Nannochloropsis oceanica, kindly provided by NECTON S.A. (Olhão, Portugal), was pre-cultured in sterile 250 mL Erlenmeyer flasks with 100 mL of filter-sterilised (pore size of 0.2 µm; Sartorius, Germany) medium. Medium composition was based on artificial seawater and micronutrients adapted from the recipe of NutriBloom plus (Phytobloom, Olhão, Portugal) resulting in: NaCl 419.2 mM; MgCl₂·6H₂O 48.2 mM; NaNO₃ 35.3 mM; Na₂SO₄ 22.5 mM; CaCl₂·2H₂O 5.4 mM; K₂SO₄ 4.9 mM; KH₂PO₄ 0.73 mM; Na₂EDTA·2H₂O 52.8 μM; FeCl₃·6H₂O 40 μM ZnSO₄·7H₂O 4 μM; MgSO₄·7H₂O 4 μM; MnCl₂·4H₂O 2 μM; Na₂MoO₄·2H₂O 0.2 μM; CoCl₂·6H₂O 0.2 μM; CuSO₄·5H₂O 0.2 μM. Maintenance cultures were cultivated under orbital agitation, at 25°C, low incident light of 20-50 µmol m^-2 s^-1 with 16:8 L:D cycles. Pre-cultures for reactor cultivation were grown in an orbital shaker HT Multitron Pro (Infors HT, Switzerland) at 120 rpms, 25°C, 100-120 µmol m^-2 s^-1 (warm-white LED) in 16:8 L:D cycles and 0.2% CO₂ enriched headspace. All cultures were maintained at pH=7.5 by the addition of HEPES at a final concentration of 20 mM.

Medium composition of the reactor was the same as for the cultures with the exception that reactor medium did not contain HEPES. Additionally, NaNO₃ and KH₂PO₄ concentrations were increased to 58.8 and 2.76 mM, respectively, to avoid nutrient limitations. Final pH medium was adjusted to 6 to avoid nutrient precipitation during reactor operation.

Pre-cultures were inoculated to heat-sterilised airlift-loop flat panel Labfors 5 photobioreactors (Infors HT, Switzerland). The reactor has an illuminated surface area of 0.08 m², a depth of 20.7 mm and a working volume of 1.8 L. Incident light was provided with a set of 260 water-cooled high-power LED lights (28 V, 2.3 Watt per LED) at 670 μmol.m^-2.s^-1 with a photoperiod of 16h of constant light and 8 hours of darkness. pH was measured and controlled at 7.5 ± 0.05 by acid (0.9 M, H₂SO₄) or base (1 M, NaOH) addition. Mixing was done by sparging air at 0.56 vvm enriched with 2% CO₂. Evaporation losses were minimised with a condenser located on top of the reactor. Temperature was measured and controlled with a water jacket directly connected to a cryostat. In this research, the following temperature setpoints were studied: 9, 13, 17, 21, 25, 29, 31 and 33°C.

Reactors were operated continuously in turbidostat mode. Light on the rear of the reactor was measured by a secondary light sensor (LI-190SA 2ϖ quantum sensor, Li-Cor, USA) and controlled at 15 μmol m^-2 s^-1, above the experimental compensation light intensity (I_ph,c) of 4 μmol m^-2 s^-1 estimated from Eq 1:

where q p h m is the specific photon supply rate for maintenance purposes assumed to be 3.53 mmol_ph gDW^-1 h^-1 for Nannochloropsis sp (Barten et al., 2022) and a_x the averaged absorption cross-section assumed to be 0.27 m² gDW^-1 at low light conditions (Janssen et al., 2018). Cultures were first grown in batch at an initial OD₇₅₀ of 0.15-0.2 and incident light of 200 μmol m^-2 s^-1. The incident light was increased stepwise with the growth of the culture, first to 400 μmol m^-2 s^-1, and then to 670 μmol m^-2 s^-1. At this light intensity, turbidostat mode started and filter-sterilised medium was added automatically whenever the light intensity at the rear dropped below the setpoint. The overflow (V_Harvest) was pumped out to an empty vessel and it was collected and measured every day two hours after the beginning of the light period. The dry weight (DW) biomass concentration and the absorption cross section (a_x) were measured to assess the physiology and stability of the culture. It was considered that the cultures reached the steady-state when the dilution rate, DW and a_x did not change more than 10% for at least 3 residence times. At steady state, samples for the measurement of DW, a_x, cell concentration and oxygen evolution were directly taken from the reactor. The experiment at 9°C was not sampled at steady-state, but once the DW and the a_x did not change more than 10%. The V_Harvest line was connected to an empty vessel in order to collect biomass for biochemical analysis. Collection was done for 4 hours under dark conditions and the vessel was placed on ice. The V_Harvest collected was measured, centrifuged (5 min, 4200 g) and washed twice with ammonium formate (0.5 M) in order to remove any traces of salts. The supernatant was discarded and the pellet was flushed with N₂ gas, sealed and stored at -20°C until further use.

After the collection of samples for one temperature, the experiment was concluded and the temperature setpoint was reduced or increased by a change of 2°C per day. After reaching the setpoint required, a new experiment started and the culture was monitored until a new steady state was achieved. Sampling at steady state was done on three different days. The experiment at 25°C was repeated in a different photobioreactor to assess the reproducibility of the results.

Dry weight was measured by assessing the difference in weight of a known volume of culture on pre-weighted Whatman glass microfiber filters (55 mm; GE Healthcare, USA). Shortly, filters were pre-washed with deionised water, dried in the oven (100°C) overnight and cooled down in the desiccator for 2 hours and weighted. A culture volume containing 2-5 mg of biomass was filtered in pre-weighted filters and washed with ammonium formate (0.5 M) to remove the excess of salts from the biomass. Filters were dried for 24h, placed in a desiccator for at least 2h and weighted afterwards. Measurements were done in triplicate.

Cell concentration was measured in a Coulter Multisizer 3 (Beckman Coulter Inc, USA). Samples with an approximate optical density at 750 nm of 0.1-0.2 were diluted 100 times in Isoton reagent (ISOTON™ II Dilutant, Beckman Coulter Inc., USA) and measured with an aperture tube of 50 μm. Cell concentration and diameter were calculated from the number of particles distributed between 2 and 5 μm after analysing 1 mL of diluted sample.

The absorption spectrum of undiluted cultures was measured in a UV-2600/2700 Spectrophotometer (Shimadzu, Japan) equipped with an integrating sphere for light scattering correction. The absorbance from biomass scattering (740-750 nm) was substracted to the absorbance from 400 to 700 nm and normalised to the DW of the sample and light path (2 mm). The averaged-specific absorption cross section was obtained after averaging the relative absorbance values from 400 to 700 nm.

The maximum quantum yield of PSII photochemistry (QY_max) was measured in diluted dark-adapted samples with a fluorometer (AquaPen-C AP-C100, Photon Systems Instruments, Czech Republic). The maximal fluorescence signal (F_m) and fluorescence baseline (F_o) from diluted cultures (OD₇₅₀ ~ 0.1-0.2) were obtained after incubation in dark conditions for 10-15 min. Maximal quantum yield of PSII was measured as follows:

Respiration and photosynthesis rates were measured by monitoring the oxygen evolution at different incident light intensities in an Oxytherm+P system (Hansatech Instruments Ltd, England). This is a liquid phase system with a cuvette of 9 mm diameter illuminated from one side with white LEDs. An O₂ electrode was prepared according to manufacturer’s instructions. Prior to measurements, the electrode was calibrated with air (100% saturation) and N₂ gas (0% saturation) with reactor medium adjusted to pH 7.5 and 40 mM final HEPES concentration. Na₂S₂O₄ (1-3 mg) was added for the 0% saturation calibration to further remove dissolved oxygen. Calibration was done every time the device was used and according to the temperature in the reactor. Temperature was controlled during the whole measurement. Prior to the analysis, the culture was diluted to a final known biomass concentration of 0.15-0.2 g L^-1 and dissolved O₂ concentration was reduced to 25% of the saturation value. Sodium bicarbonate (NaHCO₃) was added to a final concentration of 10.2 μM to prevent carbon limitation during the measurement. Oxygen evolution was measured according to the following irradiance setup plan: darkness (6 min), 10 (4), 20 (3), 40 (2, 80 (2), 160 (1), 250 (1), 350 (1), 500 (1), 750 (1), 1000 (1), 1500 (1), 2000 (1), 2500 (1), 3000 (1). Dissolved oxygen concentration (μmol_O2 L^-1) was recorded in intervals of one second and the volumetric oxygen rate (μmol_O2 min^-1 L^-1) was calculated after stabilisation of the slope.

Total fatty acid quantification was done according to (Breuer et al., 2013), with the exception that no internal standard was added prior to disruption. Shortly, 5-10 mg of freeze-dried biomass were bead-beaten (3 cycles of 60 s at 2500 rpm with 120 s break) and transferred to a fresh new tube. Sonication (10 min, 80 kHz) and a saline solution of 50 mM and 1 M NaCl at pH 7.5 were added to the disrupted extract. 3 cycles of extraction with 1 mL of chloroform and transfer to a fresh tube were applied. Chloroform was evaporated with N₂ gas and the lipid extract was sealed with parafilm and stored at -20°C until further use.

Lipid classes were separated and their FA composition was analysed as described by Gros and Jouhet, 2018, with some minor differences. Shortly, after dilution of lipid extract in 1 mL of chloroform, a 50 μL sample was methylated (1 h, 100°C) and the total fatty acids were quantified by GC-FID. The volume corresponding to 750 nmol of FA was loaded in pre-activated Thin Layer Chromatography (TLC) silica plates (105721, Supelco, Germany) and different plates were used to separate polar and neutral lipids, respectively. Migration and identification of lipid classes in TLC plates were done accordingly to Gros and Jouhet, 2018 with the difference that a solution of primuline 0.5% v/v in 8:2 acetone: deionised water was employed as a staining agent. After identification, the silica from the spots was scrapped and collected into separate glass tubes. The content of the tubes was methylated (1 h, 100°C) and quantified by GC-FID by using the FAME standard pentadecanoic acid (CAS 1002-84-2, Merck, Germany). FA quantification was based on the relative response of a standard mix of fatty acid methyl esters (FAMEs) and normalised to the initial weighted DW. The FA content for each lipid class was normalised to the sum of all the FAs to estimate the FA composition per lipid class (% mol mol^-1). The lipid class composition was based on the total content of FA found in each lipid spot normalised to the sum of all the different lipid and the number of acyl chains per lipid class. The content of FA in lipid classes (w/w) was calculated based on the FA composition, the lipid class composition and the TFAs.

The dilution rate (D, d^-1) of the turbidostats was calculated from the harvested overflow as follows:

where V_Harvest was collected after 1 day and V_r stands for the working volume of the reactor.

The volumetric absorption capacity was calculated as the product of the a_x and DW.

Photosynthesis-Irradiance (PI) curves were obtained from plotting the specific oxygen production rate (q _{O 2} ) to the incident light intensity (I_ph,o). Volumetric oxygen production rates, measured from biological oxygen monitoring experiments, were normalised to the estimated DW in the light chamber.

The measured specific oxygen production rate is the result of the cellular photosynthesis and respiration rates (Eq. 4). Photosynthesis rates can be described as a function of the incident light intensities with the Jassby and Platt model (Jassby and Platt, 1976) as follows:

where q O 2 d a r k is the specific oxygen rate associated with respiration processes, q O 2 m a x is the maximal specific oxygen production rate, α is the slope of the linear phase at light limiting conditions and Y_oph is the yield of oxygen on light. Fitting the specific oxygen production rates as a function of the incident light intensity of each experiment to Eq. 4 resulted in an estimation of q O 2 m a x , α and q O 2 d a r k for each temperature evaluated.

Descriptive statistics were used to estimate the average and standard deviations of replicate measurements. The post hoc Tukey’s range test was done to assess changes in lipid classes and FA composition per lipid class at different temperatures with a confidence level of 95%. Matlab (MATLAB, 2010) was used for parameter estimation and data visualisation. Error propagation was used in case multiple arithmetic operations were required. R (R Core Team, 2021) was used for statistical analysis.

Temperature is the parameter that exerts the most notorious effect on growth during microalgae cultivation after light. The effect of temperature on growth has been widely studied in Nannochloropsis species by several authors ( Table 1 ). Except for the turbidostat studies of Hoffmann et al., 2010, most studies were done in batch and therefore light perceived by the cultures varied with culture growth. In many of those studies, a synergistic effect of temperature and light was studied without a distinction of the influence of each parameter. In this study, the light perceived by the culture was fixed for all experiments by keeping a constant light gradient over the depth of the reactor. In this way, the effect derived from temperature is the only variable evaluated. In continuous experiments, where light is the only limiting factor, the dilution rate measured is equivalent to the growth rate of the culture according to the cell mass balance over the reactor. The steady-state dilution rates at different temperatures are represented in Figure 1A . The highest dilution rate ranged between 0.68 and 0.70 d^-1 and was found between 25 and 29°C. The dilution rate decreased sharply from 0.7 to 0 d^-1 at supraoptimal values from 29 to 33°C, while the decrease was less steep at suboptimal temperatures from 0.68 d^-1 at 25°C to 0.03 d^-1 at 9°C. The abrupt drop in growth at supraoptimal temperatures indicates that this species is extremely sensitive to temperatures near 30°C. The optimal temperatures for growth found in this research agree with previous batch experiments with other Nannochloropsis strains. The optimal temperature for growth of Nannochloropsis oceanica was found between 26 and 28°C at different incident irradiances between 34 and 80 μmol m^-2 s^-1 (Sandnes et al., 2005). An optimal temperature for growth of 24-26°C was also found for Nannochloropsis gaditana (Van Wagenen et al., 2012), Nannochloropsis sp (Sukenik, 1991) and Nannochloropsis oculata (Wei et al., 2015).

In a turbidostat operation mode, where the outgoing light intensity is fixed, the light absorption area per volume of reactor is expected to be constant since the incident light intensity was also constant at daylight. Hence, the biomass dry weight (DW) and the averaged specific absorption cross section area (a_x) of the biomass affect the volumetric light absorption capacity of the culture. The volumetric light absorption capacity was constant at 252 m² L^-1 for all temperatures studied except at 31°C, where it decreased to 233 m² L^-1 due to unknown reasons. Despite the constant light absorption, the DW and the a_x changed inversely both at low and high temperatures as shown in Figure 1B . At temperatures near optimal growth (25-29°C), biomass concentration and absorption cross section were 1.55-1.82 g L^-1 and 145-163 m² kg^-1, respectively. From 29 to 17°C, the DW increased from 1.82 to 2.33 g L^-1 and the a_x decreased from 163 to 109 m² kg^-1. The a_x increased at even lower temperatures from 109 m² kg^-1 at 17°C to 119 m² kg^-1 at 9°C. Data of absorbance spectra showed a higher absorbance between 420 and 480 nm, which indicates an increased content in carotenoids at 9°C ( Figure S1 ). At temperatures higher than 29°C, the biomass concentration remained constant at 1.6 g L^-1 while the a_x decreased from 163 to 142 m² kg^-1.

Cell concentration and cell diameter are presented in Figure 1C . Similar to the biomass DW, the cell concentration increased from 2.50E+8 cell mL^-1 at 31°C to 3.98E+8 cell mL^-1 at 9°C. Cell diameter was constant (~3.05 μm) at optimal and supraoptimal conditions but increased up to 3.31 μm at 17°C. At lower temperatures than 17°C, the diameter decreased to 3.00 μm. The decrease in diameter could explain the decrease of DW observed at 9°C. Previous short-term studies in Chlorella, Scenedesmus or Mycrocistis showed increased cell volumes and decreased chlorophyll contents at low temperatures (Maxwell et al., 1994; Staehr and Birkeland, 2006).

The photosynthetic performance was evaluated by using oxygen evolution and chlorophyll fluorescence. Photosynthesis-Irradiance (PI) curves were developed for each temperature to gain understanding on the acclimation, maximal photosynthesis rates and respiration rates. After fitting the Jasby and Platt model to the specific oxygen production rates over incident light intensity, the maximal specific oxygen production rate ( q O 2 m a x ), the maximum light utilisation coefficient (α) and the specific oxygen respiration rate in the dark ( q O 2 d a r k ) were estimated. Examples of PI curves derived at 17, 25 and 31°C including their respective fitting results are shown in Figure 2A . The q O 2 m a x indicates the maximal photosynthetic capacity of the microalga and was strongly affected by low and high temperatures. In similarity to the dilution rate, the q O 2 m a x was maximal at 25°C (1.58 μmol_O2 gDW^-1 s^-1) and decreased to 0.94 μmol_O2 gDW^-1 s^-1 at 31°C and 0.14 μmol O₂ gDW^-1 s^-1 at 9°C ( Figure 2B ). Low temperatures slow down the rates of repair of the photosynthetic apparatus and electron transport in the thylakoid membranes resulting in a redox imbalance (Mellis, 1999). As a result, fewer PSII reaction centers are functional at lower temperatures, which ultimately results in a decreased maximal photosynthesis rate (Sakshaug et al., 1997). A slower electron transfer rate coupled to reduced enzymatic activities leads to a reduced energy pool in the cell (ATP and NADPH). The limited availability of ATP and NADPH will in turn affect relevant pathways such as carbon fixation, which is then reflected on a decreasing q O 2 m a x .

The decreasing q O 2 m a x at higher temperatures was also explained mainly by limitations in the carbon fixation reactions. Despite the high thermolability of the photosystems at high temperatures, the activase activity of the Rubisco enzyme was recognised as the main bottleneck (Crafts-Brandner and Salvucci, 2000; Salvucci and Crafts-Brandner, 2004; Allakhverdiev et al., 2008). Higher temperatures were shown to reduce the specificity of the Rubisco towards CO₂ over O₂, leading to the accumulation of phosphoglycolate. Phosphoglycolate needs to be regenerated to glycerate in order to reuse the carbon in the Calvin cycle and this can only be done via photorespiration (Peterhansel et al., 2010). The regeneration of carbon costs energy in the form of ATP and reducing power, resulting in lower photosynthetic rates. The maximal specific photosynthetic rate decreased by 40% ( Figure 2B ) at 31°C while the dilution rate only decreased by 10% ( Figure 1A ). The divergence existing between the dilution rate and the photosynthetic rates at 31°C suggests the occurrence of photorespiration but also the activation of other mechanisms generating additional energy. Heat stress was shown to increase the activity of PSI, hence increasing the redox state of the PSI (Sharkey and Zhang, 2010). As a result, a cyclic flow of electrons occurs to maintain the proton gradient across the thylakoid membranes and generate additional energy in the form of ATP. The activation of electron cyclic flow at 31°C could explain the maintenance of cell growth rates near the optimal despite the decay in photosynthetic rates.

Dark respiration accounts for the sum of all the metabolic processes that consume oxygen independently of light. Respiration rates are usually assumed to be 10% of the photosynthesis rate and can change with environmental conditions (J.Geider and Osborne, 1989). The respiration rates followed a similar trend to the photosynthetic rates by increasing with temperature. The estimated q O 2 d a r k was maximal at 29°C (0.15 μmol gDW^-1 s^-1) and decreased to 0.13 μmol gDW^-1 s^-1 at 31°C and 0.02 μmol gDW^-1 s^-1 at 9°C ( Figure 2C ). The high respiration rates at optimal conditions rather than suboptimal critical conditions are indicative of an increased rate of catabolism of carbohydrate storage to support higher growth rates. The q O 2 d a r k : q O 2 m a x increased at lower and higher temperatures than 25°C suggesting higher metabolic requirements at supra- and suboptimal conditions ( Figure S2A ).

The maximum light utilisation coefficient (α) or slope of the PI curve is indicative of the transition from light-limiting to saturating conditions. α increased from 3.6E-3 μmol_O2 m² μmol_ph ^-1 gDW^-1 at 31°C to 4.7E-3 μmol_O2 m² μmol_ph ^-1 gDW^-1 at 29°C and then decreased at lower temperatures than 25°C to 2.2E-4 μmol_O2 m² μmol_ph ^-1 gDW^-1 at 9°C ( Figure 2D ). The α correlates with the yield of oxygen released per mole of photon (Y_oph), providing additional information on the metabolic state of the culture. The yield decreased from 0.026 mol_O2 mol_ph ^-1 at 29°C down to 0.002 mol_O2 mol_ph ^-1 at 9°C, further confirming a loss of photosynthetic efficiency at lower and higher temperatures than optimal ( Figure S2B ). At higher temperatures than 29°C, the loss was less apparent. These results confirm that photorespiration occurs at 31°C and that a big share of energy for growth is granted by a predominant occurrence of cyclic photophosphorilation. In general, the obtained yields were relatively low to the expected theoretical values (Sakshaug et al., 1997) and this is explained by the protocol used. First of all, the incident light intensity instead of the average light intensity were used in the x-axis of the PI curve to estimate the slope. This resulted in an understimation of the real value since the non-absorbed light is also taken in the calculations (Saroussi and Beer, 2007). Secondly, the resulting values are dependent on the mathematical expression used. While the estimation of the q O 2 m a x is unvariable to the model used, the slope is highly dependent on it. In our study, we used the Jasby and Platt model to describe the relationship between light irradiance and photosynthesis but the use of other models would yield slightly different results. This might complicate the comparison and interpretation of data between different studies.

The QY_max was also measured at each temperature and the results are plotted in Figure 2D . The QY_max was highest between 31 and 25°C (0.65-0.64) and slowly decreased from 0.64 at 21°C to 0.61 at 17°C and then reduced by 2-fold every 4°C down to 0.15 at 9°C. A low QY_max suggests a low photosynthetic performance due to the occurrence of photoinhibition. Photosynthetic organisms dissipate the excess of absorbed light (not used in photochemistry) via non-photochemical quenching (NPQ). Distinct mechanisms have been studied in eukaryotes which are categorised depending on the relaxation times. Higher energy state quenching (qE) occurs in the order of seconds while photoinhibition quenching (qI) or zeaxanthin-dependent quenching (qZ) occur at longer times (Hüner et al., 2022). The prolonged adaptation of the culture to sub- and supraoptimal temperatures favors the occurrence of quenching mechanisms of longer time scales such as qI and qZ. Previous studies showed that QY_max and α of Chlorella cultures growing at 27°C were unaffected after short exposure to 5°C indicating that low temperatures do not interfere with the excitation process (Maxwell et al., 1994). However, chronic exposure to low temperatures combined with high saturating light intensities results in oxidative stress which is accentuated at lower temperatures. The observed decrease of QY_max and α at low temperatures support the activation of non-photochemical quenching processes. Lowering the temperature from 25°C to 17°C resulted in a decrease in α while the QY_max remained relatively constant. This discrepancy between QY_max and α, where QY_max decreased by only 8% and α decreased by 57% indicates that different quenching mechanisms are activated This possibly indicated that non-photochemical quenching occurred via fast relaxation routes such as qE or other routes of electron transfer such as the cyclic electron flow for ATP generation (Sakshaug et al., 1997). At temperatures lower than 17°C, the additional decrease in QY_max indicated membrane photodamage in the thylakoids and therefore the activation of new quenching routes of longer time scales. qI is a quenching mechanism occurring under photoinhibitory conditions that induce the inactivation of photosynthesis. Moreover, decreasing the temperature of the culture results in cells experiencing conditions similar to high saturating light intensities. The xanthophyll cycle is a photoprotective mechanism present in Nannochloropsis oceanica which aims at dissipating the excess of absorbed light under high light conditions. In this cycle, the product of violaxanthin de-epoxidation (zeaxanthin) acts as a quencher dissipating the excess of aborbed energy. As shown in the appendix, the increased absorption observed at 500 nm at temperatures lower than 25°C indicates the activation of the xanthophyll cycle.

At temperatures higher than optimal, the QY_max did not change, in contrast to lower temperatures. The discrepancy between the chlorophyll fluorescence method and the oxygen evolution further evidences that processes involved in carbon fixation are more limiting than the ones involved in light reactions during photosynthesis at high temperatures. Studies in acclimated and non-acclimated wheat and cotton plants to different temperatures showed that PSII photosynthetic activity was more resistant than Calvin cycle activity (Law and Crafts-brandner, 1999). Besides an increase in photorespiration, long-term exposure to high temperatures was shown to promote the uncoupling of the reaction centres of the photosystems from the light-harvesting complexes, as well as the unstacking of grana in the thylakoids (Hemme et al., 2014). Structural changes in the photosynthetic machinery could explain the lower a_x ( Figure 1B ), which occurred without significant changes in the absorption spectrum ( Figure S1 ). Previous studies in plants reported changes in the light scattering properties of the chloroplast due to the formation of plastoglobules (Zhang et al., 2010).

Overall, these results emphasize that a different photosynthetic response occurs at low and high temperatures characterised by different mechanisms. Additionally, N. oceanica growth at lower temperatures is highly correlated to its photosynthetic activity, while temperatures higher than the optimum result in an uncoupling between growth and photosynthetic activities.

Besides, changes in photosynthesis rates, temperature variations affect intracellular lipid content and composition (Thompson, 1996; Solovchenko, 2012). Figure 3A depicts the changes in the content of total fatty acids (TFA) and the distribution between polar and neutral lipids for 4 selected temperature conditions. The temperature 29°C was representative of optimal growth, while temperatures 9 and 31°C were representative of suboptimal and supraoptimal conditions, respectively, at which growth was still present. Since temperature 17°C was observed as a tipping point in cell physiology and photosynthesis performance, lipids were also analysed and compared in the selected conditions. The TFA content remained constant at 17% w/w in the temperature range 17-31°C and decreased to 14.9% and 13.3% w/w at 13 and 9°C, respectively. Decrease in TFA was mainly due to a decrease in saturated fatty acids (SFAs), which decreased from 9.4% w/w at 31°C to 4.2% w/w at 9°C ( Table S1 ). Polyunsaturated fatty acids (PUFAs) increased from 3.9% w/w at 31°C to a stable value of 4.8% w/w at 25°C and remained stable at lower temperatures. Fractionation of TFA between polar and neutral lipids showed that the polar lipid content was highest (7.9% w/w) near the optimal temperature (29°C) and decreased to 5.5% and 6.2% at 9°C and 31°C, respectively. Concerning the neutral lipids, the content at 29°C was 8.3% w/w, increased at lower temperature up to 12.0% w/w and decreased again down to 7.4% w/w due to the reduced photosynthesis rates. At supraoptimal conditions, the content increased up to 10.8% w/w. The decrease in neutral lipids correlated well with the reduced diameter observed at critical suboptimal temperatures.

Polar and neutral lipids were further separated into lipid classes with thin layer chromatography (TLC) coupled to gas chromatography for lipid quantification. The FA content per lipid class as well as their FA composition are shown in Figures 3B , 4 , respectively. The following sections are subdivided into the different categories of polar and neutral lipids.

Nannochloropsis species are interesting from an industrial point of view since they accumulate high contents of EPA in comparison to other microalgae-producing species. Nevertheless, the economic profitability is dependent on the productivities which are currently still low for photoautotrophic systems. Lowering the temperature is regarded as a strategy to increase the EPA contents and productivities despite the growth impairment (Hoffmann et al., 2010). The total EPA content and distribution between polar and neutral lipids obtained in this research is shown in Figure 5 . The total EPA remained constant at 3.5% w/w with decreasing temperature from 29 to 9°C and only decreased to 3.0% w/w from 29 to 31°C. The decrease in total EPA content was due to a decrease in EPA in the polar fraction (2.4 to 1.7% w/w). In contrast, EPA in the neutral fraction did not change over the range of temperatures tested. Despite the constant EPA content in the polar fraction, the content between lipid classes changed with temperature indicating a dynamic mobilisation of EPA ( Figure S4 ). The increase of EPA content in DGTS, TAG and galactolipids at low temperatures indicates that these lipid classes play a relevant role in cell adaptation. Despite the changes in FA content in the polar fraction, the net content in EPA was constant and only affected at high temperatures due to a need for saturated fatty acids. The constant EPA content with the dynamics changes in lipid classes (and therefore their EPA content) indicates a tight regulation of synthesis and degradation of EPA.

Several experimental studies evaluated the effect of temperature on different species of Nannochloropsis ( Table 2 ). Most of these studies reported an increase in EPA at slightly suboptimal temperatures. In contrast, our work did not show any changes at low temperatures. Several reasons could explain this divergence including the scale, the incident light intensity, the light pattern (continuous or circadian cycles) or the operation mode (batch or continuous) used in the experiments. Batch studies employing the same microalgae species did not result in changes in the EPA content after changing the temperature setpoint from 25 to 15°C with similar incident light intensities (Sá et al., 2020). One common feature is that all studies that reported an increase of EPA were performed at incident lights ranging from 60 to 250 µmol m^-2 s^-1. Our study was done at a higher incident light intensity (670 µmol m^-2 s^-1) which is higher than the I_ph,sat reported for Nannochloropsis species (Fisher et al., 1996). Although this needs to be confirmed with further research, the incident light intensity might influence the cellular accumulation of EPA. This might be due to the fact that a low incident light combined with low temperatures could lead to a lower redox imbalance in contrast to higher incident light intensities and low temperatures.

Temperature in outdoors systems is barely controlled and fluctuation can occur in the range of 10°C over the day (Béchet et al., 2010). Besides daily variations, seasonal temperatures can also influence growth and restrict microalgae cultivation to certain periods during the year depending on the thermotolerance of the species. Despite daily variations can occur in small frames of time over the day, both the EPA and the biomass productivity of the culture can be greatly affected. Biomass and EPA productivity followed a similar profile with an optimum at temperatures between 21 and 25°C ( Figure S5 ). These results coincide with previous work which indicated that the optimal EPA productivity was found at temperatures optimal for growth (Chini Zittelli et al., 1999; Camacho-Rodríguez et al., 2014). However, a 10°C degree reduction from 25 to 15°C results in a 2-fold decrease in both biomass and EPA productivity. A different behaviour is observed at temperatures higher than optimal where the EPA productivity is more affected than biomass due to the reduced content in EPA. It is therefore indispensable to know in advance how temperature affects growth and product accumulations for optimal reactor operations at bigger scales even when control cannot be executed.