double mutant at 0 min HL has a substantially shorter F M fluorescence lifetime compared to the stt7-9 310 strain, which we assume arises from previous exposure to the growth light and slow PSII repair (Kato et al., 2012;Malnoë et al., 2014). We fit the short F M fluorescence lifetime to a residual mixture of broken (34%) and quenching (6%) RCs (Fig. 5P), which also reproduces the steady-state decrease in F V /F M value (sim.: 0.65, exp.: 0.67) at 0 min HL.
The best fit kinetic model has a q I quenching rate (k qI ) of 0.1 ps -1 . A single rate of q I site formation (0.027 315 min -1 ) successfully describes the onset of quenching in both strains, and shows a rate of quenching centres degradation that is 2 orders of magnitude times slower in stt7-9 ftsh1-1 compared to stt7-9 .
The model further supports the simultaneous irreversible closing of RCs and q I site formation. Overall, the photoinhibition model proposed here not only correctly simulates the steady-state fluorescence in both strains with a minimal kinetic model, but also reproduces the trends of protein degradation (Fig. 5O, Q). 320 The latter thus suggests that the proteolytic degradation of RC complexes is sufficient and necessary to abolish q I , and that the lag observed in Fig. 3D and E could indeed be simply interpreted as a result of energetic connectivity. We finally confirmed the self-consistency of this model by reproducing fluorescence lifetime measurements after 60 min HL treatment (Fig. 5N,P). 325 Here we have demonstrated that a minimal model containing two types of photoinhibited RCs and three kinetic rates can reproduce both fluorescence measurements reporting on the ultrafast process of light harvesting and immunoblotting data describing protein degradation on the minutes timescale (see Supplementary Discussion and Fig. S19 for more details). Therefore, the model links kinetics spanning 14 orders of magnitude and provides a predictive framework for understanding a broad range of 330 photoinhibition-related events in vivo. Crucially, this model is parsimonious, successfully accounting for the q I formation-and RC closure upon photodamage, without a necessity of evoking two quenching sites and mechanisms proposed earlier (Matsubara and Chow, 2004;Zavafer et al., 2019). 345 (D) Types of RCs used in modelling and their characteristics. An 'open RC' has both an initial reversible charge separation (RP1) and a slower irreversible charge separation (RP2). A 'closed RC' has a slower rate of initial charge separation (RP1) with a faster rate of recombination compared to open RCs and does not have the ability to perform irreversible charge separation. A 'quenching RC' has no rate of charge separation but has a separate non-photochemical quenching mechanism with a rate kqI. A 'broken RC' has no charge separation or quenching pathways.

350
(E) Kinetic model of photoinhibition used for the modelling. The rates were obtained by fitting the data from panels F-K.  It is of little surprise that photoinhibition, damaging one of the most complex enzymes in biology, is itself a complicated process. Understanding PSII damage and repair is nonetheless crucial due to the importance of photoinhibition in crop and aquatic photosynthesis (Ainsworth and Ort, 2010;Chen et al., 2020;Long et al., 1994;Murata et al., 2007). However, the molecular nature of q I has been rarely addressed in the literature (Krieger et al., 1992;Matsubara and Chow, 2004;Richter et al., 1999;Zavafer et al., 2019). 370 In this work we have simultaneously analysed several aspects of this process which will be discussed here to provide a picture of the formation and relaxation of q I (Fig 7).
Connectivity between PSII complexes and its influence on F V /F M Historically, oxygen evolution, spectroscopy, and biochemical measurements were used to quantify PSII 375 damage (Chow and Aro, 2005;Hippler et al., 1998;Tyystjärvi, 2013). In particular, changes in the photochemical yield of PSII assessed through fluorescence measurements (e.g. F V /F M ) are commonly used to quantify photoinhibition. However, the decrease of F V /F M is not linearly related to the accumulation of broken PSIIs, but rather it describes the global maximal photochemistry. This is due to the energetic connectivity, which results in an increase of the antenna size of the active centres by pigments belonging 380 to the photoinhibited complexes (up to 50 nm apart (Amarnath et al., 2016)). Our simulations indicate that formation of damaged centres is faster than the decrease of F V /F M and slower than the apparent F M changes (Fig. 6), in line with previous experiments (Oguchi et al., 2009).

390
Mechanism of q I and potential heterogeneity of photodamage It is now generally accepted that a two-step model best explains the initial PSII photodamage (Hakala et al., 2005;Ohnishi et al., 2005;Tyystjärvi, 2013). First, direct absorption of blue or ultraviolet light by the manganese ions of the OEC damages the donor side of PSII. Then, donor side-impaired PSII trigger changes on their acceptor side. These changes trigger protein oxidation (Kale et al., 2017) and cleavage 395 (Aro et al., 1993). Our results agree with the above model, but indicate that the two steps take place in parallel rather than sequentially, in agreement with a previous proposal (Oguchi et al., 2009(Oguchi et al., , 2011. The oxygen dependence of q I formation (Fig. 4) and the slower F V /F M decrease in anoxia (Fig. S12) suggest that in our experimental conditions the acceptor-side damage is faster than the donor-side photoinhibition, but that 400 in the absence of O 2 the latter still takes place. Crucially, photoinhibition linearly correlates with light intensity ((Tyystjärvi, 2013); Fig. S13) and RC closure takes place simultaneously with the q I site formation (Fig. 5). We therefore consider that the acceptor-side damage involves oxidation of PSII RC pigment(s) through in situ singlet oxygen sensitization proceeding by PSII charge recombination. As demonstrated by Rehman et al. (Rehman et al., 2013), 1 O 2 formation by 3 P 680 (a by-product of charge 405 recombination) linearly correlates with light intensity, in line with this hypothesis (Vass and Cser, 2009). The parallel model implies that the two steps of photoinhibition can be separated. An argument in agreement with that is provided by the anoxic photoinhibition experiment: we observed the appearance of a light-induced quenching of small amplitude (Fig. S11) in samples which undergone HL treatment. This quenching hints at a capacity of P 680 + formation due to the dysfunctional donor side, and thus a partially 410 active acceptor side thanks to the absence of oxygen sensitization. The dependence of photoinhibition and fluorescence parameters on light colour support the same conclusion (Oguchi et al., 2009).
Interestingly, an intrinsic regulation of the rate of 1 O 2 production is built into the PSII, decreasing the rate of PSII charge recombination (Johnson et al., 1995;Rehman et al., 2013). It depends both on the presence 415 of functional donor-and acceptor sides of PSII (Brinkert et al., 2016;Johnson et al., 1995). The midpoint redox potential (E m ) of Q A quinone increases by around 115 mV when OEC is impaired and by 75 mV in the absence of the HCO 3 between Q A and Q B . As a consequence, direct recombination between P 680 is favoured over a repopulation of the P 680 * state. Crucially, such charge recombination could in principle also constitute the q I mechanism -when the donor side of PSII is impaired upon 420 photoinhibition, radiationless recombination would result in a quenching of fluorescence (Krieger et al., 1992). Our experiments, however, show that it is an unlikely scenario. For the recombination mechanism to work, an oxidised Q A quinone is necessary, while anoxia induced post-photoinhibition -which nonphotochemically reduces this cofactor (Mus et al., 2005) -had no influence on the quenching. Conversely, oxygen presence was necessary to form the q I site within the PSII reaction centre. Formation 425 of 1 O 2 is well established upon photoinhibition and ROS have been shown to damage the protein scaffold (Kale et al., 2017). Pigment oxidation also results in a shortening of chlorophyll excited state lifetime (Lingvay et al., 2020). Finally, singlet oxygen concentration increases linearly with light intensity, as does the rate of q I formation (Fig. S13). Thus, a hypothesis where 1 O 2 formed via 3 P 680 directly oxidises one of the RC pigments, impairing charge separation capacity and forming the q I site, is consistent with all the 430 observations.
The kinetic model of photoinhibition (Fig. 5) provides insights into the heterogeneity of PSII RC composition during HL treatment. Importantly, it demonstrates that even in the case of a null F V /F M value, a combination of quenching and broken RCs is present in the thylakoids (Fig. 5L, M). Nonetheless, 435 our model supports a homogenous damage mechanism, with a simple conversion of active RCs to quenchers, followed by their degradation and loss of q I sites (but see also Supplementary discussion and Fig. S19). This finding succeeds in describing notably the kinetically different behaviour between F M and F V /F M . Additionally, the close match between the appearance of broken centres and PSII core subunits degradation suggests that the latter process abolishes the quenching capacity in vivo. 440

Is photoinhibition photoprotective?
Quenching is often associated with photoprotection. In the case of photoinhibition-related quenching, the situation is more complex, as damage is a prerequisite for quenching. The linear dependence of photoinhibition on light intensity and first-order kinetics of PSII function loss (see discussion in 445 (Santabarbara et al., 2002;Tyystjärvi, 2013) and references therein) support the donor-side damage mediated by the direct light absorption by the OEC. This damage cannot be prevented by quenching, but only by photoprotective adaptations such as screens on the leaf surface (Hakala-Yatkin et al., 2010). This interpretation becomes slightly more nuanced in the scope of the two-step photoinhibition model: the donor-side mechanism might be inevitable, but the acceptor-side impairment could potentially be 450 alleviated. However, it is important to stress that F V /F M measurements do not distinguish between these two mechanisms (Fig. S12) -using quenching as the observable might be beneficial for suture studies (Fig. 4). Ultimately, while PSII photodamage studies are invariably done in the absence of PSII repair, in natural conditions limiting ROS production thanks to q I protects the repair machinery (Nishiyama et al., 2006) 455 and alleviates damage in the steady-state (Roach and Krieger-Liszkay, 2019;Roach et al., 2020). As such, q I can positively influence the recovery from photoinhibition, even if not significantly preventing PSII damage (Santabarbara et al., 2001). In particular, PSII assembly provides cues that agree with the need of PSII acceptor-side photoprotection and ROS formation decrease. The abovementioned changes in Q A E m were recently shown to dampen the singlet oxygen production before PSII became fully assembled 460 (Johnson et al., 1995;Zabret et al., 2020).
Crucially, however, by damaging the PSII RC complex photoinhibition prevents the formation of long-lived quenching in LHCII (Fig. S4). The duration of quenching in HL in the absence of PSII core (Tian et al., 2015) suggests that it is difficult to relax this quenching which can thus be detrimental to photosynthetic 465 efficiency following HL exposure (as suggested in the case of too-slowly-relaxing q E (Kromdijk et al., 2016)). This in turn highlights that locating q I in the replaceable PSII RC is another feat of PSII, whose design and function do not cease to surprise (Brinkert et al., 2016;Zabret et al., 2020).
Potential mechanisms of D1 damage 470 Whether offering substantial photoprotection or not, quenching centres are damaged and thus need to be repaired (Baena-González et al., 1999;Chow and Aro, 2005;Hippler et al., 1998;Li et al., 2018), starting with the degradation of the D1 protein. The mechanism of D1 degradation is debated, with at least three known processes involved.
The luminal subunits of PSII (PsbO/P/Q; (Shen et al., 2019)) shield the D1 peptide from proteolysis by DEG. Their dissociation was indeed shown to correlate with D1 cleavage (Eisenberg-Domovich et al., 485 1995;Hundal et al., 1990), and it could constitute the rate-limiting step of degradation. On the other hand, the relatively slow D1 cleavage might be due to the limited capacity of PSII complexes to migrate to the non-appressed regions of the thylakoids. The repair cycle was proposed to take place in the stroma lamellae or grana margins, to account for the accessibility of damaged D1 to the bulky stromal subunits of FtsH (Andersson and Aro, 2001;Chow and Aro, 2005;Hundal et al., 1990;Uthoff and Baumann, 490 2018). In such case, the stt7-9 effect slowing down RC peptides degradation could be due to the impairment in membrane fluidity and movement of damaged PSII RCs. We have demonstrated that the quenching is lost due to proteolysis (Figs 3, 5), and we propose two mechanisms which can account for that. On one hand, initial cleavage of D1 might result in a loss of q I capacity. The microenvironment of the hypothesized oxidised RC pigment could influence q I or the 495 excitation energy transfer to this site. Another possibility is that the RC complexes with cleaved D1 that still remain assembled (at least a fraction of them is shown in Fig. S10) retain the ability to perform quenching, and only complete degradation of the RC core removes the q I site (see Fig. 7, inner polar plot). However, the quantification of the latter process is difficult due to the complexity of solubilisation of membrane complexes, in particular containing cleaved peptides. If furthermore the RC complex 500 degradation rapidly follows the initial cleavage of D1, distinction between these two processes might prove difficult. Finally, while we do observe degradation of CP47, this is likely a result of the absence of de novo D1 synthesis (blocked consistently with lincomycin) and thus their assembly partner (de Vitry et al., 1989). Out of RC proteins upon regular photoinhibition, D1 has by far the fastest turnover (Li et al., 2018;Minai 505 et al., 2006;Ohad et al., 1984;Schnettger et al., 1994), confirming that in many cases, replacement of D1 is the only requirement for PSII repair after photodamage. The experiments in ftsh1 mutant background provide another advantage in our understanding of photoinhibition-related fluorescence decrease. The stability of observed quenching for at least several hours in the absence of degradation (Figs 3, 5) strongly suggests that q I is the sole quenching mechanism 510 observed, as indicated in the initial timepoints by time-resolved fluorescence (Fig. 1). This is preserved despite the changes in the ultrastructure of thylakoids changes upon photoinhibition (Barbato et al., 1992;Hundal et al., 1990).

Outlook 515
In this work, we used integrated approaches to study photoinhibition and the related q I quenching in vivo.
A consistent description of all observables across multiple timescales was achieved using a modelling approach in the context of energetic connectivity. We were able to exclude a number of potential quenching sites thanks to spectrally-and time-resolved fluorescence measurements and the use of multiple mutants. We propose that the q I site is formed by a singlet oxygen-mediated attack on one of the PSII 520 RC pigments, simultaneously closing the RC and forming a quenching site. This process at once decreases the F M fluorescence level and raises F 0 . Quenching RCs are then lost through PSII degradation mediated by FtsH. This unified model does not require separate processes to account for the contrasting kinetics of steady-state fluorescence parameters, nor a heterogeneity in quenchers. Our model of photoinhibition where donor-and acceptor-side damage take place independently, with only the latter forming a quenching 525 site, will help understanding this complicated process from a novel perspective. Initial photodamage to PSII supercomplex (core dimer, C2; and antenna, LHCII) can be described by two mechanisms: the 530 excess-light induced ROS formation and oxidation of PSII subunits and pigments, as well as the direct OEC damage (Tyystjärvi, 2013). qI (inner polar plot of the average fluorescence lifetime in vivo), established with halftime of ~15 minutes, is concomitant with photodamage. As a result of PSII RC complex degradation the quenching is lost and the lifetime of fluorescence increases back to the initial FM state. Following D1 cleavage the PSII monomerises but remains at least partly assembled despite the endoproteolysis of its reaction centre protein, and translocates to non-appressed regions of the 535 thylakoids (or else the appressed regions unstack) (Barbato et al., 1992;Hundal et al., 1990). Transmembrane thylakoid FtsH protease degrades the PSII RC complex. PSII repair with de novo synthesis of at least D1 subunit takes place to reestablish the initial state.

Spectroscopy
Fluorescence and transient absorption measurements were done using the JTS-10 apparatus (BioLogic). In fluorescence mode, detecting white LED flashes were filtered through a narrow interference filter (520 565 nm, 10 nm full width at half-maximum [FWHM], 10 mm) and triggered within short (~300 μs) dark periods of actinic illumination ('dark pulse' mode). The detection was done with a longpass filter (cutoff at 670 nm, 10 mm, Schott). Actinic light was provided from both sides of the cuvette in a custom-build holder, and was set to a subsaturating value of 150 μmol photons / m 2 / s (630 nm peak) to accurately capture decreasing PSII photochemistry in HL. Saturating pulses of 200 ms provided the same actinic 570 LEDs were used throughout (15 mmol photons / m 2 / s). The cells were spun during the measurements with a magnetic stirrer. Steady-state fluorescence parameters were corrected to exclude the PSI contribution (3% at F M ) in data shown in Fig. 5 to account for the absence of PSI in the model membranes.
In absorption mode, P 700 and plastocyanin redox changes were monitored upon 'dark pulse' using detecting 575 flashes passing through a 705 nm (10 nm FWHM) and 730 nm (10 nm FWHM) filters, respectively, and the actinic light was cut from the detectors with Schott RG695 longpass filters. DCMU (10 μM) was added for P 700 experiments. PSI:PSII ratio and PSII antenna size were determined as described previously (Nawrocki et al., 2016Tian et al., 2019). 580 Oxygen emission recordings were done simultaneously with fluorescence measurements using a microoptode (UniSense) stuck inside the culture cuvette. The optode was calibrated in Min medium flushed with air (O 2 saturated) and in the presence of glucose-and glucose oxidase (anoxia). Cell absorption spectra were measured using a Cary 4000 spectrophotometer (Varian) fitted with a Diffuse Reflectance Accessory to account for light scattering. 585 Fluorescence spectra of cells and of BN gel pieces at 77 K were measured with a F-CCDZEN fluorimeter (BeamBio). A LED peaking at 470 nm was used for the excitation.
Time-resolved fluorescence (Fig. 5,S3) was recorded at room temperature with a time-correlated singlephoton counting setup (FluoTime 200,PicoQuant). 10 x 10 mm quartz cuvette was used and the cells 590 were stirred using a magnetic stirrer during the measurements. Excitation was provided by a 438 nm laser working at 10 MHz repetition rate and was set to 30 μW using a neutral density filter. Longpass optical filter was placed in front of the detector to prevent scattered light from reaching the photodiode. Detection was done at 680 ± 8 nm with 4 ps bins. DCMU (10 μM) was added to close PSII (F M state). The instrument response function (IRF) was measured through the decay of pinacyanol iodide in methanol (6 595 ps lifetime; ). IRF obtained was 88 ps FWHM in the same conditions as sample measurements. Data analysis was done with the FluoFit software and the decays were deconvoluted with the IRF and 3 exponential decay functions to yield τ average values. In Fig. 5N and 5P, the PSI-related component (~60 ps) was excluded from the plot to show uniquely the PSII-related contribution to fluorescence decay. 600 Streak camera measurements (Figs 1, S15-S18) (Van Stokkum et al., 2008) at room temperature were done as described previously (Tian et al., 2019). Sample was stirred during the measurements. 400 nm laser excitation was used (Vitesse Duo, Coherent) and the power at the ~50 μm diameter spot was set to 15 μW with 250 MHz repetition rate. DCMU (10 μM) was added to close PSII (F M state). Averages of 100 images (10 s integration each) were background-and shading-corrected. 605 Global analysis of the streak camera images was done as described in (Holzwarth, 1996;van Stokkum, 2018). GloTarAn software was used for the data analysis (Snellenburg et al., 2012), and linked analysis of 3 biological replicas each at 3 timepoints of photoinhibition was performed.

Biochemistry 610
Total protein extracts and thylakoids were isolated as described in (Ramundo et al., 2013). Blue native polyacrylamide gel electrophoresis (BN-PAGE) was performed according to (Chen et al., 2019;Wittig et al., 2006). The resolving gel of BN-PAGE used was 4.1-14% T (w/v), while the stacking gel was 3.6% T (w/v), where T stands for the total concentration of acrylamide and bisacrylamide monomers. Blue native gels were cut into strips and loaded onto an SDS (sodium dodecyl sulfonate) gel for the second 615 dimension (Schägger, 2006). 4% T (w/v) was used for the stacking gel and 12% T (w/v) for the resolving gel. After running, the gels were either stained with a Coomassie blue solution (0.1% Coomassie Blue R250 in 10% acetic acid, 40% methanol, and 50% H 2 O) for 2 hr, and de-stained with a de-staining solution (10% acetic acid, 40% methanol, and 50% H 2 O) for 3 x 2 hr; or used for immunoblotting. 620 Immunoblotting was performed as described in (Nawrocki et al., 2020). 1 μg Chl thylakoid or total protein extract was loaded onto a pre-cast gel (4-12% T (w/v) Bis-Tris Plus, Invitrogen). After electrophoresis, the proteins were transferred to a nitrocellulose membrane. The SDS gel after 2D BN-PAGE was also used for the protein transfer procedure. The membrane was then blocked with 10% (w/v) milk in TTBS (20 mM Tris, pH 7.5, 150mM NaCl, 0.1% Tween 20) for 1 hr and then incubated with different primary 625 antibodies overnight in the dark at 4°C. All antibodies were purchased from Agrisera. The membrane was then washed with TTBS for 4 x 10 min and was incubated with the secondary antibody (goat anti-rabbit IgG) for another 1 hr. The membrane was rewashed with TTBS for 4 x 5 min and was developed for chemiluminescence (with agent SuperSignal West Pico, Thermo Scientific) using a LAS 4000 Image Analyzer (ImageQuant). 630

Modelling
The excitation energy transfer rate matrix was constructed using a membrane model originally developed for A. thaliana (Amarnath et al., 2016). Excitation-energy transfer was simulated using generalized Forster theory combined with a collection of previously described Hamiltonians and protein structures (Amarnath 635 et al., 2016). Open (k cs = 1.56 ps -1 , k rc = 0.00625 ps -1 , k irr = 0.00192 ps -1 ) and closed (k cs = 0.00323 ps -1 , k rc = 0.00218 ps -1 ) reaction centres had the same electron transport rates as reported previously (Amarnath et al., 2016;Bennett et al., 2013). Damaged reaction centres, which are incapable of charge separation, are either 'broken' and have no additional rates of transport, or 'quenching' in which case a new loss channel arises with a rate of k qI . All calculations presented in the main text use k qI = 0.1 ps -1 . To 640 construct a rate matrix for a specified number of open, closed, broken, and quenching reaction centres a random number generator was used to determine which of the 128 reaction centres in the membrane belonged to each type. For F M (F 0 ) simulations, the number of open (closed) reaction centres is taken to be 0. Finally, the time evolution of the excitation population was simulated using 645 dP(t)/dt = KP(t) and a fourth order Runge-Kutta integration method with timesteps 0.2 ps. The excitation population was time evolved to 6 ns where residual chlorophyll excitation was negligible. The steady-state fluorescence observables were computed using the simulated fluorescence yield and then averaged over twenty 650 realizations of RC assignments that satisfy the population distribution determined from the kinetic model. The experiments shown were performed in stt7-9 strain, in the presence of lincomycin, which inhibits chloroplast translation and PSII repair. (A) Absorption difference plots corresponding to P 700 oxidation at various timepoints of photoinhibition. The traces were acquired in the presence of DCMU (to obtain the maximal photooxidisable P 700 quantity), which was added after photoinhibition. Traces obtained at 705 nm were corrected for the 970 plastocyanin signal contribution at 730 nm. Black boxes depict dark period during the protocol; grey boxes -moderate actinic light illumination (150 μmol photons / m 2 / s); red boxes -saturating pulses (15 mmol photons / m 2 / s). (B) Quantification of the changes in maximal photoxidisable P 700 (upon saturating pulse) quantity during photoinhibition. The timepoint at 95' is a control not exposed to HL. 975 To account for the variable rates of the acceptor-side photoinhibition in the PSII antenna mutants (due to lower effective photochemical rate of PSII during illumination), the development of the quenching is shown as a function of decreasing O 2 1035 evolution capacity rather than as a function of time. (B) Immunoblot of selected thylakoid membrane proteins in WT strains and in light harvesting mutants before (-) and after (+) 90 min HL treatment in the presence of lincomycin. The coloured symbols correspond to those in panel (C). Note that Agrisera AS09 408 ⍺-LHCBM antibody used recognizes all major LHCII in Chlamydomonas. 1040 (C) "Antenna size" (maximal light-limited photochemical rate) of PSII (D) PSI:PSII reaction centre ratio measured in vivo .
(E) Accumulation of LHCSR3 during photoinhibition in the stt7-9 strain. CP47 is shown to highlight the ongoing photoinhibition, and the gamma subunit of ATP synthase is shown as a loading control. Grey boxes -low light illumination (15 μmol photons / m 2 / s); red boxes -HL period (1500 μmol photons / 1045 m 2 / s). (F)   Shown experiments were performed in the presence of lincomycin. ftsh1-3(pSL18-FTSH1) is the ftsh1-3 strain complemented with the genomic sequence of ftsh1. The experiments were performed on the stt7-9 strain, in the presence of lincomycin. Grey box -low light illumination (15 μmol photons / m 2 / s); red boxes -HL period (1500 μmol photons / m 2 / s). 1120  (A) and (B). Shaded areas represent the 95% confidence intervals from linear fits of the data forced through the intercept.  Supplementary discussion 1160 Fitting of kinetic parameters The parameters defining the kinetic model (Fig. 5), i.e. the kinetic rates between varying RC types, were fit to reproduce the steady state fluorescence measurements (F 0  was determined at every timepoint for both the stt7-9 mutant and the stt7-9 ftsh1-1 double mutant strains. We used the root-mean squared error across the two strains to describe the quality of fit. We included the F V /F M term in order to bias the fit towards solutions that reproduced the relative F 0 and F M 1170 behaviour. We also added the average standard deviation of each experimental measurement ( E ) to the denominator of the error in order to decrease the sensitivity of the fit in proportion to the uncertainty in the measurement.
The four-parameter kinetic models (described below) showed a rugged error landscape. As a result, we 1175 determined our optimal parameters using the dual annealing minimizer with a gradient descent optimization method implemented in the Scipy optimize library (Virtanen et al., 2020).

Testing Different Kinetic Models
The linear model presented in Fig. 5 succeeds in fitting the experimental data. Other, more elaborate 1180 kinetic models will also reproduce this data. We have tested different kinetic models with four or fewer parameters and found they provide the same or lower quality reproduction of the experimental data. Fig. S19A (left) shows the three-parameter kinetic model used for calculations presented in the main text, denoted as the linear model. In this model, the rate between quenching and broken RCs was allowed to vary between the two strains which we depict schematically as two arrows between those boxes in the 1185 kinetic diagram. We compared the three-parameter model to additional models of higher complexity by calculating the minimal error for each kinetic model across a range of quenching rates (Fig. S19B). In particular, we hypothesized that there could be a light induced mechanism that directly converts active to broken RCs (Fig. S19A, right). We then considered a four-parameter model which, like the linear model, allowed the rate from quenching to broken RCs to vary between the two strains (Model-4A). The linear 1190 model shows a sharply defined minimal error for a quenching rate (k qI ) of 0.1 ps -1 , while model-4A has essentially equivalent error to the linear model for low quenching rates (k qI <0.1 ps -1 ) but returns similarly small errors for a broad range of quenching rates. The lack of any substantial improvement in the overall fit to the fluorescence measurements in Model-4A is notable given the greater flexibility available for fitting. We also tested other four-parameter kinetic models where the rates from active to quenching RCs 1195 and active to broken RCs were allowed to change between the two mutant strains (Model-4B/C, Fig.  S19A right). Strikingly, the overall error for both of these models increased compared to the simpler linear model. This suggests that the flexibility in describing the connection between quenching and broken reaction centres is critical to describing the experimental observations. 1200