Over-reduction of the electron transport chain, which occurs when light absorption exceeds the immediate capacity for CO₂ fixation, can result in electrons “spilling over” to oxygen, leading to the production of Reactive Oxygen Species (ROS). Most directly, this includes singlet excited oxygen (¹O₂), superoxide (O2-), hydrogen peroxide (H₂O₂), and hydroxyl radicals (HO). ROS generation can be triggered in the chloroplast by many environmental factors, including high light, salinity, drought, pathogens, etc. Therefore, plants have evolved a plethora of ROS scavenging mechanisms to minimize the harmful effects of increased ROS levels (Pospíšil, 2012; Foyer, 2018). Although ROS are also generated in other cell compartments, such as mitochondria and peroxisomes, the main sites of ROS production in the chloroplast are photosystem I (PSI) and photosystem II (PSII; Tripathy and Oelmüller, 2012; Pospíšil, 2016).

Oxygen can be reduced to O2- by electrons derived from both photosystems (Zulfugarov et al., 2014; Pospíšil, 2016; Takagi et al., 2016). O2- can be converted to H₂O₂ and O₂, a process that is catalyzed by the enzyme superoxide dismutase (SOD) in the chloroplast stroma (Pospíšil, 2012). H₂O₂ can subsequently be converted to water, in the so called water-water cycle, via a series of redox reactions (Awad et al., 2015)_. Although the scavenging of H₂O₂ can act as an alternative electron sink, the electron flux through this pathway is quickly saturated (Driever and Baker, 2011). Thus, excess H₂O₂ from the chloroplast can pass to the nucleus, where it modulates gene expression and triggers plant acclimation responses (Figure 1) (Exposito-Rodriguez et al., 2017).

¹O₂ is mainly produced via energy transfer from triplet excited chlorophylls (³Chl^*) to oxygen in PSII (Krieger-Liszkay et al., 2008; Pospíšil, 2016). Unlike H₂O₂, ¹O₂ is believed not to diffuse to the nucleus, due to its short lifetime (~200 ns; Skovsen et al., 2005). However, ¹O₂ produced in PSII has been shown to directly react with carotenoids and thylakoid lipids, causing oxidative damage. Oxidative products of carotenoids have been shown to have a signaling role in stress responses (Hideg et al., 1998; Triantaphylidès and Havaux, 2009; Ramel et al., 2012).

Arguably, the most important photo-protective processes in plants are collectively measured by the parameter Non-photochemical quenching (NPQ). This term includes different components that exhibit distinct activation and deactivation kinetics (Standfuss et al., 2005; Johnson and Ruban, 2009; Niyogi, 2009; Ruban et al., 2012; Derks et al., 2015). The major, and fastest, component of NPQ is high energy state quenching (qE). qE is activated by the formation of a pH gradient (ΔpH) between the thylakoid lumen and the chloroplast stroma, and involves protonation of PSII subunit S (PsbS) and the de-epoxidation of zeaxanthin from violaxanthin through the xanthophyll cycle (Pascal et al., 2005; Takizawa et al., 2007; Ruban et al., 2012).

An increase in light intensity results in more protons being transferred from the chloroplast stroma to the thylakoid lumen via photosynthetic electron flow and, as a result, the pH in the thylakoid lumen drops. Protonation of PsbS leads to conformational changes in LHCII, which in turn increase the amount of energy that is quenched as heat, preventing ROS overproduction and protecting PSII from photodamage (Henmi et al., 2004; Vass, 2012; Zavafer et al., 2015). Under fluctuating light conditions, changes in ΔpH are a transient signal for qE, forming and decaying within seconds. However, the interconversion of violaxanthin and zeaxanthin in response to changes in irradiance occurs over a longer timescale (minutes).

Cyclic electron transport (CET) is another regulatory process involved in photoprotection in plants (Finazzi and Johnson, 2016; Yamori et al., 2016; Yamamoto and Shikanai, 2019), which plays an important role in generating the ΔpH required to trigger qE (Suorsa et al., 2016; Nakano et al., 2019). In CET, electrons are transferred from ferredoxin back to the plastoquinone (PQ) pool, and subsequently to PSI through Cyt b₆f and plastocyanin (PC). Thus, CET imports protons from the stroma to the lumen for ATP generation, but without net production of NADPH (Breyton et al., 2006; Joliot and Johnson, 2011). CET helps to balance the ATP:NADPH ratio under circumstances where the rate of consumption of reducing equivalents is reduced. In Arabidopsis, two main pathways for cyclic electron flow have been identified (Munekage et al., 2008; Shikanai, 2014). In the antimycin A-sensitive pathway, electrons are transferred from ferredoxin to PQ via a pathway involving the PGR5/PGRL1 complex (Munekage et al., 2002; DalCorso et al., 2008). In contrast, in the antimycin A-insensitive pathway, the electron transfer to the PQ pool is facilitated by NADH:plastoquinone oxidoreductase (NDH; Joliot and Johnson, 2011; Shikanai, 2014). For an extensive review of CET, see Nawrocki et al. (2019).

Of the two pathways of CET, the antimycin A-sensitive pathway has been particularly linked to PSI photoprotection under fluctuating conditions (Suorsa et al., 2012; Yamamoto and Shikanai, 2019). Studies in the pgr5 mutant of Arabidopsis show that PGR5 participates in photosynthetic control of Cyt b₆f (Nandha et al., 2007), protecting PSI from photodamage under low/high light cycles (Suorsa et al., 2012, 2013; Yamamoto and Shikanai, 2019). In addition, PGR5 plays a role in the acceptor-side regulation of PSI. Accordingly, it was recently observed that over-expression of PGR5 in the C4 plant Flaveria bidentis, enhances the electron sink downstream of PSI, increasing photoprotection (Tazoe et al., 2020). The water-water cycle has also been suggested as a PSI photoprotective mechanism under fluctuating light conditions, although its activity is strongly species dependent (Huang et al., 2019b; Yang et al., 2020). In this cycle, stromal antioxidants enzymes catalyze ROS conversion into water (Asada, 1999), a process which was recently suggested to be more relevant for PSI photoprotection than CET in angiosperms (Sun et al., 2020).

In addition to qE, another important NPQ component in plants is qT, a form of quenching associated with state transitions (for a review see Minagawa, 2011). This process regulates the distribution of excitation energy between both photosystems, under conditions where the incident light favors the excitation of one over the other. To mitigate such changes, plants can adjust the energy excitation between PSII and PSI within minutes, by altering the distribution of light harvesting proteins between them. The imbalance in the excitation level of the photosystems is sensed through changes in the redox state of the PQ pool (Lemeille and Rochaix, 2010). In particular, reduction of the PQ pool and binding of PQH₂ at the Qo site of the Cyt b₆f , activates a specific kinase (STN7) that phosphorylates LHCII trimers (Wollman and Lemaire, 1988; Vener et al., 1995; Zito et al., 1999; Depège et al., 2003; Bellafiore et al., 2005). Phosphorylation induces LHCII detachment from PSII and partial (or total) attachment to PSI, triggering transition to State II (Kyle et al., 1983; Larsson et al., 1983). Meanwhile, when the PQ pool is oxidized, dephosphorylation of LHCII triggers the opposite phenomenon and transition to State I (Pribil et al., 2010; Shapiguzov et al., 2010). The signals produced from the redox state of the PQ pool are naturally transient, however, evidence shows a direct and rapid regulation of chloroplast gene expression in response to changes in PQ redox state (Pfannschmidt et al., 1999).

Although the classical view of state transitions has been associated with spectral changes in the quality of the incident light, thylakoid phosphorylation can also be triggered dynamically by changes in light intensity (Tikkanen et al., 2010; Grieco et al., 2012; Mekala et al., 2015). While under low white light intensity LHCII phosphorylation levels are maximal, under high light conditions LHCII phosphorylation is down-regulated and PSII core phosphorylation increases (Tikkanen et al., 2010). These opposite states do not change the relative excitation of PSII and PSI, but their regulatory function is related with the maintenance of an equal excitation pressure between both photosystems (Tikkanen et al., 2010). The main kinases and phosphatases involved in this phosphorylation pathway are STN7/STN8 and TAP38/PPH1, respectively; being their regulation particularly relevant under low light conditions (Tikkanen et al., 2010; Mekala et al., 2015). When light intensity increases, other regulatory mechanisms (such as NPQ) become more important for photoprotection (Tikkanen et al., 2010; Grieco et al., 2012). Under fluctuating light conditions, a role for STN7-dependent phosphorylation was also found in PSI photoprotection, through the maintenance of the redox stability of the electron transport chain (Grieco et al., 2012).

Both STN7 and STN8 are also capable of phosphorylating a range of proteins in the chloroplast (Schönberg et al., 2017), extending their involvement in the short-term response to fluctuating light onto further processes of acclimation. The regulatory pathways related with thylakoid protein phosphorylation have a key role in the photosynthetic responses to a changing environment, and their participation in the signals transduction pathway for acclimation needs further elucidation (reviewed by Grieco et al., 2016).

Ion homeostasis in the chloroplast is relevant to light sensing, not only due to its effect on enzymatic activity, but also due to its contribution to the regulation of the proton and electric potentials across the thylakoid membrane (Finazzi et al., 2015). Proton motive force (PMF), the driver of ATP synthesis, consists of two components: ΔΨ, the electrical potential gradient that is built due to ions moving in and out of the thylakoid lumen, and ΔpH. When proton concentration significantly increases in the thylakoid lumen, qE is activated, leading to the loss of energy as heat (Henmi et al., 2004; Vass, 2012; Zavafer et al., 2015). To maintain ATP production without promoting acidification of the lumen, which leads to NPQ activation, fluxes of counter ions (Cl⁻ influx, Mg²⁺ and K⁺ efflux) regulate the ΔΨ component of the PMF (Carraretto et al., 2013; Armbruster et al., 2014; Herdean et al., 2016).

Finetuning ΔpH and ΔΨ to better suit different environmental conditions can facilitate fast modulation of photosynthetic activity under fluctuating light conditions. For instance, a transporter that has been linked to fast photosynthetic regulation in Arabidopsis is AtVCCN1, which transports Cl⁻ ions into the chloroplast lumen (Herdean et al., 2016). Influx of Cl⁻ ions into the lumen triggers an increase in the ΔpH/ΔΨ ratio, by decreasing H⁺ efflux from the thylakoid membranes, inducing a faster NPQ response under sudden increases in light intensities. By contrast, potassium influx to the lumen via the K⁺ antiport (KEA3), has been identified as an important factor in the transition from high to low light (Armbruster et al., 2014, 2016; Galvis et al., 2020). KEA3 transfers K⁺ into the lumen and H⁺ out to the chloroplast stroma, decreasing ΔpH but maintaining the ΔΨ necessary for ATP production. KEA3 activity accelerates NPQ relaxation during the transition to low light, leading to a fast recovery of CO₂ assimilation (Armbruster et al., 2014).

Another K⁺ transporter, the two-pore K⁺ channel (TPK3), has been suggested to play a pivotal role in thylakoid ultrastructure organization and plant growth in Arabidopsis (Carraretto et al., 2013). TPK3 exports K⁺ and Ca⁺ ions, and is thought to modulate fast regulation of PMF to optimize photosynthetic activity under different light environments (Carraretto et al., 2013). However, recent results obtained by Höhner et al. (2019) showed that TPK3 is localized in the tonoplast and is not involved in photosynthetic regulation. These authors suggest the involvement of an as yet unknown additional K⁺ channel in photosynthetic acclimation to fluctuating light.

The light-induced enzymatic activation of the Benson-Calvin cycle was first discovered by Buchanan and colleagues in the 1960s, showing that CO₂ fixation was activated by light (reviewed by Buchanan et al., 2002; Michelet et al., 2013). This light activation pathway, called the ferredoxin/thioredoxin (Fd/TRX) system, regulates carbon metabolic pathways through post-translational redox modifications (reviewed by Ruelland and Miginiac-Maslow, 1999; Lemaire et al., 2007; Michelet et al., 2013; Nikkanen et al., 2017). Thioredoxins (TRX) in the chloroplast are reduced mainly by ferredoxin (Fd, the PSI electron acceptor), via an enzyme called Ferredoxin-Thioredoxin reductase (FTR). Once reduced, TRX can reduce disulfide bonds in different stromal target proteins, placing the Fd/TRX at the crossroads between the “light” and “dark” reactions of photosynthesis (Ruelland and Miginiac-Maslow, 1999; Lemaire et al., 2007).

Activation of the Fd/TRX system will directly depend on the redox state of the chloroplast, meaning that changes in the photosynthetic electron flow will activate/deactivate different target enzymes under changing light regimes. This on/off switch acts as a significant regulatory process, leading to the adjustment of the carbon metabolism under different conditions. However, most studies on the regulatory role of Fd/TRX have been conducted under continuous light conditions, and research on their involvement in fluctuating light responses is limited (Collin et al., 2004; Nikkanen and Rintamäki, 2014; Geigenberger et al., 2017).

A study performed on the Arabidopsis knockout mutants trxm1/m2 showed the role of thioredoxins in the short-term responses to fluctuating light (Thormählen et al., 2017). Mutant plants showed alterations in the light activation of the enzyme malate dehydrogenase (MDH) and the malate/oxaloacetate (Mal/OAA) shuttle, a higher NPQ and a lower PSII quantum efficiency. This phenotype was only evident under fluctuating light conditions, with these alterations being more pronounced with increasing numbers of high-low light cycles. By contrast, no phenotypic differences were seen between the mutants and the WT plants under constant light.

Stomatal responses play a critical role in the availability of CO₂ for carbon fixation, and it has been shown that stomatal dynamics limit photosynthesis under fluctuating light (Qu et al., 2016; Papanatsiou et al., 2019; De Souza et al., 2020; Kimura et al., 2020). Since stomatal responses are slower than photochemical and biochemical regulatory changes, a sudden change in light intensity could cause chloroplast CO₂ concentration to decrease (Huang et al., 2015; Vialet-Chabrand et al., 2017). A decrease in CO₂ concentration (and thus carbon fixation) implies that fewer electrons are being directed to the Benson-Calvin cycle, favoring the over-reduction of the electron transport chain and triggering ROS generation. At the same time, a decrease in CO₂ availability also increases O₂ binding to Rubisco, its oxygenase activity and photorespiration (Huang et al., 2015). Interestingly, it was recently shown that stomatal opening and closure dynamics can acclimate to different growth light regimes, anticipating future variations in light, and adjusting CO₂ availability to the prevailing light condition (Matthews et al., 2018).

Photorespiration is a metabolic pathway that recycles 2-phosphoglycolate (2PG), a toxic product of the oxygenase activity of Rubisco, into 3-phospholycerate (3PGA; reviewed by Foyer et al., 2009; Bauwe et al., 2010; Eisenhut et al., 2019). This recycling requires several enzymatic steps that are distributed across three different organelles: the chloroplast, the peroxisome, and the mitochondrion. Although the photorespiratory pathway can represent a substantial loss of CO₂ fixation, its involvement in photoprotection, nitrogen assimilation, and abiotic stress responses make it a crucial process for plants (reviewed by Foyer et al., 2009; Bauwe et al., 2010; Timm and Bauwe, 2013; Voss et al., 2013; Eisenhut et al., 2019). Nevertheless, its participation under fluctuating light conditions has not been extensively studied (Huang et al., 2015; Schneider et al., 2019). Huang et al. (2015) showed that, under fluctuating light conditions, a strong activation of the photorespiratory pathway allows the consumption of reducing equivalents, decreasing the reduction pressure of the electron transport chain and avoiding ROS generation. In addition, RuBP regeneration is also accelerated, favoring carbon fixation under these circumstances.

When plants are grown at higher irradiances, they typically develop leaves with a high capacity for photosynthesis (see Walters, 2005). Fully developed leaves transferred from low to high light can also increase their photosynthetic capacity, typically over a period of a week (Athanasiou et al., 2010; Dyson et al., 2015). This acclimation response involves extensive changes across the whole of the leaf proteome, with marked increases in the concentration of Rubisco and other enzymes involved in the Benson-Calvin cycle, as well as down-stream enzymes involved in carbon assimilation (Miller et al., 2017).

Schneider et al. (2019) observed an upregulation of some Benson-Calvin cycle enzyme genes, such as fructose-1,6-bisphosphate aldolase 1 (FBA1) and sedoheptulose-1,7-bisphosphatase (SBSPASE), in Arabidopsis plants subjected to fluctuating light for 3 days. These enzymes were previously shown to participate in the regulation of the metabolic flux of carbon in plants (Lefebvre et al., 2005; Uematsu et al., 2012; Simkin et al., 2015, 2017). In addition, SBSPASE and FBA1 were also found to be regulated by the Fd/TRX system (Breazeale et al., 1978; Sahrawy et al., 1997; Dunford et al., 1998), suggesting a fine-tuning regulation of this long-term acclimation by a short-term mechanism. However, despite the increased activity and/or concentration of their Benson-Calvin enzymes, when compared to constant light conditions, plants under fluctuating light do not necessarily show an enhancement of their CO₂ fixation capacity (Watling et al., 1997; Vialet-Chabrand et al., 2017; Schneider et al., 2019). Studies show that proteomic and transcriptomic changes in response to fluctuating light do not always align, suggesting a role of post-transcriptional regulations in the modulation of long-term acclimation responses (Athanasiou et al., 2010; Dyson et al., 2015; Miller et al., 2017; Schneider et al., 2019; Niedermaier et al., 2020).

Furthermore, as part of the acclimation response of carbon metabolism to fluctuating light, an increase in the expression of photorespiratory genes, and their corresponding protein content, was also observed in Arabidopsis (Schneider et al., 2019; Niedermaier et al., 2020). This metabolic response was shown to be particularly significant under high light fluctuation periods (Huang et al., 2015). Under low light fluctuating regimes, an increase in the photorespiratory pathway was deemed insignificant (Kono et al., 2014), possibly due to a lower accumulation of reducing equivalents.

In addition to metabolic alterations, changes in the thylakoid membrane protein composition play an important role in light acclimation (reviewed by Walters, 2005; Anderson et al., 2012; Kaiser et al., 2018; Johnson and Wientjes, 2020). For instance, plants grown under high light have been observed to have a lower PSII/PSI ratio, but higher concentrations of Cyt b₆f and ATPase (reviewed by Evans, 1988; Eskins et al., 1991; Walters and Horton, 1994; Bailey et al., 2001; Walters, 2005). Furthermore, high light may also reduce the amount of LHCII and increase the chlorophyll a/b ratio, which is related with changes in light harvesting complexes concentration and photosystems ratio (Leong and Anderson, 1984; Yang et al., 1998; Bailey et al., 2001). By contrast, LHCII concentration increases when light exposure is limiting for plant growth, although under these conditions, a compensating decrease in PSII levels is also observed (Evans, 1988; Bailey et al., 2001). Some, but not necessarily all, of these responses are seen when plants are exposed to step changes in irradiance. In Arabidopsis, transfer from low to moderately high light resulted in an increase in Cyt b₆f and ATPase, without measurable changes in chlorophyll content or the total amount of LHC proteins (Athanasiou et al., 2010; Miller et al., 2017).

In addition to acclimation to overall light intensity, when plants are exposed to different light qualities, the protein composition of the thylakoid membranes may also change (reviewed by Anderson et al., 1988). Long-term acclimation responses include changes in LHCII concentration, and Chl a/b and PSII/PSI ratio (Chow et al., 1990; Kim et al., 1993; Walters and Horton, 1995; Murchie and Horton, 1998). Such changes, occurring within days; help balance the electron transport rate under situations where either photosystem is preferentially excited by light. Importantly, these alterations differ from state transitions, which occur in seconds to minutes and do not include changes in thylakoid membrane composition, only re-distribution of LHCII and photosystem macro-organization.

Overall, when light intensity or quality change, plant acclimation responses tend to balance light absorption and assimilation (Rott et al., 2011; Yamori et al., 2011). Nevertheless, understanding the effect of fluctuating light in natural environments is far more complex. Sunflecks will have a direct impact on light intensity, but natural shade cast by vegetation will also affect the incident light quality. Thus, natural light fluctuations in the ecosystems result in complex inputs, which might induce contradictory output responses. For instance, the PSII/PSI ratio will change in opposite directions with a decrease in light intensity or with exposure to a high far-red/red ratio, conditions that can be imposed by vegetative shading (Murchie and Horton, 1998; Bailey et al., 2001). Consequently, predicting a thylakoid membrane specific response to fluctuating light in natural environments is not an easy task. Furthermore, many of these responses are species-dependent (Murchie and Horton, 1998; Yin and Johnson, 2000).

Similar to what happens following sudden increases in light intensity, frequent exposure to oscillating periods of high light induces over-reduction of the electron transport chain, triggering an increase in ROS production and photoinhibition (Shimakawa and Miyake, 2018; Huang et al., 2019a). Thus, long-term acclimation to fluctuating light can also involve an enhancement of photoprotective mechanisms. For instance, Schneider et al. (2019) observed an up-regulation of H₂O₂ scavenging enzymes, such as glutathione peroxidase (GPX7) and catalase (CAT2), in Arabidopsis leaves exposed to 3 days of fluctuating light. In agreement, these plants also increased their ascorbate pool size, indicating an improvement in ROS scavenging and antioxidant response (Schneider et al., 2019).

Fluctuating light can increase PSBS content and the concentration of pigments of the xanthophyll cycle, leading to a strengthening of the photoprotective capacity of NPQ (Barker et al., 1997; Niinemets et al., 1998; Alter et al., 2012; Caliandro et al., 2013). This acclimation response was demonstrated to be, at least partially, regulated at the transcriptional level (Schneider et al., 2019). The relevance of NPQ as a long-term acclimation response to fluctuating light was recently shown using tobacco transgenic lines overexpressing PSBS and zeaxanthin epoxidase (ZEP) and violaxanthin de-epoxidase (VDE), the key enzymes in the xanthophyll cycle (Kromdijk et al., 2016). These plants showed a higher CO₂ assimilation compared to the WT, leading to a higher dry mass accumulation under field conditions.

In addition to NPQ, genes related to CET were also upregulated in response to fluctuating light (Schneider et al., 2019). Proteomic results obtained under the same conditions, and by the same authors, suggest a specific role of the NDH-like complex in this long-term acclimation mechanism (Niedermaier et al., 2020). Nevertheless, mutants lacking PGR5 were shown to suffer strong PSI photoinhibition under fluctuating light (Suorsa et al., 2012; Kono and Terashima, 2016), meaning that the involvement of the antimycin A-sensitive CET pathway in this acclimation process cannot be ruled out. It is worth mentioning that a higher CET flux will contribute to a higher ΔpH, facilitating NPQ generation under photoinhibitory conditions (Munekage et al., 2002). Thus, the involvement of CET in this long-term acclimation response will not only avoid PSI photoinhibition, but also increase thermal dissipation. Consequently, short-term responses to sudden increases in light intensity are also being improved by this long-term response.

Retrograde signals originating from photosynthesis have been studied extensively, aiming to describe chloroplast-nucleus communication and regulation of gene expression (Pfannschmidt et al., 1999; Fey et al., 2005b; Wilson et al., 2006; Foyer et al., 2012; Karpiński et al., 2013; Gollan et al., 2015; Matsubara et al., 2016; Leister, 2019). Within these signaling pathways, the participation of ROS originating from photosynthetic electron flow, has been widely discussed (Pfannschmidt et al., 2009; Ramel et al., 2012; Szechyńska-Hebda and Karpiński, 2013; Kim, 2020). For instance, H₂O₂ can diffuse from the chloroplast, creating a signal cascade ultimately affecting nuclear gene expression (Maruta et al., 2012). Meanwhile, oxidation of β-carotene by ¹O₂ has been shown to lead to the formation of the volatile compound β-cyclocitral (β-CC), which triggers changes in nuclear gene expression under stress conditions (Ramel et al., 2012; Havaux, 2014; Tian, 2015). Nevertheless, the specific factors participating in this ROS-related signal transduction pathways are still far from being completely elucidated (reviewed by Leister, 2019; Kim, 2020).

In addition, we can think the chloroplast as an environmental sensor for plants, with the redox state of the electron transport chain being a key gear for sensing fluctuations in the environment. Amongst the components of the electron transport chain, the involvement of the PQ pool in retrograde signaling has been suggested, facilitating a rapid physiological response to changing light conditions (El Bissati and Kirilovsky, 2001; Fey et al., 2005a; Bräutigam et al., 2009). Different proteins have been shown to be regulated by the redox state of PQ, and in a ROS-independent manner (Adamska and Kloppstech, 1991; Kimura et al., 2003; Yabuta et al., 2004), and a link between the redox state of the PQ pool and photosystem gene expression has even been observed (Pfannschmidt et al., 1999; El Bissati and Kirilovsky, 2001).

Although it is clear that ROS retrograde signaling and chloroplast redox changes participate in light acclimation, it is still not yet understood how these signals are integrated under fluctuating light. The short lifetime of these chloroplast redox changes imply that other factors might be integrating the redox variations over time. Puthiyaveetil et al. (2008) suggested that a key protein participating in retrograde signaling is the Chloroplast Sensor Kinase (CSK) (Figure 2; Puthiyaveetil et al., 2008). This protein is widely seen in photosynthetic organisms and has been shown to bind an iron-sulfur cluster, which can sense changes in the chloroplast redox state (Ibrahim et al., 2020). The activation/deactivation of the kinase regulates the expression of different photosynthetic genes (Puthiyaveetil et al., 2008; Ibrahim et al., 2020), supporting its participation in the adjustment of the stoichiometry of photosynthetic complexes under different light conditions. Interestingly, CSK gene expression was found to be upregulated under fluctuating light in Arabidopsis (Schneider et al., 2019).

The STN7 kinase phosphorylates LHCII and some PSII subunits upon changes in light conditions, triggering state transitions and regulating PSII turnover (Bellafiore et al., 2005; Tikkanen et al., 2006, 2010; Wagner et al., 2008; Pietrzykowska et al., 2014). Although STN7 involvement in short-term regulatory responses to fluctuating light has been well-described, it was also shown that stn7 mutants are unable to undergo forms of long-term acclimation under changing light regimes (Bonardi et al., 2005; Pesaresi et al., 2009). When stn7 mutants were grown under fluctuating light they exhibited reduced growth and a lower seed yield and were incapable of adjusting their thylakoid composition to the conditions experienced. In agreement, STN7 gene expression was increased after 3 days of fluctuating light in Arabidopsis (Schneider et al., 2019), again highlighting its potential involvement in long-term acclimation. Thus, STN7 seems to play a critical role in both short- and long-term responses to fluctuating light regimes in plants, either directly or indirectly (Bonardi et al., 2005; Wagner et al., 2008; Bräutigam et al., 2009; Leister, 2019).

Although it is still unknown which proteins participate downstream of STN7 in the long-term acclimation response, one of the putative proteins is TSP9. TSP9 is a plant specific nuclear-encoded protein, found in the thylakoid membranes, which is phosphorylated by STN7 upon illumination (Carlberg et al., 2003; Zer and Ohad, 2003). The phosphorylated form dissociates from the thylakoid membrane and has been proposed to act as a signaling molecule regulating gene expression under changing light conditions (Carlberg et al., 2003; Zer and Ohad, 2003; Fristedt et al., 2009). Nevertheless, downregulation of TSP9 in Arabidopsis plants did not affect the long-term response to light changes (Pesaresi et al., 2009), but its mutation was shown to affect state transitions and NPQ (Fristedt et al., 2009). These results suggest the involvement of TSP9 in the short-term responses to fluctuating light, but not in long-term acclimation. The downstream factors involved in the STN7 long-term acclimation signaling pathway, are still elusive (Leister, 2019).

Mathematical modeling within biology is ruled by two paradigms: empirical modeling and mechanistic modeling. Empirical modeling, also known as statistical modeling, fits a model to the data without considering the underlying biological processes (Table 1). Through the observation of a repeated pattern, it is assumed that future events of the same type will result in the same pattern. A straightforward example of an empirical model is the non-rectangular hyperbola of net carbon gas exchange fitted to light response curves (Johnson and Murchie, 2011). While the fit of this model has stood the validation test of time, possible underlying biological mechanisms have only recently been discussed (Retkute et al., 2015; Herrmann et al., 2020). Stegemann et al. (1999) constructed an empirical model relating fluctuating diurnal changes in light intensity to net photosynthesis. By fitting their model parameters to data obtained from two different tree species, they successfully estimated carbon uptake without explaining the mechanisms behind their photosynthetic responses to light.

Empirical models are a powerful tool when the underlying processes are not known and form the premise of machine learning algorithms (Kotsiantis et al., 2007; Angelov and Gu, 2019). The reliability of empirical models improves vastly with the amount of input data available (Kotsiantis et al., 2007), a limitation that is becoming less hindering in the current ‘omics era. Nonetheless, extrapolation of empirical models is difficult and, typically, good predictions cannot be made outside of the range of previously measured values. For example, an empirical model of photosynthesis with parameters fitted to a specific light and temperature regime is unlikely to be transferable to another light and temperature regime, and, instead, the model parameters must be estimated anew (Herrmann et al., 2020).

By contrast, mechanistic models, albeit harder to construct, have several advantages over empirical models (Table 1). Mechanistic models break down a system into smaller components, and the processes by which these components interact with one another are then captured by mathematical equations. Mechanistic models require an in-depth understanding of the system components and their interactions in space and time. However, once a mechanistic model is constructed and its parameters are successfully estimated, few input data are required for outcome prediction. Furthermore, if the same mechanisms apply under different conditions, or outside the range of the initial input values, the model can be applied beyond the range of the initial training data (Geritz and Kisdi, 2012; Ratti, 2018).

Kinetic models of metabolic pathways are examples of mechanistic models: a pathway is broken down into its metabolites and the way in which these metabolites interact with one another can, for example, be described by mass action law or Michaelis-Menten kinetics (Schallau and Junker, 2010). Many successful kinetic models of photosynthesis have been built, and their ability to capture fluctuating light conditions is discussed in the next section. Kinetic models, as with all others mechanistic models, are limited by our knowledge of the system under study.

Dynamic models generally encompass time-dependent-models that capture changes over time (Table 1). These models employ a set of ordinary differential equations, or partial differential equations, considering one or more independent variables. Dynamic models can be empirical or mechanistic; however, in biochemistry, dynamic models appear most commonly in the form of kinetic models. Kinetic models of biochemical pathways are mechanistic, dynamic models, as they consider changes in metabolite concentrations over time. Due to a combinatorial explosion of the parameter estimation, dynamic models are generally limited to a handful of equations.

Multiple dynamic models of photosynthesis and the Benson-Calvin cycle reactions exist (Farquhar et al., 1980, 2001; Harley and Tenhunen, 1991; Poolman et al., 2000); however, most of them have not been applied to study fluctuating light acclimation. Kirschbaum et al. (1997) and Pearcy et al. (1997) are among the few to have extended the original Farquhar et al. (1980) model to study fluctuating light regimes over a time frame of seconds to hours. Mott and Woodrow (2000), however, addressed the question of nitrogen resource allocation under fluctuating light regimes by using a much simpler model of rubisco and rubisco activase. In addition, Porcar-Castell et al. (2006) incorporated both regulatory and feedback mechanisms in their dynamic model of PSII, and were able to validate experimentally obtained photochemical and non-photochemical quantum yields under fluctuating light. However, none of these models consider a long-term acclimation of plants to changing light regimes, as they have been parametrized for a much shorter timescale.

Retkute et al. (2015) employed a semi-empirical dynamic model to describe carbon uptake over time as a function of light availability and a constant maximum photosynthetic capacity (Pmax). Pmax is calculated to give the maximum possible carbon uptake over the time-weighted average of a light pattern, representing the acclimation state of the plant. How plants shift from one acclimation state to another, and alter their Pmax accordingly, was discussed by Herrmann et al. (2020) using a time- and temperature-dependent model. However, the sensors and signals that trigger a new photosynthetic acclimated state remain elusive.

As discussed in section Signals Transduction Pathways for Long-Term Acclimation to Fluctuating Light, and highlighted in our previous studies (Dyson et al., 2015; Herrmann et al., 2020), carbon fluxes between the cytosol and the chloroplast seem to be important factors in the photosynthetic acclimation responses of plants. In particular, the resulting changes in sugar vs. starch production have been shown to be crucial for acclimation to different light regimes (Dyson et al., 2015). Modeling the changes in carbon metabolism under fluctuating light could be useful to identify key signals leading to acclimation responses in plants. For instance, using a simple kinetic model and a sensitivity analysis of the model parameters, Nägele and Weckwerth (2014) analyzed the control of sugar homeostasis in plants, and suggested that allosteric effectors alone can account for a considerable readjustment of metabolic homeostasis.

Metabolic control analysis (MCA) quantifies the extent to which fluxes, or concentrations, depend on the model parameters. Thus, when applied to kinetic models, MCA provides a valuable tool for identifying parameters, and thus enzymes, that exert the greatest metabolic control over the fluxes (or species concentrations) in a defined model (ap Rees and Hill, 1994; Poolman et al., 2000). The fact that detailed regulatory information can be included in these dynamic kinetic models represents one of their greater advantages; although at the same time limits the size and complexity over which they can be feasibly solved. Often, a trade-off between the level of mechanistic detail and the feasibility to solve the model is required (Harley and Tenhunen, 1991).

Photosynthetic acclimation to a sustained change in light regime typically occurs over multiple days. As Athanasiou et al. (2010) observed, a new photosynthetic state is reached only 1 week after a change in exposure from low to high light. Acclimation responses to changes in light regimes are typically not represented by kinetic models, which tend to be parametrized over a timescale of seconds to hours. Instead, genome-scale steady-state models which tend to operate over multiple days and weeks can be used to study photosynthetic acclimation (Table 1; Herrmann et al., 2019a).

Genome-scale, steady-state, metabolic models, employ what are known as constraint-based modeling (CBM) techniques (Lewis et al., 2012), and operate under the assumption that internal metabolite concentrations are constant over time. This approximation is generally valid over longer time-frames, because the changes in metabolic flux leading to new equilibrium states are usually faster when compared to acclimation responses. Whilst these types of models are able to capture the final acclimated steady-state of plant metabolism, they fail to incorporate the mechanisms that initiate and lead to that new steady-state.

More recent variations of CBM techniques aim to overcome these limitations by employing dynamic CBM techniques (Mahadevan et al., 2002; Grafahrend-Belau et al., 2013). Shaw and Cheung (2018), for example, built a dynamic multi-tissue model by using the output of one steady-state model as the input of another steady-state model. By defining a steady-state model at each time point, they were able to effectively analyse resource allocation in plants over days and weeks. The model by Shaw and Cheung (2018) is based on the diel model first published by Cheung et al. (2014), which combined both a day-time (light-dependent) steady-state model and a night-time (light-independent) steady-state model. Dynamic models constructed of steady-state models are thus able to incorporate time-dependent changes, but typically consider changes over a timescale of multiple days.

Different ‘omics datasets can be used to incorporate enzyme regulatory mechanisms into genome-scale stoichiometric models, making it, for instance, possible to study redox changes at a resolution of second to hours with longer process steady-state models (Jamshidi and Palsson, 2010). This kind of analysis, however, has yet to be applied to plants exposed to fluctuating light.

Incorporating data measured over different timescales, and both discrete (e.g., state switching) and continuous scales (e.g., sink metabolite accumulation), poses an immense challenge for the study of photosynthetic acclimation under fluctuating light conditions. In order to incorporate models describing different processes over different periods of times, dimensionality reduction techniques will be necessary. These techniques are designed to reduce model complexity by discarding components that have little effect on the overall outcome of interest (Hummer and Szabo, 2015; Snowden et al., 2017). Successful dimensionality reduction should lead to the identification of essential model components required for predictive power; each fine-tuned according to the timescale over which it must operate, and the magnitude for which can shows significant effects on the system itself.

Purvis et al. (2009) use dimensionality reduction to combine multiple small-scale kinetic models in the human platelet P2Y1 signaling system, and convert it into a single holistic model. Employing known dimensionality reduction techniques (Hummer and Szabo, 2015; Ali Eshtewy and Scholz, 2020) on fast existing kinetic models of photosynthesis, may pave the way for their incorporation into slower process models (such as the genetic changes involved in light acclimation). Unfortunately, neither dimensionality reduction nor model validation techniques are frequently employed in plant sciences. Model validation, both at the experimental and theoretical level (Hasdemir et al., 2014), will need to be done before a model is deemed suitable to be integrated holistically.

Hybrid models, which incorporate both discrete and continuous information, are starting to gain attention in other disciplines (Henzinger, 2000; Bortolussi and Policriti, 2008). A simple analogy for such a hybrid automaton is a thermostat, whereby the law of thermodynamics are described by ordinary differential equations (continuous) but the state of the heater is either on or off (discrete). Thus, we can imagine multiple metabolite concentrations changing on a continuous scale in response to environmental fluctuations, which could emerge in an on/off output response (such as the ones described in sections Signals Transduction Pathways for Long-Term Acclimation to Fluctuating Light and System Outputs: Long-Term Acclimation to Fluctuating Light of the present review, respectively). However, such a system has yet to be identified in plant acclimation to fluctuating light.

The alternative to the mechanistic approaches described above, would be to take an empirical modeling approach. With an ever-increasing and overwhelming amount of multi-omic data available, there are numerous supervised learning algorithms that could be applied to identify “biomarkers” of a given acclimation stage (Mjolsness and DeCoste, 2001; Saeys et al., 2007). While the identification of molecular predictors is promising, these approaches typically do not reveal any information about the mechanisms by which the identified molecules trigger the final acclimated state of the plant. This empirical approach, however, does hold the potential for validating existing hypotheses or generating new hypotheses for experimental validation. For example, if both STN7 and TSP9 were identified as predictors for light acclimation, this would support the idea that TSP9 acts downstream of STN7 in promoting a long-term acclimation response.

Finally, stochastic models can account for random variations in inputs and result in a probability distribution of potential outcomes (Table 1; Guerriero et al., 2014; Retkute et al., 2018). The application of stochastic models to photosynthesis remains limited as of today but holds a great potential for identifying potential thresholds for acclimation. For instance, one could imagine a metabolite concentration that fluctuates in response to environmental changes triggering an acclimation process only once a given threshold concentration is passed. This could be the case of the proposed ROS and carbon-related metabolites discussed above (see section Signals Transduction Pathways for Long-Term Acclimation to Fluctuating Light).

To facilitate the integration of different modeling techniques, rigorously standardized tools are required. For instance, the open platform www.e-photosynthesis.org hosts a collection of dynamic plant models, translated to the Systems Biology Mark-up Language (SBML), and provides a good starting point for any modeler interested in photosynthetic acclimation. Currently, the project consists largely of model parametrized over short timescale, but hopefully it will be extended to longer-time physiological processes in the future, such that an effective integration of the two will be possible.