Integration of physiologically relevant photosynthetic energy flows into whole genome models of light‐driven metabolism

SUMMARY

Characterizing photosynthetic productivity is necessary to understand the ecological contributions and biotechnology potential of plants, algae, and cyanobacteria. Light capture efficiency and photophysiology have long been characterized by measurements of chlorophyll fluorescence dynamics. However, these investigations typically do not consider the metabolic network downstream of light harvesting. By contrast, genome‐scale metabolic models capture species‐specific metabolic capabilities but have yet to incorporate the rapid regulation of the light harvesting apparatus. Here, we combine chlorophyll fluorescence parameters defining photosynthetic and non‐photosynthetic yield of absorbed light energy with a metabolic model of the pennate diatom Phaeodactylum tricornutum. This integration increases the model predictive accuracy regarding growth rate, intracellular oxygen production and consumption, and metabolic pathway usage. Through the quantification of excess electron transport, we uncover the sequential activation of non‐radiative energy dissipation processes, cross‐compartment electron shuttling, and non‐photochemical quenching as the rapid photoacclimation strategy in P. tricornutum. Interestingly, the photon absorption thresholds that trigger the transition between these mechanisms were consistent at low and high incident photon fluxes. We use this understanding to explore engineering strategies for rerouting cellular resources and excess light energy towards bioproducts in silico. Overall, we present a methodology for incorporating a common, informative data type into computational models of light‐driven metabolism and show its utilization within the design–build–test–learn cycle for engineering of photosynthetic organisms.

Significance Statement

Whole genome models of metabolism are improved by the addition of parameters associated with measured photo‐physiology. The approach is demonstrated in the photosynthetic alga, Phaeodactylum tricornutum, but this approach could be applied to any plant system.

INTRODUCTION

There is great interest in characterizing light‐driven metabolism as a result of the ecological importance and engineering potential of phototrophic microorganisms and plants. Oxygenic photosynthesis utilizes light energy to generate an oxidized protein complex capable of extracting electrons from water at photosystem II (PSII), at the same time as re‐energizing the extracted electron to reduce NADP⁺ at photosystem I (PSI). These ‘light harvesting’ reactions drive electron transport, ATP generation, and subsequent CO₂ fixation through the Calvin–Benson–Bassham cycle (CBBC) in addition to the other energy consuming reactions throughout the cell.

Light absorption by a photosynthetic cell is not constant. Because light fluxes can vary across the day and as a result of ecological or climatological features, photosynthetic microorganisms often absorb more photons than can be utilized by metabolism during these natural fluctuations. If this excess energy is not dissipated, over‐reduction of the photosynthetic electron transport chain (ETC) occurs. The resulting formation of reactive oxygen species causes damage to proteins, lipids, and nucleic acids (Dietz et al., ²⁰¹⁶; Niyogi, ²⁰⁰⁰). Photoinhibition is a product of this damage, and it decreases photosynthetic efficiency because of damage from excess light capture. The PSII D1 subunit is the primary photoinhibition target in the photosynthetic ETC (Edelman & Mattoo, ²⁰⁰⁸). A complex repair cycle characterized by removal, degradation, and de novo synthesis is constitutively active to counter this damage and it is energetically expensive (Nixon et al., ²⁰⁰⁵). To prevent photoinhibition, excess energy can be dissipated upstream of the photosynthetic ETC complexes via a variety of mechanisms encompassing non‐photochemical quenching (NPQ), which harmlessly converts excitation energy to heat (Nicol et al., ²⁰¹⁹). NPQ not only protects the photosynthetic system from oxidative stress, but also reduces the apparent efficiency of light‐biomass conversion as a smaller fraction of captured light energy enters the broader metabolic network.

Here we coin the term excess electron transport (EET) as an additional important physiological feature at the intersection of photophysiology and bioengineering. This comprises several components that act either as shunts within the ETC (Jallet, Cantrell, & Peers, ²⁰¹⁶; Ware et al., ²⁰²⁰) or downstream of the photosynthetic machinery within the broader metabolic network. It relieves over‐reduction of the photosynthetic ETC by dispelling electrons generated by excess light (Jallet, Cantrell, & Peers, ²⁰¹⁶). Usually, these reactions are considered metabolically ‘futile’ as the electrons are deposited on elemental oxygen to generate water, for example. However, they are important in relieving photosynthetic ETC over‐reduction. There is interest in the bioengineering field to redirect these electrons away from metabolic futility towards bioproducts of interest, at the same time as maintaining the beneficial effects on ETC redox balance (Lassen et al., ²⁰¹⁴; Levering et al., ²⁰¹⁵). Harnessing excess reductant can convert endogenous carbon sinks, such as carbohydrates, into more energy dense products such as lipids. Additionally, recently, it was shown that engineered reductant sinks can actually increase carbon fixation and overall photosynthetic efficiency (Santos‐Merino et al., ²⁰²¹). Thus, downregulating evolutionarily beneficial processes for photosynthetic individuals in favor of mass culture productivities offers promising avenues for increasing bioproduct and biofuel efficiency (Peers, ²⁰¹⁴). Quantitative characterization of the push–pull of light capture upstream and dissipation in the metabolic network downstream of the photosynthetic ETC would enable design and optimization of these engineered reductant sinks.

Properly accounting for EET facilitates this bioprocess optimization and provides insight into photoprotection strategies. Previous work in photosynthetic microorganisms (diatoms and green algae) has used photophysiology parameters derived from chlorophyll fluorescence measurements to estimate EET (Wagner et al., ²⁰⁰⁶). In this previous framework, EET was calculated as the difference between the total absorbed photons and the excitation energy required for biomass production and cellular maintenance. Additionally, the fraction of total absorbed light energy lost upstream of the photosynthetic ETC was estimated using chlorophyll fluorescence data, which have long been employed to assess phototrophic physiology (Krause & Weis, ¹⁹⁹¹).

Chlorophyll fluorescence primarily quantifies the fate of absorbed light energy directed to PSII; however, there is evidence of contributions from PSI as well (Giovagnetti et al., ²⁰¹⁵; Pfündel et al., ²⁰¹³). This excitation energy has three primary fates: it can perform photochemistry at PSII, it can be dissipated as heat through NPQ processes, or it can be dissipated by other, less well characterized non‐radiative and fluorescence processes (NO). All of these can be quantified through the use of pulse amplitude modulation (PAM) chlorophyll fluorimetry (Kramer et al., ²⁰⁰⁴). When these values are normalized to the total excitation energy routed to PSII, they are annotated as the quantum yields Y(II), Y(NPQ), and Y(NO), respectively, the sum of which is always one. These techniques have unveiled the diverse photoprotective strategies employed by photosynthetic microorganisms to include extensive NPQ in the diatom Phaeodactylum tricornutum (Lavaud et al., ²⁰⁰²). However, these important aspects of photosynthesis have not been integrated into models of total cellular metabolism.

Constraint‐based modeling coupled with flux balance analysis (FBA) has successfully been employed to characterize and engineer a wide range of biological systems (Bordbar et al., ²⁰¹⁴; Küken & Nikoloski, ²⁰¹⁹). Constraint‐based modeling relies on a reconstruction of the metabolic content of the organism of interest, which, when performed at the whole‐organism level, results in a genome‐scale knowledge base. Leveraging this genome‐scale reconstruction to compute cellular phenotypes via FBA results in a genome‐scale model (GEM). There have been several advances in the metabolic modeling of photosynthetic organisms to include plants (de Oliveira Dal'Molin et al., ²⁰¹⁰; Poolman et al., ²⁰¹³), cyanobacteria (Broddrick et al., ²⁰¹⁶; Broddrick, Welkie, et al., ²⁰¹⁹), green algae (Chang et al., ²⁰¹¹; Zuñiga et al., ²⁰¹⁷), and diatoms (Levering et al., ²⁰¹⁶). Typically, most models integrate light harvesting into photosynthetic GEMs by assuming a linear relationship between light absorption and photophosphorylation (Shastri & Morgan, ²⁰⁰⁵), However, recent modeling in the diatom P. tricornutum quantified growth rates, excitation energy partitioning between the photosystems, and cross‐compartment energetic coupling of the chloroplast and mitochondrion (Broddrick, Du, et al., ²⁰¹⁹). However, that study still used simplified assumptions regarding light harvesting, possibly affecting the accuracy of absolute fluxes predicted by the model.

The metabolic network that underpins GEMs is assembled from the reactant and product stoichiometry of biochemical reactions; thus, it should be feasible to couple the representation of chlorophyll fluorescence parameters as a fraction of light energy routed to PSII as a stoichiometrically balanced biochemical equation. Such a framework would enable the explicit integration of chlorophyll fluorescence data and EET as a constraint on photosynthetic metabolic processes towards an increased understanding of photoacclimation, photoprotection, and bioengineering of phototrophic metabolism.

RESULTS

Cell physiology of P. tricornutum at low and high light

Phaeodactylum tricornutum was acclimated and cultured at a high light irradiance of 600 μmol photons m⁻² sec⁻¹ (HL, n = 4) and a low light irradiance of 60 μmol photons m⁻² sec⁻¹ (LL, n = 3). The range of growth rates for P. tricornutum was 0.026–0.029 (n = 3) and 0.052–0.053 (n = 4) h⁻¹ at LL and HL, respectively. Cell volume differed by approximately 10% [202 ± 43 versus 184 ± 47 μm³ at HL (n = 94) and LL (n = 46), respectively] and dry cell weight was also similar between the cultures (Table S1). There were differences in the chlorophyll content at LL and HL, as is typical for microalgae (Falkowski & Owens, ¹⁹⁸⁰). Total chlorophyll per cell [chlorophyll a (chla) and chlorophyll c (chlc)] at LL was 2.8‐fold higher than HL (Table S1), which resulted in a three‐fold increase in the cell‐normalized absorption coefficient at LL compared to HL (a*_cell) (Figure 1a). The consistency in the pigment mass‐normalized absorption coefficient under both light conditions (a* _pigm) (Figure 1b) suggested efficient photoacclimation between these light levels. The chla to chlc ratio varied slightly from 6.0 at LL to 5.2 at HL. Overall, the chlorophyll content per cell was consistent with previous observations of photoacclimation in P. tricornutum (Broddrick, Du, et al., ²⁰¹⁹; Nymark et al., ²⁰⁰⁹).

Figure 1.

Photophysiology of P. tricornutum acclimated to low and high light.

(a) Cell‐specific absorption coefficient. (b) Pigment‐specific absorption coefficient. The pigment mass includes chlorophyll a and chlorophyll c. Shaded areas represent one standard deviation from the mean (HL, n = 4; LL, n = 3). (c) Cell‐specific P _O versus QF curve. (d) Fraction of closed reaction centers (1 – qL) versus QF curve. (e) Chlorophyll fluorescence parameters versus quantum flux for cells acclimated to low light. (f) Chlorophyll fluorescence parameters versus quantum flux for cells acclimated to high light. Vertical dashed lines represent the mean quantum flux received by the cultures at the experimental irradiance. Abbreviations and definitions: LL, low light; HL, high light; QF, quantum flux; Y(II), quantum efficiency of photosystem II; NPQ, non‐photochemical quenching; Y(NO), unregulated, non‐radiative dissipation of excitation energy. Data based on n = 3 biological replicates for LL and n = 4 biological replicates for HL.

Photophysiology of P. tricornutum at low and high light

Using a rapid light curve (RLC) protocol (Jallet, Caballero, et al., ²⁰¹⁶), we concurrently determined chlorophyll fluorescence parameters and oxygen evolution. Model parameters include the photon uptake rate (quantum flux, QF) and the oxygen evolution rate (P _O), similar to recent modeling efforts in cyanobacteria and diatoms (Broddrick, Du, et al., ²⁰¹⁹; Broddrick, Welkie, et al., ²⁰¹⁹). We report both the maximum QF (QF _max) and the mean QF (QF _mean) representative of the highest and the average photon capture rates, accounting for cellular self‐shading across the full culture vessel path length. This calculation was also performed for the PAM sample cuvette as these measurements required high cell densities to generate sufficient fluorescence signal, resulting in significant self‐shading. QF _mean was then used as the independent variable for P _O and chlorophyll fluorescence parameters versus QF plots.

The growth rate differences between HL and LL acclimated cultures were largely attributed to differences in photophysiology between the two conditions. Despite the HL acclimated cells absorbing 3.3 times more photons than the LL acclimated cells, the HL maximum oxygen evolution rate was only 1.3‐fold higher (P _O max) (Table 1). However, this ratio increased to 1.7‐fold at the mean oxygen evolution rate (P _O mean), quantifying the impact of self‐shading from the increased pigment content at LL on overall productivity. The P _O versus QF curve initial slopes were 9.5 × 10⁻² and 9.8 × 10⁻² mol O₂ mol photon⁻¹ for HL and LL, respectively. The initial slope of the total chlorophyll (chla+chlc)‐normalized‐P _O versus PAR curves were also consistent [2.6 × 10⁻⁴ mol O₂ mol photon⁻¹ m² mgChl⁻¹ for both HL and LL Figure (S1)].

Table 1.

Comparison of photophysiology in Phaeodactlyum tricornutum acclimated to low and high light

	QF max a	QF mean a	P O max b	P O mean b	Fv/Fm	Y(II) c	Y(NPQ) c	1 – qL c	P O max (HL:LL)	P O mean (HL:LL)
Low light	0.23	0.15	0.014 ± 0.000	0.010 ± 0.000	0.68	0.63	0.00	0.25	1.3	1.7
High light	0.77	0.59	0.018 ± 0.002	0.017 ± 0.002	0.63	0.32	0.00	0.69

QF _max, maximum quantum flux received by the cross‐section of cells closest to the light source; QF _mean, the mean quantum flux absorbed across the entire culture path length; P _O max, the net oxygen evolution rate at QF _max; P _O mean, the net oxygen evolution rate at QF _mean; Fv/Fm – maximum quantum efficiency of PSII; Y(II), quantum yield of PSII at QF _mean; Y(NPQ), nonphotochemical quenching at QF _mean; 1 – qL, redox state of the plastoquinone pool at QF _mean.

^a fmol photons cell⁻¹ sec⁻¹.

^b fmol O₂ cell⁻¹ sec⁻¹.

^c Value at QF _mean.

Trends in the chlorophyll fluorescence parameters plotted against quantum flux (PAM versus QF) were similar between the two light regimes (Figure 1d–f). The effective quantum yield of PSII, Y(II), decreased rapidly with increasing QF for both light regimes. Non‐photochemical quenching [Y(NPQ)] dissipated almost no excitation energy at either HL or LL (Table 1; Figure 1e,f). Because Y(II) accounts for the fraction of QF performing photochemistry and Y(NPQ) is the fraction of QF lost as heat, the balance, Y(NO), accounts for the QF dissipated in unregulated, non‐radiative processes. Y(NO) accounted for approximately 68 and 37% of PSII‐directed excitation energy at QF _mean for HL and LL, respectively. The fraction of closed reaction centers (1 – qL) is a proxy for the redox state of the plastoquinone pool and it was almost identical across the entire QF range for HL and LL acclimated samples (Figure 1d). Additionally, under both light conditions, Y(NPQ) activated at a QF of approximately 0.7 fmol photons cell⁻¹ sec⁻¹, 1 – qL values of 0.70–0.75, and approximately 90% of their respective maximum photosynthetic rates (Figure S2).

We explored the possibility of carbon limitation in our RLC experiments. Ideally, the dissolved inorganic carbon (DIC) concentration at each timepoint in the RLC would be the same as the air sparged experimental culture. However, the RLC protocol adds variability to the DIC value as the cultures are reconstituted in fresh media. The fresh media does not capture the steady‐state DIC values of the air sparged culture, which is a combination of cellular metabolism (carbon fixation and respiration) and DIC replenishment from the sparged air. Additionally, the dark incubation prior to RLC measurements may increase the DIC in solution as the cells respire stored carbon in the dark. Finally, each step of the RLC changes the DIC concentration in the sealed sample vessel, potentially resulting in carbon limitation at higher irradiance values later in the RLC as the fixed amount of DIC is consumed by photosynthesis. To define the upper limit of P _O, we assessed the light capture efficiency with an excess of inorganic carbon by repeating the P _O versus QF and PAM experiments supplementing the samples with 5 mm bicarbonate. The resulting oxygen evolution was 15% higher at QF _mean for the HL acclimated condition, and unchanged for the LL condition (Figure S3a,b). At the same time, PAM results were comparable between samples with and without added bicarbonate (Figure S3c). These data suggest any carbon limitation was insufficient with respect to affecting quantum efficiency but may have affected the maximum photosynthetic rate.

Simulating photoautotrophic growth of P. tricornutum at low and high light

Photophysiology constraints, maintenance, and biomass composition

Accurate model simulations require the application of appropriate constraints. Previous photoautotrophic modeling efforts established oxygen evolution and photon uptake constraints yielding reasonable simulation predictions (Broddrick et al., ²⁰¹⁶; Broddrick, Du, et al., ²⁰¹⁹; Broddrick, Welkie, et al., ²⁰¹⁹). However, because of cell‐to‐cell shading, the biomass‐normalized uptake rate for light decreases as the biomass increases (assuming a fixed photon flux and culture volume). Thus, we simulated photoautotrophic growth at low and high light by translating the photophysiology results into modeling constraints, leveraging previous modeling methods (Broddrick, Du, et al., ²⁰¹⁹; Broddrick, Welkie, et al., ²⁰¹⁹) similar to dynamic flux balance analysis (Mahadevan et al., ²⁰⁰²). We used the genome‐scale model iLB1034 (Broddrick, Du, et al., ²⁰¹⁹), the biomass macromolecular composition of which uses previously determined values for P. tricornutum (Jallet, Caballero, et al., ²⁰¹⁶). We updated this biomass composition with the pigment fraction updated based on our current experimental data (Table S1) and added additional reactions to the model (Data S1). The final version used for simulations consisted of 1035 genes, 1720 metabolites, 2171 reactions, and six cellular compartments (cytosol, extracellular space, plastid, mitochondrion, peroxisome, and thylakoid).

The culture duration (12 h for HL, 24 h for LL) was segmented into 20‐min pseudo‐steady‐state intervals. At the beginning of each interval, the model was constrained, and biomass production was simulated over the 20‐min pseudo‐steady‐state period. The first constraint quantified the amount of light absorbed by the cells. This photon uptake constraint was the calculated quantum flux derived from the intersection of a*_cell (Figure 1a), the incident photon flux (the experimental PAR intensity and the emission spectrum of the fluorescent lamp used during culturing, Equation S9), the culture path length, and the cell density at the time of the simulation (Equation S7). This constraint placed quantitative bounds on the model's photon exchange reaction (model reaction: EX_photon_e), which is the ‘metabolite’ used for photochemistry at both PSI and PSII. Oxygen evolution is the calculated P _O rate at a given quantum flux (the photon uptake constraint) based on the experimental P _O versus QF curves (Figure 1c). This constraint placed a quantitative bound on the model's oxygen exchange reaction (model reaction: EX_o2_e). This simulation methodology naturally accounted for the decrease in photon uptake as increased biomass resulted in cell‐to‐cell shading (Figure S4a,b). Non‐growth associated maintenance (NGAM) was set to the dark respiration rates (0.83 and 0.34 mmol O₂ g DW⁻¹ h⁻¹ for HL and LL, respectively) because this value has been shown to result in accurate predictions of photoautotrophic growth (Broddrick, Du, et al., ²⁰¹⁹; Broddrick, Welkie, et al., ²⁰¹⁹).

Chlorophyll fluorescence as a constraint on photochemical yield

Next, we incorporated the chlorophyll fluorescence data with our modeling construct. For the system to maintain steady‐state flux balance, the GEM must account for all photons captured by the cells (total quantum flux). However, not all excitation energy performs photochemistry. Chlorophyll fluorescence parameters account for the excitation energy routed to PSII. The fraction of excitation energy allocated to photochemistry [Y(II)], regulated non‐photochemical quenching [Y(NPQ)], and unregulated non‐photochemical quenching [Y(NO)] can be formulated as canonical biochemical reaction stoichiometry. Because GEMs are built using biochemical reaction stoichiometry, we added a pseudo‐reaction to the model that imposes this fractionation between Y(II), Y(NPQ), and Y(NO) (Figure 2). Model simulations predict the excitation energy split between PSII and PSI to satisfy the reductant and ATP needs for biomass production. The chlorophyll fluorescence parameters apply a constraint on this excitation energy split because only the Y(II) fraction can perform photochemistry at PSII, which places an upper bound on the total flux through the photosynthetic electron transport chain. Y(NPQ) excitation energy is lost as heat; however, there are currently no mechanisms to link this heat loss to the broader metabolic network. Additionally, Y(NO) represents non‐radiative and fluorescence dissipation of excitation energy, the mechanisms of which are poorly characterized. Thus, both Y(NPQ) and Y(NO) are allowed to leave the system through simple demand reactions but are included as separate processes to allow future mechanistic incorporation into metabolic modeling.

Figure 2.

Incorporation of chlorophyll fluorescence parameters in the genome‐scale model.

Values for the pseudo‐reaction are representative of the high light acclimated samples. Abbreviations: hν, photon flux; QF, quantum flux; PSI_QF, quantum flux allocated to photosystem I; Fdx_ox/red, oxidized/reduced ferredoxin; PC_ox/red, oxidized/reduced plastocyanin; PQ, oxidized plastoquinone; PQH₂, reduced plastoquinone; Y(NO), unregulated non‐photochemical quenching of PSII allocated quantum flux; Y(NPQ), regulated non‐photochemical quenching of PSII allocated quantum flux; Y(II), photochemical quenching of PSII allocated quantum flux.

Photosystem II D1 subunit damage as a maintenance cost

An additional constraint added to the model accounted for photodamage of the PSII D1 subunit. The D1 subunit is the primary photoinhibition target in the photosynthetic ETC (Edelman & Mattoo, ²⁰⁰⁸), which is repaired by removal, degradation, and de novo synthesis (Nixon et al., ²⁰⁰⁵). To account for this repair cost, we determined the D1 damage rate at the experimental irradiance for both LL and HL acclimated cells by comparing the maximum quantum yield of PSII (F _v/F _m) with and without lincomycin, a plastid protein synthesis inhibitor (Figure S5a,b). We determined the D1 damage first‐order rate constant to be between 5 × 10⁻⁴ and 7 × 10⁻⁴ (n = 3) and 2.22 × 10⁻² and 2.52 × 10⁻² (n = 3) for LL and HL acclimated samples, respectively. Western blot analysis determined the D1 protein to be approximately 1.1% of total protein for both LL and HL. From these values, we calculated the D1 damage constraint to be 7 ± 2 × 10⁻⁶ and 2.53 ± 0.51 × 10⁻⁴ mmol D1 g DW⁻¹ h⁻¹ for LL and HL, respectively. These values were added as a maintenance energy reaction to the model (model reaction: NGAM_D1_u) requiring 360 mmol ATP and 720 mmol GTP to be consumed at the plastid ribosome to polymerize 1 mmol of D1 protein. The ATP requirement for this reaction was increased to 720 mmol ATP mmol D1⁻¹ to account for the cost to degrade the 360 amino acid‐long damaged D1 protein via an ATP‐dependent zinc metalloprotease, FtsH (Domingues et al., 2012; Nixon et al., ²⁰⁰⁵; Nymark et al., ²⁰⁰⁹). A summary of these values is provided in Table S2.

Constraining cyclic electron flow around photosystem I

Diatoms, to include P. tricornutum, have been shown to have minimal cyclic electron flow around PSI, limited to approximately 5% of linear electron transport (LET) (Bailleul et al., ²⁰¹⁵). To incorporate this physiological limitation, we simulated photoautotrophic growth at high photon fluxes [PAR: 1150 μmol photons m⁻² sec⁻¹ (QF: 1.27 fmol photons cell⁻¹ sec⁻¹) and PAR: 700 μmol photons m⁻² sec⁻¹ (QF: 1.94 fmol photons cell⁻¹ sec⁻¹) for HL and LL, respectively] to extract the model‐derived maximum LET [equivalent to the flux through PSI (model reaction: PSI_u)], with the flux bounds through the cyclic electron flow reaction (CEF_h) set to zero. The maximum LET was determined to be 27.0 and 44.8 mmol e^– g DW⁻¹ h⁻¹ for HL and LL, respectively. This resulted in maximum cyclic electron flow (CEF) rates of 1.4 and 2.2 mmol e^– g DW⁻¹ h⁻¹ for HL and LL, respectively. These values were used to set upper bounds on CEF flux in during model simulations.

Growth rate simulations

With the suite of photophysiology constraints incorporated into the GEM, we simulated photoautotrophic growth for both light conditions. To account for experimental variability on the model predictions, we simulated growth using a parameter space that included the mean and plus or minus one standard deviation of the experimentally determined a*_cell (Figure 1a), P _O versus QF curves (Figure 1c), and dry cell weight (Table S1). Simulated growth rates were consistent with experimental values (Figure 3). The model predicted a LL mean growth rate of 0.027 ± 0.002 h⁻¹ (n = 27 parameter combinations), in good agreement with the experimental range of 0.026–0.029 h⁻¹ (n = 3). For the HL condition, the model predicted a mean growth rate of 0.046 ± 0.004 h⁻¹ (n = 27 parameter combinations) compared to an experimental range of 0.052–0.053 h⁻¹ (n = 4). The predicted mean growth rate for the HL acclimated condition was underestimated by approximately 12%; however, adjusting the P _O versus QF curve based on the bicarbonate spiked data (Figure S3a) resulted in a simulated mean growth rate of 0.052 ± 0.005.

Figure 3.

Experimental versus simulated growth rates for P. tricornutum acclimated to low and high light.

Experimental values represent the growth rate for independent biological replicates (HL, n = 4; LL, n = 3). For the simulation values, the data points represent the simulated growth rate using the mean, +1 standard deviation and −1 standard deviation of the O₂ versus QF curves, a*_cell and cell dry weight (n = 27 parameter combinations). Abbreviations: Exp, experimental; Sim, simulated; +HCO₃ ⁻, simulated, bicarbonate spiked.

Hierarchy of photophysiology constraints for simulating photoautotrophy

We assessed the relative contribution of various constraints on the accuracy of model‐predicted growth rates, specifically the impact of photon uptake, oxygen evolution, D1 repair, and Y(II) on growth rate predictions (Table 2). The photon uptake and oxygen evolution constraints are described above (see section on ‘Photophysiology constraints, maintenance, and biomass composition’). The D1 repair constraint is a requirement for the model simulation to regenerate the PSII D1 subunit at a fixed rate (7 ± 2 × 10⁻⁶ and 2.53 ± 0.51 × 10⁻⁴ mmol D1 g DW⁻¹ h⁻¹ for LL and HL, respectively), as described above. This constraint places a quantitative bound on a non‐growth associated maintenance reaction in the model that represents the polymerization of 360 amino acids (i.e. the length of the P. tricornutum D1 polypeptide). The model reaction accounts for the 1 ATP required for degrading each amino acid in the damaged D1 protein as well as the 1 ATP and 2 GTP molecules consumed per amino acid polymerized at the chloroplast ribosome [total: 720 mmol ATP (mmol D1)⁻¹; 720 mmol GTP (mmol D1)⁻¹; model reaction: NGAM_D1_u]. The Y(II) constraint forces all photosynthetic excitation energy directed to PSII to be separated into one of three fates: photochemistry at PSII [Y(II)], NPQ, or NO. For a given quantum flux, the Y(II), Y(NPQ), and Y(NO) values are determined based on PAM versus QF experimental data (Figure 1e,f) and used to set the stoichiometric coefficients on the chlorophyll fluorescence pseudo‐reaction upstream of PSII (model reaction: PHO_PSIIt_u). This constrains PSII photochemistry to only the excitation energy in the Y(II) fraction.

Table 2.

Accuracy of growth rate predictions for different constraints. Simulations detailed in the first column describe the constraints that were included in the model for growth rate determination

Experimental	Growth rate [μ (h−1)]
	HL	LL
	0.052–0.053	0.026–0.029
Simulations
hν	0.138	0.038
P O	0.052	0.027
hν, P O, D1	0.052	0.027
hν, Y(II)	0.073	0.03
hν, P O, D1, Y(II)	0.052	0.027
hν, P O, DM20%	0.052	0.027

Constraints: hν, photon uptake; PO, oxygen evolution rate; D1, PSII D1 protein repair; Y(II), PSII quantum yield; DM20%, previous modeling assumption of 20% of absorbed photons lost upstream of the photosystems (Broddrick, Du, et al., 2019). Growth rates are in units of h−1.

For the LL simulations, the model was parameterized with the mean values for P _O versus QF, a*_cell, and dry cell weight. For the HL simulations, we used the mean values for all but P _O versus QF, where we used the bicarbonate spike‐adjusted data because it most accurately represented the growth rate of the experimental cultures (Figure 3). Constraining the photon uptake (hν) alone overestimated the growth rate for all conditions (+40 and +160% compared to experimental mean for LL and HL, respectively) (Table 2). This was expected because it assumes all the light captured by the cell is converted into photochemical energy. Using the oxygen evolution rate (P _O) as the sole constraint resulted in accurate growth rate predictions (within the range of experimentally observed growth rates for both LL and HL). Combining constraints on photon uptake, P _O, and D1 repair requirements, as well as photon uptake, P _O, D1 repair, and photochemical yield [Y(II)], did not change the predicted growth rates compared to P _O alone. Constraining the simulations with only the photon uptake and Y(II) resulted in an accurate prediction at LL (+5% compared to experimental mean). For HL acclimated cells, the prediction was more accurate than photon uptake alone but still over‐estimated the growth rate (+40% compared to experimental mean). The numerical values used to either set bounds on model reactions or the Y(II) stoichiometry in the PSII photon fractionation pseudo‐reaction (Figure 2) are reported in Figure S4a–g.

Constraining D1 repair and Y(II) affects metabolic pathway predictions in P. tricornutum

Next, we investigated how the D1 repair requirement and Y(II) constraints affected predictions of metabolic pathway usage. Intracellular reaction fluxes provide insight into the metabolic phenotype and form the basis of GEM‐based bioengineering strategies. For previous modeling efforts in P. tricornutum acclimated to high light, the predicted rate of intracellular oxygen consumption (defined as the difference between gross oxygen production at PSII and net oxygen evolution into the culture medium) was higher than experimental measurements (Broddrick, Du, et al., ²⁰¹⁹). The previous effort assumed that up to 20% of the photon flux could be dissipated upstream of the photosystems (hereafter the 20% assumption). Although this assumption did not affect the predicted growth rate (Table 2), we hypothesized the over‐estimation of intracellular oxygen consumption could affect metabolic pathway activation and absolute flux values. Additionally, D1 protein repair cost was not included in these previous simulations. Thus, we simulated photoautotrophic growth in P. tricornutum with both the previous 20% assumption and our experimentally derived Y(II) values and D1 damage rates [hereafter Y(II) constrained]; comparing the model predictions with experimental O₂ exchange values derived from membrane inlet mass spectrometry (MIMS).

Y(II) constraints increase the accuracy of intracellular oxygen consumption

We defined the model predicted intracellular oxygen consumption as the ratio of cellular oxygen evolution and the PSII oxygen generation rates. For the LL acclimated condition, both the Y(II) constrained and 20% assumption resulted in equivalent predictions of intracellular O₂ consumption [24–33% and 24–35% of total PSII O₂ generation for the Y(II) constrained versus the 20% assumption, respectively; ranges based on flux variability analysis]. These values were consistent with experimentally determined values in LL acclimated P. tricornutum [35 ± 5% (Broddrick, Du, et al., ²⁰¹⁹)]. For P. tricornutum acclimated to HL, the predicted intracellular oxygen consumption values were dramatically different between the 20% assumption and the Y(II) constrained simulation (63–67% and 40–46%, respectively; ranges based on flux variability analysis). Using MIMS, we experimentally determined the fraction of PSII O₂ generation consumed by light‐independent mechanisms (maintenance), consumed by light‐dependent mechanisms (EET), and evolved (net P _O) by cells acclimated to high light (600 μmol photons m⁻² sec⁻¹). For this HL condition, the MIMS experiment determined that 42 ± 5% of the PSII generated O₂ was consumed via intracellular consumption compared to the Y(II) constrained simulation predicted range of 40–46%. Additionally, the MIMS experiment determined 15 ± 3% of PSII generated O₂ was consumed via light‐independent mechanisms (maintenance) compared to the Y(II) constrained simulation predicted range of 13–14%. Finally, MIMS measured 26 ± 4% of PSII generated O₂ was consumed via light‐dependent mechanisms compared to the Y(II) constrained simulation predicted range of 26–32%. Overall, although constraining the model with chlorophyll fluorescence data did not affect growth rate predictions, it did result in highly accurate predictions of intracellular oxygen consumption. Table S3 summarizes the Y(II) constrained model predicted photosynthetic parameters.

Cross‐compartment metabolic coupling

Next, we investigated how incorporating chlorophyll fluorescence data affected predictions of cross‐compartment coupling. Previous modeling in P. tricornutum hypothesized excess reductant was shunted from the chloroplast to the mitochondrion (Broddrick, Du, et al., ²⁰¹⁹), consistent with experimental evidence of energetic coupling of these compartments (Bailleul et al., ²⁰¹⁵; Murik et al., ²⁰¹⁹). This previous work suggested photorespiration, branch‐chain amino acids, and an ornithine‐mediated chloroplast‐mitochondrion shunt were the dominant mechanisms for cross‐compartment coupling (Broddrick, Du, et al., ²⁰¹⁹). However, the Y(II) constrained model, validated with the O₂ values from the MIMS experiment, suggested that P. tricornutum acclimated to high light has a substantially reduced effective quantum yield at the experimental irradiance [Y(II)] (Table 1). Thus, excess excitation energy not performing photochemistry (e.g. NPQ) likely reduces the need for cross‐compartment coupling.

We investigated this hypothesis by comparing model predicted intracellular fluxes using the 20% assumption and the Y(II) constrained models. First, we determined the total EET in the system, defined as the excitation energy captured in excess of biomass and maintenance requirements (units: mmol electrons g DW⁻¹ h⁻¹), as a function of QF for both the 20% assumption and the Y(II) constraints. The results reflected the inaccuracy of the 20% assumption for both LL and HL conditions (Figure 4a). The biomass‐normalized EET flux was consistent between LL and HL conditions, and both sets of constraints up to a QF value of approximately 0.3 fmol photons cell⁻¹ sec⁻¹. For the 20% assumption, the model‐predicted EET flux continued to increase linearly with QF. By contrast, EET in the Y(II) constrained model began to plateau (Figure 4a), reflecting the decrease in Y(II) and increase in NPQ as QF increased (Figure 1e,f).

Figure 4.

Chlorophyll fluorescence and D1 damage constraints affects model predictions for cross‐compartment metabolic coupling.

(a) Predicted EET as a function of quantum flux for cells acclimated to LL (triangles) or HL (circles). Open markers: simulations where a fixed 20% of captured photons are lost upstream of the photosystems; filled markers: simulations with Y(II) and D1 repair constraints. Vertical dashed lines represent the mean quantum flux received by the cultures at the experimental irradiance. (b) Total metabolic flux shunted to the mitochondrion via different metabolic pathways for P. tricornutum acclimated to high light. Black bars: Simulations where a fixed 20% of captured photons are lost upstream of the photosystems; white bars: simulations with Y(II) and D1 repair constraints; gray bars: simulations with Y(II) and D1 repair constraints and NGAM routed to PTOX. Abbreviations: AOX, alternative oxidase; BCAA branched‐chain amino acid; Glyco_m, mitochondrial glycolysis; Orn, ornithine shunt; PR, photorespiration (reducing, glycine cleavage system; oxidizing, glyoxylate transaminases); PTOX, plastid terminal oxidase; ROS, reactive oxygen species detoxification; TCA, mitochondrial tricarboxylic acid cycle.

Next, we investigated the intracellular EET pathways connecting the plastid and mitochondrion for cells acclimated to HL. The 20% assumption predicted 107% more photosynthetically derived electrons were shuttled to the mitochondria compared to the Y(II) constrained simulation. This difference in cross‐compartment energetic coupling resulted in similar cross‐compartment shuttles, as previously reported (Broddrick, Du, et al., ²⁰¹⁹); however, the absolute flux through these pathways was altered (Figure 4b). The Y(II) constrained simulations predicted decreased flux through all pathways, apart from mitochondrial glycolysis, compared to the 20% assumption simulations. We also explored whether the compartmentalization of light‐independent O₂ consumption (NGAM) affected EET predictions because subcellular location of the NGAM reaction can change metabolic flux predictions. Our NGAM equation was formulated as a mitochondrial‐targeted quinone oxidase. We changed the NGAM to a plastid terminal oxidase (PTOX), a thylakoid membrane‐localized plastoquinone oxidase. Simulations with PTOX as the NGAM reaction resulted in a slight decrease in absolute flux values predicted in mitochondrial EET pathways commensurate with the reduction in photosynthetically derived electrons leaving the plastid ETC. However, the overall trends were consistent with mitochondrial targeted NGAM simulations. An unexplored, potential cross‐compartment shuttle suggested by these simulations was the amino acid lysine. However, lysine catabolic flux in the mitochondrion was similar for all conditions, suggesting that this pathway may not be used for EET.

Our simulation predictions suggested three routes for plastid‐derived reductant consumption: the mitochondrial alternative oxidase (AOX), scavenging of reactive oxygen species (ROS), and conversion of reduced carbon skeletons (glutamate and alanine) to more oxidized forms (alpha ketoglutarate and pyruvate) during glyoxylate transaminase reactions. These last two categories function to detoxify glycolate produced via photorespiration. A unique feature of photorespiration is that it performs both reduction and oxidation of mitochondrial cofactors. Glycine produced by transaminase reactions during the detoxification of glycolate was predicted to be consumed by the glycine cleavage system producing NADH. This reductant then helped fuel the AOX reaction. This linkage between photorespiration, ROS, and AOX is consistent with studies showing AOX to be activated by ROS stress and important in maintaining redox balance in P. tricornutum (Murik et al., ²⁰¹⁹).

Model‐based exploration of bioengineering potential

Biomass macromolecule reductant cost and EET

GEMs account for every known biochemical reaction in the organism and can calculate accurate assessments of resource requirements for biomass components and bioproducts (Dinh et al., ²⁰¹⁸). We calculated the fraction of linear electron transport (LET) used to biosynthesize each biomass macromolecular fraction (Table 3). The results provided insight into the relative reductant cost of each macromolecular component, which ranged from 177 mmol e⁻ g DW⁻¹ of carbohydrate to 411 mmol e⁻ g DW⁻¹ of membrane lipids. Using these values and the amount of excess photosynthetically generated reductant in the system, we calculated the theoretical yield of different biomass components if that EET could be rerouted for biomass biosynthesis. The predictions were inversely related to reductant cost with 20.0 mg carbohydrates g DW⁻¹ h⁻¹ being the highest yield and membrane lipids the lowest at 8.6 mg lipids g DW⁻¹ h⁻¹ (Table 3).

Table 3.

Photosynthetically generated electron requirements for different biomass components for cells acclimated to high light

Biomass component	Biomass percent	%LET	Reductant cost c	Theoretical yield d
Protein	71.5	43.9	299	23.7
Structural carb	6.1	2.2	176	40.3
DNA	0.3	0.1	163	43.5
Membrane lipids	3.1	2.6	409	17.3
Pigments	2.2	1.7	377	18.8
Plastid lipids	3.8	2.7	346	20.5
RNA	2.8	1.4	244	29.0
Storage a	10.2	5.3	253	28.0
EET	NA	28.1	NA	NA
Other b	NA	12.0	NA	NA

EET, alternative electron transport; LET, linear electron flow; NA, not applicable.

^a 2.1–1.0 molar ratio of a β‐1,3‐glucan and triacylglycerol (16:1(9Z)/16:1(9Z)/16:0).

^b Maintenance, vitamins, and cofactors.

^c Millimole photosynthetically generated electrons per gram dry weight of molecule class (mmol e⁻ g DW⁻¹).

^d Milligram dry weight component per gram dry weight biomass per hour (mg g DW⁻¹ h⁻¹).

Designing photoautotrophic bioproduction strategies

Intracellular metabolic rewiring for bioproduction

Finally, we used the model in the design step of the design–build–test–learn (DBTL) (Carbonell et al., ²⁰¹⁸) paradigm towards model‐driven engineering strategies to produce high value bioproducts. Bioengineering of photosynthetic microbes has typically targeted fuel or nutraceuticals. These targets are derived from three different precursors: fatty acids, aromatic amino acids, and terpenoids (Brey et al., ²⁰²⁰; Kumar et al., ²⁰²⁰). We evaluated engineering intracellular pathways to increase flux through plastid fatty acid biosynthesis (hexadecanoate), the shikimate pathway (chorismate), and isoprenoid precursors (isopentenyl pyrophosphate). In our simulation, downregulation of EET provided extra reductant for bioproducts; however, carbon and other elements were also diverted from other biomass components. We simulated diverting up to 50% of cellular biomass to bioproduct synthesis, in increments of 10%, and evaluated changes in the intracellular reaction fluxes that could enable light‐drive production of these compounds (Figure 5).

Figure 5.

Metabolic engineering potential of P. tricornutum acclimated to high light.

Changes in metabolic reaction flux towards the bioproducts hexadecanoate, isopentenyl pyrophosphate and chorismate are shown on the flux map. The heatmap above the reactions indicate an increase or decrease in flux towards chorismaI (c), hexadecanoate (h), or isopentenyl pyrophosphate (i). Values represent the difference between the baseline simulation fluxes [Y(II) constrained] and bioproduct formation with 30% of biomass rerouted to the desired product. Graphs indicate bioproduct yield as a function of %biomass diverted as well as the carbon‐normalized yields. Abbreviations are based on the BiGG Models database (King et al., ²⁰¹⁶). Abbreviations (reactions): ACCOAC, acetyl‐CoA carboxylase; ENO, enolase; FBA, fructose 1,6‐bisphosphate aldolase; FBP, fructose 1,6‐bisphosphase; GAPDH, glyceraldehyde 3‐phosphate dehydrogenase; PDH, pyruvate dehydrogenase; PGAM, phosphoglycerate mutase; PGK, phosphoglycerate kinase; PYK, pyruvate kinase; RUBISC, ribulose‐1,5‐bisphophate carboxylase; RUBISO, ribulose‐1,5‐bisphophate oxygenase; TKL2, transketolase 2. Abbreviations (metabolites): 13dpg, d‐glycerate 1,3‐diphosphate; 2 pg, 2‐phospho‐d‐glycerate; 2pglyc, 2‐phosphoglycolate; 3 pg, 3‐phospho‐d‐glycerate; 3psme, 5‐O‐(1‐carboxyvinyl)‐3‐phosphoshikimate; accoa, acetyl‐CoA; co2, carbon dioxide; dhap, dihydroxyacetone phosphate; e4p, d‐erythrose‐4‐phosphate; f6p, beta‐d‐fructose 6‐bisphosphate; fdp, beta‐d‐fructose 1,6‐bisphosphate; g3p, glyceraldehyde 3‐phosphate; malcoa, malonyl‐CoA; o2, oxygen; pep, phosphoenolpyruvate; pyr, pyruvate; rb15bp, ribulose 1,5‐bisphosphate; skm5p, shikimate 5‐phosphate; xu5p, d‐xyulose 5‐phosphate.

The model predicted a linear increase in product yield as a function of increased biomass diverted to bioproducts. Production rates were 0.15, 0.50, and 0.37 mmol bioproduct g DW⁻¹ h⁻¹ fraction_biomass ⁻¹ for hexadecanoate, isopentenyl pyrophosphate, and chorismate, respectively. When normalized to the number of carbons in each of these end products, the yields were 2.4, 2.5, and 3.7 mmol C fixed in bioproduct g DW⁻¹ h⁻¹ fraction_biomass ⁻¹ for hexadecanoate, isopentenyl pyrophosphate, and chorismate, respectively. There were no major differences in predicted EET as a result of bioproduct synthesis. We compared the baseline EET [Y(II) constrained simulations] to the EET of the production strains with 30% of biomass diverted to bioproduct synthesis. The baseline EET was 3.13 mmol e⁻ g DW⁻¹ h⁻¹ compared to 3.08, 3.21, and 3.09 mmol e⁻ g DW⁻¹ h⁻¹ for hexadecanoate, isopentenyl pyrophosphate, and chorismate, respectively.

The flux simulations identified metabolic pathways where rerouting of flux is required for bioproduct synthesis. We compared model‐predicted metabolic flux routing at an intermediate biomass diversion value of 30%. All three metabolites, hexadecanoate, isopentenyl pyrophosphate, and chorismate, require plastid glycolytic precursors (acetyl‐CoA, pyruvate, and phosphoenolpyruvate, respectively). Additionally, chorismate and isopentenyl pyrophosphate both require CBBC intermediates (d‐erythrose‐4‐phosphate and glyceraldehyde‐3‐phosphate, respectively) for biosynthesis. These requirements were evident in the reaction flux differences between the reference simulation [Y(II) constrained] and the bioproduct simulations (Figure 5). Hexadecanoate biosynthesis required the largest flux rerouting through lower plastid glycolysis as all the carbon required for its biosynthesis is sourced from acetyl‐CoA. For chorismate, six of its 10 carbons come from lower glycolysis, requiring increased flux through the reactions phosphoglycerate mutase and enolase. The remaining four carbons are sourced through the CBBC resulting in a slight increase in flux through these reactions, to include a predicted increase in carbon fixation at Rubisco. Although initiation of isopentenyl pyrophosphate biosynthesis utilizes the lower glycolytic metabolite pyruvate, the model predicted that the source of this metabolite was recycling of carbon from plastid–mitochondrial metabolic coupling, not redirection of flux away from the CBBC. Our previous modeling in P. tricornutum suggested that the metabolites carrying photosynthetically derived reductant, delivered to the mitochondrion for oxidation via EET pathways, were returned to the plastid as pyruvate (Broddrick, Du, et al., ²⁰¹⁹). Our simulations here suggest that isopentenyl pyrophosphate biosynthesis can use this returning pyruvate to initiate product formation. This result shows how GEMs can result in non‐intuitive flux routing towards engineering bioproducts.

Optimal productivity depends on tradeoffs in culture density and light penetration

Designing aqueous photoautotrophic engineering strategies requires an assessment of how product yield and cell‐to‐cell shading are inversely affected by culture density. Low density cultures are optically thin, maximizing light harvesting, but there are fewer cells producing the desired product. By contrast, dense cultures can have orders of magnitude more individual cellular factories, but poor light penetration may limit productivity. We explored this tradeoff for theoretical cultures engineered to produce isopentenyl diphosphate, chorismate, or hexadecanoate with either 20 or 60% of biomass diverted to bioproduction under high and low light conditions (Figure 6).

Figure 6.

Biproduct yield versus time and inoculation density.

Bioproduction envelope for a theoretical culture gown for 5 days at (a) 600 μmol photons m⁻² sec⁻¹ or (b) 60 μmol photons m⁻² sec⁻¹ with 60% of biomass diverted to isopentenyl diphosphate production and varying inoculation densities. Comparison of cumulative production versus inoculation density after 5 days for cultures with either 60 or 20% of biomass diverted to one of three bioproducts. (c) 600 μmol photons m⁻² sec⁻¹, (d) 60 μmol photons m⁻² sec⁻¹. Abbreviations: ipdp, isopentenyl diphosphate; chor, chorismate; hdca, hexadecanoate.

Tradeoffs in light penetration and culture density were predicted to drive overall productivity. For high light acclimated cultures diverting 60% of biomass to the bioproduct, a maximum productivity of 1.69 g of isopentenyl diphosphate (28.2 mg %biomass⁻¹) was achieved at an inoculation density of 4 × 10⁷ cells ml⁻¹ after 5 days of growth (Figure 6a). This contrasts with the low light acclimated cultures which achieved a maximum productivity of 0.22 g isopentenyl diphosphate (3.7 mg %biomass⁻¹) at an inoculation density of 1.2 × 10⁷ cells ml⁻¹ (Figure 6b). Additionally, for the LL acclimated cultures, there was insufficient light penetration to overcome maintenance requirements at inoculation densities above approximately 6 × 10⁷ cells ml⁻¹; in contrast to 2 × 10⁸ cells ml⁻¹ for HL acclimated cultures (Figure 6a,b).

The fraction of biomass diverted also affected overall productivity and optimal inoculation density. For both light levels, and all three bioproducts, when 20% of the biomass was diverted, both the maximum productivity (normalized to %biomass diverted for comparison) and the optimal inoculation density decreased compared to 60% biomass diversion (Figure 6c,d). This result is linked to the increased fraction of resources going into cell division, as opposed to bioproduction, which increased the rate of light attenuation because of cell shading. Despite the linear relationship between bioproduct yield and %biomass diverted at a single time point (Figure 5); these results captured the non‐linear dynamics of light penetration over time as a function of culture density, which is relevant to bioprocess design.

DISCUSSION

Cell and photophysiology

In the present study, we characterized photoautotrophic metabolism in P. tricornutum through integrated chlorophyll fluorescence measurements and genome‐scale modeling. Our observations in P. tricornutum were consistent with photophysiology under fluctuating and sinusoidal light (Wagner et al., ²⁰⁰⁶) and photoacclimation (Nymark et al., ²⁰⁰⁹). P. tricornutum exhibited efficient photoacclimation with the quanta absorbed per pigment remaining consistent between LL and HL (Figure 1b). This efficiency was also observed when looking at the initial slope of the cell‐normalized P _O versus QF curves (Figure 1c) and the chlorophyll‐normalized P _O versus PAR curves (Figure S1), which were consistent between both LL and HL cultures. This efficiency across a range of photoacclimation conditions likely contributes to the ecological success of diatoms in dynamic light environments (Behrenfeld et al., ²⁰²¹).

Our interpretation of photophysiology was heavily influenced by analyzing the P _o and PAM data with QF as the independent variable. The 1 – qL versus QF curves (Figure 1d), the shape of the chlorophyll fluorescence parameters versus QF curves (Figure 1e,f), and the D1 content as a fraction of total protein (Table S2) were consistent between LL and HL. Contrasting PAM versus QF (Figure 1e,f) with PAM versus PAR (Figure S6a,b) illustrates how interpreting photophysiology from a QF perspective affects conclusions about photophysiology.

We observed very little dissipation of excitation energy via NPQ at the experimental QF values (Figure 1e,f). For the HL conditions, this lack of NPQ was coupled with an effective quantum yield of PSII [Y(II)] value of 0.32 (Table 1), suggesting the presence of alternative dissipative mechanisms [Y(NO)]. NPQ is an important excitation energy dissipation mechanism under dynamic light conditions (Lavaud et al., ²⁰⁰²; Olaizola et al., ¹⁹⁹⁴; Wagner et al., ²⁰⁰⁶); however, our results suggest these other dissipative mechanisms are sufficient to prevent photoinhibition under stable light environments. Overall, these data suggest P. tricornutum employs a photoacclimation strategy emphasizing rapid utilization and dissipation of light energy. At times, the overall photosynthetic apparatus is under‐utilized (e.g. LL acclimated cultures). However, consistency in total PSII content per cell and the redox state of the plastoquinone pool versus quantum flux (Figure S2a,b) suggest that P. tricornutum can immediately respond to an increase in available photon flux without the need to biosynthesize additional macromolecules.

Modeling photoautotrophy and the hierarchy of constraints

Translating the QF to a photon uptake constraint and P _O into an oxygen evolution constraint in the GEM resulted in accurate predictions of photoautotrophic growth (Figure 3). Growth rate under the HL condition was underestimated by 12%, likely as a result of the beginning of carbon limitation in the sample during short‐term measurements of photosynthetic capacity (Figure S3a). We chose not to spike in exogenous bicarbonate for our oxygen evolution measurements as we were interested in measuring photosynthetic parameters relevant to our culturing conditions. However, our HL growth rate underestimation suggests the rapid light curve protocol did not completely recapitulate the experimental culture carbon environment. At the same time, including several mM NaHCO₃ ⁻ in these assays to avoid carbon limitation, as is standard protocol, likely overestimates the available inorganic carbon for photosynthesis in an air sparged experimental culture. Future modeling efforts need to address these factors to ensure accurate simulation parameters.

From our efforts to establish the hierarchy of constraints, it was clear oxygen evolution (P _O) is the dominant constraint on the system, resulting in accurate growth rate predictions as the sole constraint. Water splitting at PSII is the only reaction in the model with net oxygen production and is stoichiometrically coupled to photosynthetically derived electrons. All oxygen consuming reactions in the model are linked to electron consumption either directly (e.g. oxidoreductases, epoxidases, fatty acid desaturases, etc.) or indirectly (e.g. photorespiration of 2‐phosphoglycolate, hydrogen peroxide detoxification, etc.). Thus, net cellular oxygen evolution is proportional to the amount of photosynthetically derived reductant that can form biomass macromolecular components. Additionally, the difference between gross oxygen production at PSII and net cellular oxygen evolution is proportional to EET, which was a primary motivator for refining PSII flux via inclusion of chlorophyll fluorescence parameters. Still, it should be noted that P _O alone did not result in accurate predictions of intracellular oxygen dynamics, as seen in the modeling analysis of the MIMS data.

The D1 damage constraint did not affect the predicted growth rate. This is likely a result of the excess EET in the system, which could be rerouted to nucleotide triphosphate (ATP and GTP) biosynthesis, to include mitochondrial ATP synthesis, if necessary, to satisfy the D1 degradation and biosynthetic costs. The NTP maintenance cost (GTP and ATP) was calculated to be 0.36 ± 0.07 and 1 ± 0.2 × 10⁻² mmol NTP g DW⁻¹ h⁻¹ at high light and low light, respectively (Table S2). This is likely a conservative value because it assumes 100% of the amino acids from the damaged D1 protein can be reused after proteolysis. Still, we had to increase the D1 damage constraint at high light to > 10.0 mmol NTP g DW⁻¹ h⁻¹ and >2.1 mmol NTP g DW⁻¹ h⁻¹ at low light before EET was exhausted and a decrease in the predicted growth rate was observed. However, the D1 damage constraint requires the model to deliver GTP to the plastid, which affects predictions regarding intracellular metabolic pathway activation. Thus, although this constraint may not affect predictions of growth rate, it likely increases the accuracy of metabolic reaction flux predictions.

An important finding from the hierarchy was that photon uptake and chlorophyll fluorescence constraints alone accurately predict growth rate at low acclimation irradiances (Table 2). It is possible that, when the maximum photosynthetic rate is light limited, as is likely with our low light acclimated samples, accurately constraining the total photon input into the system results in a similar system‐level constraint as the oxygen evolution constraint. However, when the maximum photosynthetic rate is limited by metabolic reaction kinetics, such as the carbon fixation rate, photon uptake and chlorophyll fluorescence constraints no longer mirror the P _O constraint. This may explain why this set of constraints overestimated the growth rate of our high light acclimated samples. This is a testable hypothesis and should be explored in future modeling efforts. Still, accurate predictions of biomass accumulation under light limited conditions opens the possibility for non‐invasive monitoring of culture health and productivity for biotechnology applications. A passive sampling window with integrated chlorophyll fluorescence and spectral absorption analysis, coupled with data on surface irradiance, would recreate the chlorophyll fluorescence and photon uptake constraints. Because bioproduction conditions are usually high density, resulting in low cell‐specific quantum flux, integrating non‐invasive sampling with our modeling construct could result in accurate measurements of photoautotrophic metabolism in these settings.

Chlorophyll fluorescence constraints increased the accuracy of intracellular metabolic processes

Constraining biomass accumulation with QF and P _O automatically predicted relevant photosynthetic parameters in a manner similar to previous investigations (Jakob et al., ²⁰⁰⁷; Wagner et al., ²⁰⁰⁶). Importantly, GEMs also predict the optimal distribution of excitation energy between PSI and PSII, an advantage compared to previous work where it was assumed that 50% of the absorbed quanta were directed to each photosystem. Our simulations predicted a two‐fold increase in excitation energy utilized by PSII under HL compared to LL, with approximately 76% of absorbed photons directed to PSII. However, there was a similar number of charge separation events at both photosystems (Table S3). This is consistent with the observation that there is minimal cyclic electron flow (CEF) around PSI in P. tricornutum (Bailleul et al., ²⁰¹⁵), which would result in roughly equivalent charge separations at both photosystems. Additionally, although we constrained the upper bound of CEF flux based on experimental values (Bailleul et al., ²⁰¹⁵), the simulation results suggested this constraint was redundant. CEF flux at both HL and LL (0.55 and 0.67 mmol e^– g DW⁻¹ h⁻¹ for HL and LL, respectively) were below the maximum CEF rates (1.4 and 2.2 mmol e^– g DW⁻¹ h⁻¹ for HL and LL, respectively). It is likely that the bifurcation of excitation energy between PSII and PSI, and, thus, CEF, is captured by the QF, P _O, and chlorophyll fluorescence constraints.

Our model derived ETR differs from methodologies that assume equal excitation energy routed to both photosystems and a lower predicted quantum demand at experimental QF values (Table S3). For an organism such as P. tricornutum that does not employ extensive CEF (Bailleul et al., ²⁰¹⁵), the advantage of this approach is diminished. However, for microalgae, cyanobacteria, or plants that dynamically reroute excitation energy between the photosystems and adjust their biomass macromolecular composition as part of their photoacclimation strategy [e.g. Chlamydomonas reinhardtii (Davis et al., ²⁰¹³; Lucker & Kramer, ²⁰¹³; Walker et al., ²⁰²⁰)], implicit calculation of excitation routing between photosystems using our modeling framework will result in better approximations of ETR.

Incorporating chlorophyll fluorescence as a GEM constraint increased the accuracy of model predictions. Capturing photon loss upstream of the photosystems recapitulated intracellular oxygen production and reductant‐mediated oxygen consumption rates. This is likely because significant excitation energy is lost as Y(NO), especially at high light. Without the chlorophyll fluorescence constraint, the model would route that excitation energy to photochemistry, increasing LET and requiring more intracellular electron and oxygen consumption via EET to not violate the oxygen evolution constraint. As a result, these new constraints resulted in accurate predictions of intracellular oxygen consumption (validated by MIMS), and affected predictions of excess reductant in the system (EET), as well as predictions regarding cross‐compartment metabolic coupling (Figure 4a,b).

The plateau in EET flux as a function of QF (Figure 4a) was correlated with NPQ activation (Figure 1f), suggesting that saturation of EET pathways triggers this photoprotective mechanism. Our results suggest the following model of photophysiology in P. tricornutum in a stable light environment: up to a QF of approximately 0.3 fmol photons cell⁻¹ sec⁻¹, Y(NO) dissipates excess excitation energy until the steady‐state reduction of the plastoquinone pool is approximately 50%. At this point, EET is activated to facilitate re‐oxidation of the photosynthetic electron transport chain. At a QF of approximately 0.6 fmol photons cell⁻¹ sec⁻¹, the EET pathways are saturated, the steady‐state reduction of the plastoquinone pool reaches 70%, and NPQ is activated to assist in dissipating captured photons. Therefore, the model accurately recreates the onset of NPQ that occurs when light absorption outpaces the ability to utilize this energy within metabolism. Interestingly, this model is consistent for cells acclimated to both LL and HL, and Phaeodactylum is known to maintain high capacity for NPQ under both light conditions (Taddei et al., ²⁰¹⁸).

Currently, experimentally derived photoautotrophic metabolic flux values for P. tricornutum do not exist; thus, our flux predictions are hypotheses that still require validation. Still, the approach outlined in this study is generally applicable to all phototrophic genome‐scale model simulations and previous efforts using experimentally‐derived electron transport efficiencies, as opposed to PAM, showed good agreement with ¹³C metabolic flux analysis (Broddrick, Welkie, et al., ²⁰¹⁹).

The interest in integrating light‐driven metabolism into bioengineering and synthetic biology necessitates an iterative framework for bioprocess development. The DBTL paradigm has enabled rapid increases in bioproduct titers (Carbonell et al., ²⁰¹⁸; Petzold et al., ²⁰¹⁵). Computational tools are integral to these workflows. Relevant to the design step of the process, GEMs have been extensively used to rationally engineer metabolism to generate a wide variety of phenotypes (Bang et al., ²⁰²⁰; Czajka et al., ²⁰²¹; Li et al., ²⁰¹⁹). We explored implementing the modeling framework towards light‐driven bioproduct formation, as adoption of these approaches is underrepresented in phototrophic systems. Quantifying the reductant cost of cellular macromolecules provided insight into the theoretical yield of different compound classes (Table 3), with protein biosynthesis being the predominant energy sink. EET pathways, which were found to consume 29% of LET in HL, serve to oxidize the photosynthetic electron transport chain and resupply low energy cofactors to autotrophic metabolism. However, they are generally viewed as wasteful as this energy could be utilized to increase biomass yields (Peers, ²⁰¹⁴). We utilized our model to estimate the metabolic consequences of redirecting metabolism towards important chemical precursors. The overproduction of plastid fatty acids (hexadecanoate), the shikimate pathway (chorismate), and isoprenoid precursors (isopentenyl pyrophosphate) were only partially fueled by reductant that would normally be dissipated by EET, showing there is still considerable potential to engineer primary photosynthetic metabolism to increase bioproducts yields. However, the linearity of the product formation versus fraction of redirected biomass (Figure 5) suggests the system is carbon limited, not reductant limited. Thus, increasing the available DIC during culturing may result in synergistic flux modes where the EET is more efficiently utilized for bioproduction without a linear decrease in biomass production.

Furthermore, our modeling suggests that increasing the flux of reduced carbon to metabolic precursors of interest may require significant engineering of central carbon metabolism. For instance, increasing the production of our three selected metabolites increased the flux of 3‐phospho‐d‐glycerate (3 pg) through the plastid glycolytic pathway (Figure 5). Additionally, the increased flux of carbon to isoprenoid biosynthesis or through the shikimate pathway increased the demand for CBBC sourced biosynthetic intermediates (glyceralde‐3‐phosphate and d‐erythrose‐4‐phosphate, respectively). Photosynthetic microorganisms, including diatoms, redirect carbon and energy to storage molecules under conditions of nutrient deprivation, such as nitrogen limitation (Alipanah et al., ²⁰¹⁵; Park et al., ²⁰¹⁵). This phenotype forms the basis for much of the interest in biofuel applications of these phototrophs (Levering et al., ²⁰¹⁵). Thus, a reasonable strategy and potential future direction, is the diel separation of carbon fixation and bioproduct formation. Rerouting these metabolites from sugar polymer degradation via the pentose phosphate pathway or mitochondrial β‐oxidation of lipids (Jallet et al., ²⁰²⁰) may alleviate some of the pressure on the CBBC.

While pathway engineering alone is often sufficient for successful bioengineering of heterotrophic microorganisms, photoautotrophic systems must account for additional design parameters, with light availability being of paramount importance. Our simulations demonstrated the utility of assessing the impact of light environment and inoculation density on overall productivity (Figure 6). Light ultimately drives the formation of biomass and, expectedly, modeled yields of product are lower for cells inoculated in low light versus high light. This is likely attributable to the fact the light availability cannot overcome maintenance energy requirements of dense cultures. Focusing on the high light simulations, we observed that initial inoculation density has a major effect on the yield of desired bioproduct and that this is non‐linear (Figure 6c). Reduced yields at increasing inoculation density are due to self‐shading of the culture that reduce the amount of energy available for product formation. While our simulation held constant variables such as light level, inorganic carbon availability, and photophysiology, the modeling construct presented here can incorporate time dependent changes in these, as well as other variables, towards comprehensive bioprocess design.

The next step is to build and test these strains to initiate the first iteration of the DBTL cycle. During the test phase, the model constraints provide a roadmap for relevant process parameters and a framework to evaluate process performance, including assessments on EET usage efficiency. It is important to note our design simulations do not include possible changes in photophysiology as a result of strain engineering (e.g. an increase in P _O). However, physiological outputs from the testing of the strain designs proposed can be re‐integrated into the modeling framework to include changes in experimental constraints. This contributes to the learn step of the DBTL cycle, closing the loop on the first iteration and enabling an updated design strategy for the next iteration.

CONCLUSIONS

In the present study, we combined chlorophyll fluorescence parameters defining photosynthetic and non‐photosynthetic yield of absorbed light energy with a metabolic model of P. tricornutum. This integration increased the model predictive accuracy regarding growth rate, intracellular oxygen production and consumption, and metabolic pathway usage. Through the quantification of EET, we uncovered the sequential activation of non‐radiative energy dissipation processes, cross‐compartment electron shuttling, and non‐photochemical quenching as the rapid photoacclimation strategy in P. tricornutum. The photon absorption thresholds that triggered the transition between these mechanisms were consistent at low and high incident photon fluxes, providing insights into the mechanisms that drive the ecological success of this important class of primary producers. Quantification of EET allowed us to assess engineering strategies for rerouting cellular resources and excess light energy towards bioproducts. Taken together, our results show integrating relevant measurements of photosynthetic physiology with genome‐scale models results in quantitative predictions of condition‐specific phenotypes. This paves the way for iterative design and real‐time process control of photobioproduction platforms.

EXPERIMENTAL PROCEDURES

Cell culture and physiology

Culture conditions, cell physiology measurements and dimensions, and pigment extraction all followed standard procedures and the details can be found in Data S1.

Cellular absorption coefficients (a*_cell and a*_pigm)

Cellular absorption coefficients were determined based on published protocols (Moore et al., ¹⁹⁹⁵). Approximately 5 × 10⁷ cells were collected by filtration onto a GF/A glass microfiber filter (21 mm in diameter). The filter was cut to fit in a 1‐cm path length cuvette and placed against the inside of the cuvette. Absorbance spectra were collected used a Cary 60 UV–visible spectrophotometer (Agilent, Santa Clara, CA, USA) in scan mode (350–800 nm, scan interval of 1 nm). In total, four technical replicates were collected per sample and averaged. To decrease the noise from filter scattering, the absorbance spectra were smoothed using a Savitzky–Golay filter (width = 17 nm, polynomial = 2nd order) as implemented in scipy (Virtanen et al., ²⁰²⁰). The resulting spectra were blank subtracted against the appropriate media and normalized to an OD₇₅₀ value of 0. The wavelength specific absorption coefficient was determined, along with correcting for filter amplification according to:

$$a_{\lambda }=2.303\left(A\left({\mathrm{OD}}_{\lambda }\right)+B\left({\mathrm{OD}}_{\lambda }^2\right)\right)$$ (1)

where OD_λ is the absorbance at a given wavelength and A and B are species‐specific coefficients for the β‐factor correction (0.388 and 0.616 for A and B, respectively) (Finkel & Irwin, ²⁰⁰¹). The cell‐normalized absorption coefficient (a*_cell, units: cm² cell⁻¹) and the pigment‐normalized coefficient (a*_pigm, units: cm² μg⁻¹ pigments) were determined by dividing a _λ by either the total number of cells deposited on the filter or the total pigment mass, respectively, and then multiplying the resulting value by the filter area onto which the cells were deposited (2.1 cm² for the 21 mm diameter GF/A filter).

Simultaneous oxygen evolution and chlorophyll fluorescence parameters

RLCs were performed as outlined previously (Broddrick, Welkie, et al., ²⁰¹⁹; Jallet, Caballero, et al., ²⁰¹⁶). Briefly, a Dual PAM 100 fluorometer (Heinz Walz GmbH, Pfullingen, Germany) in a temperature controlled custom cuvette holder and a FireSting Optical Oxygen Meter (Pyro Science GmbH, Aachen, Germany) were used for the simultaneous measurement of chlorophyll fluorescence and oxygen evolution. Cells from approximately 30 ml of culture were pelleted by centrifugation (3000  g for 10 min at the experimental temperature). Cell pellets were resuspended in fresh media to the target cell density (HL: 2 × 10⁷ cells ml⁻¹, LL: 1 × 10⁷ cells ml⁻¹) and kept in the dark for 10 min prior to analysis. For select experiments, the cells were reconstituted in fresh media supplemented with 5 mm sodium bicarbonate. Dark respiration rates were collected for approximately 10 min prior to running RLCs. A red actinic light (635 nm) was used to provide a saturating pulse (600 ms, 10 000 μmol photons m⁻² sec⁻¹) for fluorescence measurements. Illumination steps and durations are provided in Data S1.

The chlorophyll fluorescence parameters F _v/F _m, Y(II), 1 – qL, and NPQ were determined as described (Kramer et al., ²⁰⁰⁴; Schreiber et al., ¹⁹⁹⁵). Shading in the round cuvette was accounted for by calculating the attenuation across the cuvette path length as described previously (Broddrick, Welkie, et al., 2019). Details are provided in Data S1. The resulting QF value was used as the independent variable in plots of oxygen‐based photosynthesis (P _O).

Time‐course measurements of oxygen evolution were exported from the FireSting O2 Logger software as a .txt file. The oxygen evolution rate was determined by taking the slope of the O₂ versus time plot for the each illumination step using the python scipy package (scipy.stats.linregress) (Virtanen et al., ²⁰²⁰) and normalized to cell counts. Dark period respiration rates were determined by taking the slope of the O₂ versus time plot for the last 2 min of the 10‐min dark acclimation period.

MIMS

Membrane inlet mass spectrometry was measured similarly to (Broddrick, Du, et al., ²⁰¹⁹) and (Ware et al., ²⁰²⁰). Samples corresponding to 4 μg chlorophyll a ml⁻¹ were collected and resuspended in fresh f/2 media. Next, 2 ml of culture was loaded into a cylindrical cuvette and oxygen evolution and consumption was measured using a quad mass spectrometer (PrismaPlus QMG220, Quadera v4.62; Pfeiffer Vacuum, Aßlar, Germany). Dissolved gas was pulled through a silicon membrane, connected to the mass spectrometer via a stainless‐steel tube with a vapor trap (ethanol and dry ice). Cells were bubbled with N₂ to deplete oxygen from the suspension to approximately 50% of atmospheric concentrations. The suspensions were injected with ¹⁸O₂ (catalog. no. 490474; Sigma‐Aldrich, St Louis, MO, USA) and mixed for 10–15 min until equilibration was achieved. ¹⁸O₂ was then purged from the sample using a stopper and gas consumption was measured in the dark for 5 min to calculate the respiration rate. Cells were then illuminated with a blue measuring light (Dual‐PAM‐100; Heinz Walz GmbH) to achieve a 0.2 V fluorescence signal for 15 min to relax photoprotective processes (Fisher et al., ²⁰²⁰). A white LED programmed to achieve 600 or 60 μmol photons m⁻² sec⁻¹ (measured with a ULM‐500, US‐SQS/L attachment; Heinz Walz GmbH) was used to illuminate cultures to their corresponding in situ light intensity for 7 min. The slopes of oxygen consumption (¹⁶O₂, m/z 32; ¹⁸O₂, m/z 36) and evolution (¹⁶O₂, m/z 32) were calculated according to Beckmann et al. (2009) using the mass charge (m/z) change on a per second basis, calculated over the last 30s of illumination. Argon (m/z 40) was used to normalize oxygen concentrations, minimizing the effects of pressure change and abiotic gas consumption (Bailleul et al., ²⁰¹⁵).

Genome‐scale metabolic modeling

Model and constraints

We used the P. tricornutum genome‐scale model (GEM) iLB1034 (Broddrick, Du, et al., ²⁰¹⁹) and simulations were performed in a similar manner to (Broddrick, Welkie, et al., ²⁰¹⁹). Model designators also denote the subcellular localization of the reaction or metabolite via a one‐letter code: cytosol (c), extracellular space (e), plastid (p), mitochondrion (m), peroxisome (x), and thylakoid (u). Details are provided in Data S1.

Integrating chlorophyll fluorescence measurements into the genome‐scale model

Chlorophyll fluorescence parameters were incorporated into the GEM based on the experimental values for Y(II), Y(NPQ), and Y(NO). Total photons captured by the system, equivalent to QF, was determined based on the incident photon flux irradiance, culture density, and cellular absorption coefficient, a*_cell (Equation S9). This photon fraction set the constraint on model photon exchange reaction (model reaction: EX_photon_e). These photons are allowed to enter the cytoplasm (model metabolite: photon_c) and then the plastid (model metabolite: photon_h) via transport reactions (model reactions: PHOt_e and PHOt_h for the cytosol and plastid, respectively). The model simulation then predicts the fraction of plastid photons (photon_h) routed to PSI and PSII, based on metabolic needs and stoichiometric constraints. For PSII, the fraction of absorbed photons available to perform photochemistry was constrained to account for chlorophyll fluorescence measurements via:

$$\mathrm{PHO}\_\text{PSIIt}\_\mathrm{u}:\text{photon}\_\mathrm{h}\to \mathrm{A}\text{photon}\_\mathrm{YII}\_\mathrm{u}+\mathrm{B}\text{photon}\_\text{YNPQ}\_\mathrm{u}+\mathrm{C}\text{photon}\_\mathrm{YNO}\_\mathrm{u}$$ (2)

where PHO_PSIIt_u is the model reaction name; photon_h is the excitation energy available to be routed to both PSII and PSI (QF); and A, B, and C are the fraction values of Y(II), Y(NPQ), and Y(NO), respectively, parameterized by the calculated QF value (Equation S9) and the experimental values for Y(II) and Y(NPQ) [Equations S12 and S13 for Y(II) at HL and LL, respectively; Equation S14 for Y(NPQ)]. Y(NO) was assumed to be the residual fraction of QF (Equation S15). The model PSII reaction, which performs water splitting (model reaction PSII_u), could only use the metabolite photon_YII_u (Equation (2) for charge separation, limiting photochemistry to the Y(II) fraction. The metabolites photon_YNPQ_u and photon_YNO_u were allowed to leave the system via demand reactions (model reactions DM_photon_YNPQ_u and DM_photon_YNO_u, respectively).

Determination and integration of damage to the D1 subunit of PSII

Because D1 damage is mitigated by de novo synthesis of a new subunit (Nixon et al., ²⁰⁰⁵), we determined the D1 damage rate with and without the plastid ribosome inhibitor, lincomycin, using the protocols of Campbell & Tyystjärvi (2012). Complete experimental details and modeling steps are provided in Data S1.

Genome‐scale model simulations

The simulation was performed by maximizing the biomass objective function reaction using the parsimonious FBA function (Lewis et al., ²⁰¹⁰) as implemented in cobrapy (Ebrahim et al., ²⁰¹³). Details for all simulations, including biomass accumulation, constraints evaluation, and methods used to formulate the bioengineering applications of our model, are provided in Data S1. All simulation codes, models, flux simulations, and metabolic maps are provided in Data S1.

AUTHOR CONTRIBUTIONS

JTB, BOP, and GP concieved and designed the study. JTB, MAW, and DJ collected and analyzed the data. All authors contributed to the writing and editing of the manuscript and have reviewed the final version for accuracy and integrity.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Supporting information

Table S1. Physiology parameters of P. tricornutum acclimated to low and high light.

Table S2. Photosystem II D1 subunit damage rates and resulting maintenance cost for P. tricornutum acclimated to high and low light.

Table S3. Predicted excitation energy flow in P. tricornutum acclimated to low and high light. Values are for simulations constrained to account for 100% of absorbed quanta. ΦCO₂: quantum yield of net carbon fixation; ΦO₂: quantum yield of net oxygen evolution. Abbreviations: QF, quantum flux; ETR, electron transport rate; PSII, photosystem II; PSI, photosystem I; Y(NO), unregulated excitation energy dissipation; NPQ, non‐photochemical quenching.

Figure S1. Chlorophyll‐normalized photosynthetic rate versus quantum flux. Chlorophyll‐normalized P _O versus PAR curve. Abbreviations and definitions: LL, low light; HL, high light; PAR, photosynthetically available radiation. Data based on n = 3 biological replicates for LL and n = 4 biological replicates for HL.

Figure S2. Correlation between physiology parameters at low and high light acclimation. (a) Oxygen evolution, 1 – qL [fraction of closed reaction centers (RCs)] and Y(NPQ) versus quantum flux for cells acclimated to low light. (b) Oxygen evolution, 1 – qL and Y(NPQ) versus quantum flux for cells acclimated to high light. Vertical dashed line indices the quantum flux value where Y(NPQ) exceeds 5% of the F _v/F _m fraction.

Figure S3. Impact on photosynthetic parameters with and without 5 mm bicarbonate. (a) Cell‐specific P _O versus QF curve for P. tricornutum acclimated to high light with and without 5 mm sodium bicarbonate. Vertical dashed lines represent the quantum flux received by the cultures at the experimental irradiance. (b) Cell‐specific P _O versus QF curve for P. tricornutum acclimated to low light with and without 5 mm sodium bicarbonate. Vertical dashed lines represent the quantum flux received by the cultures at the experimental irradiance. (c) Chlorophyll fluorescence measurements (PAM) of P. tricornutum acclimated to high light with and without bicarbonate additions. Filled in symbols: +5 mm sodium bicarbonate; empty symbols: −5 mm sodium bicarbonate. Abbreviations: HL, high light; QF, quantum flux.

Figure S4. Numerical constraints for different simulation conditions. Photon uptake constraints at (a) high light and (b) low light. The model reaction EX_photon_e was set to the numerical constraint indicated by the curve at each simulation time point. For the 20% assumption simulations, the quantity of photons allowed to leave the system upstream of the photosystems are also shown as a constraint on the model reaction DM_photon_c. Oxygen evolution constraints at (c) high light and (d) low light. The model reaction EX_o2_e was set to the numerical constraint indicated by the curve at each simulation time point. Photochemical yield [Y(II)] at (e) high light and (f) low light. For simulations where Y(II) was constrained, the stoichiometry for Y(II) in model reaction PHO_PSIIt_u was set to the numerical constraint indicated by the curve at each simulation time point. (g) Summary table of all constant and variable constraints used in various simulations. High light and low light constraints are separated by the ¦ symbol.

Figure S5. Maximum quantum yield of photosystem II with and without lincomycin treatment. Maximum quantum yield of photosystem II (F _v/F _m), with and without lincomycin treatment, for cells acclimated to (a) high light (n = 3) and (b) low light (n = 3).

Figure S6. (a) Chlorophyll fluorescence parameters versus PAR for cells acclimated to low light. (b) Chlorophyll fluorescence parameters versus PAR for cells acclimated to high light. Vertical dashed lines represent the PAR at the experimental irradiance. Abbreviations and definitions: LL, low light; HL, high light; PAR, photosynthetically available radiation; Y(II), quantum efficiency of photosystem II; NPQ, non‐photochemical quenching; Y(NO), unregulated, non‐radiative dissipation of excitation energy. Data based on n = 3 biological replicates for LL and n = 4 biological replicates for HL.

Data S1. Materials and methods.

DATA AVAILABILITY STATEMENT

All of the code used to analyze and generate the results for the present study, along with input data, is available in the Supplementary Information and on the project GitHub page: github.com/JaredTBrod/PAM_GEMs.