Scale‐down of oxygen and glucose fluctuations in a tubular photobioreactor operated under oxygen‐balanced mixotrophy

Oxygen‐balanced mixotrophy (OBM) is a novel type of microalgal cultivation that improves autotrophic productivity while reducing aeration costs and achieving high biomass yields on substrate. The scale‐up of this process is not straightforward, as nonideal mixing in large photobioreactors might have unwanted effects in cell physiology. We simulated at lab scale dissolved oxygen and glucose fluctuations in a tubular photobioreactor operated under OBM where glucose is injected at the beginning of the tubular section. We ran repeated batch experiments with the strain Galdieria sulphuraria ACUF 064 under glucose pulse feeding of different lengths, representing different retention times: 112, 71, and 21 min. During the long and medium tube retention time simulations, dissolved oxygen was depleted 15–25 min after every glucose pulse. These periods of oxygen limitation resulted in the accumulation of coproporphyrin III in the supernatant, an indication of disruption in the chlorophyll synthesis pathway. Accordingly, the absorption cross‐section of the cultures decreased steeply, going from values of 150–180 m2 kg−1 at the end of the first batch down to 50–70 m2 kg−1 in the last batches of both conditions. In the short tube retention time simulation, dissolved oxygen always stayed above 10% air saturation and no pigment reduction nor coproporphyrin III accumulation were observed. Concerning glucose utilization efficiency, glucose pulse feeding caused a reduction of biomass yield on substrate in the range of 4%–22% compared to the maximum levels previously obtained with continuous glucose feeding (0.9 C‐g C‐g−1). The missing carbon was excreted to the supernatant as extracellular polymeric substances constituted by carbohydrates and proteins. Overall, the results point out the importance of studying large‐scale conditions in a controlled environment and the need for a highly controlled glucose feeding strategy in the scale‐up of mixotrophic cultivation.


| INTRODUCTION
Microalgae are considered a promising feedstock in the transition towards sustainable production systems and the biobased economy (Fernández et al., 2021). These unicellular photoautotrophic (from now on: autotrophic) organisms initially raised considerable interest due to their natural ability to fix atmospheric carbon dioxide (CO 2 ) at higher rates than land plants (Chelf et al., 1993). Nevertheless, the development of large-scale autotrophic cultivation has been restrained by a number of factors that limit the economic feasibility of bulk products Khan et al., 2018). Carbon fixation is dependent on light availability, but light penetration is poor in concentrated cultures due to mutual cell shading. A common practice to circumvent this problem usually involves reactor operation at low biomass concentration, incurring higher downstream processing costs. Besides light limitation, microalgal cell growth might also be limited by an insufficient supply of CO 2 and accumulation of oxygen (O 2 ). For this reason, ensuring an adequate gas-liquid transfer is essential to avoid suboptimal productivities. This is usually approached in two ways: CO 2 -enriched gas supply and high aeration flow rates, both having an impact in the final costs of the process (Langley et al., 2012;Ruiz et al., 2016).
A strategy to tackle the yet elevated production costs of autotrophic cultivation consists of the addition of organic carbon substrates to illuminated microalgal cultures (Pang et al., 2019). In this way, autotrophic metabolism is combined with the heterotrophic counterpart in a mixotrophic regime. Light dependency is reduced as growth can proceed in the dark zones of the reactor. This increases productivity and biomass concentration at the expense of the organic carbon source, while maintaining the possible benefits of autotrophic carbon fixation. If substrate supply is adjusted to match photosynthetic oxygen production, then oxygen consumption and production can be balanced by means of intracellular "gas recycling" (Grama et al., 2016). Most CO 2 resulting from the oxidation of the substrate is, in turn, fixed by the photosynthetic machinery, achieving up to 90% efficiency in the utilization of the organic carbon. Altogether, this type of mixotrophy, denominated OBM, enables to operate photobioreactors without any gas exchange during daylight, obtain higher biomass concentration and achieve maximal carbon conversion efficiency (Abiusi et al., 2020a(Abiusi et al., , 2020b(Abiusi et al., , 2021(Abiusi et al., , 2022. So far, we have successfully developed OBM in a lab-scale reactor, constituted by a cylindrical vessel with a working volume of 2 L and mechanical stirring at 500 rpm (Abiusi et al., 2020a). Such system can be considered ideally mixed, meaning that the concentrations of biomass, organic substrate, and dissolved oxygen (DO) are virtually the same in the entire volume. When the organic carbon source is supplied, it is almost immediately depleted after becoming in contact with the liquid.
In this manner, a stable DO level is achieved while maintaining organic substrate concentration close to 0. On the other hand, in larger systems, mixing is far from ideal. For example, one of the most common reactor configurations for large-scale cultivation of microalgae is tubular photobioreactors, characterized by plug flow fluid dynamics (Acién Fernández et al., 2001). Based on the combination of this type of fluid behavior and algal growth, both oxygen and organic substrate gradients are expected along the tubes. Achieving oxygen balance in such scenario might thus be more complex than at lab-scale. Accordingly, the algal cells might experience a changing environment with unknown effects on their mixotrophic metabolic equilibrium, entailing a scale-up risk for this process.
The aim of this work was to evaluate the performance of OBM in tubular photobioreactors before carrying out the actual scale-up. This issue was tackled by developing a scale-down strategy that reproduced mixotrophy in a tubular photobioreactor. Scale-down strategies are a common procedure in heterotrophic bioprocesses (Noorman, 2011) and have also been applied in tubular photobioreactors operated autotrophically (Rosello Sastre et al., 2007), but never mixotrophically. The idea behind is to mimic large-scale fluctuating conditions in a lab-scale setup, mainly focusing on sugar and oxygen availability. This allows to gain understanding about the physiological response of the cells in a changing environment and how that affects the overall performance of the process. The information gathered can then be used to derisk the subsequent scale-up design. Our approach consisted of feeding glucose in pulses of different length, creating gradients of oxygen and organic carbon substrate that resembled those expected in a tubular photobioreactor. Consistently with our previous research, we chose Galdieria sulphuraria for this study as its acidophilic nature prevents bacterial growth, a great advantage in a mixotrophic cultivation system with an organic carbon source (Abiusi et al., 2021(Abiusi et al., , 2022. As such, we consider this species as an ideal candidate for the scale-up of OBM.

| Photobioreactor setup
Experiments were carried out in a 3-L stirred-tank bioreactor with an internal diameter of 0.13 m (Applikon). The liquid height was maintained at 0.165 m, resulting in a working volume (V PBR ) of 2 L and a cylindrical illuminated surface (IS) of 0.067 m 2 . The vessel was illuminated homogeneously with an average continuous PFD of 507 ± 45 μmol m −2 s −1 , as described in detail in Abiusi et al. (2020a). Continuous stirring at 500 rpm was applied during all experiments. The temperature was controlled at 37°C and water evaporation was prevented with a condenser fed with water at 2°C. pH was maintained at 1.8 by automatic base addition (2 M NaOH). DO was measured by a VisiFerm DO ECS 225 DO sensor (Hamilton). The sensor was calibrated inside the reactor containing medium at the aforesaid working temperature and pH. The temperature was automatically accounted for during the measurements by means of the built-in thermometer in the DO sensor. Pressure during calibration and subsequent reactor operation was slightly above atmospheric levels (<1% difference, data not shown). This difference was considered negligible when processing the measured data. Dinitrogen gas (N 2 ) and air sparging were applied to obtain 0% and 100% DO reads, respectively. During autotrophic and mixotrophic adaptation phases, the reactor was sparged with air enriched with 2% v/v CO 2 at a flow rate of 1 L min −1 using Smart TMF 5850 S mass flow controllers (Brooks Instruments). Volume increase due to glucose external supply and/or base addition was compensated by daily sampling.
2.3 | Repeated batch cultivation of G. sulphuraria G. sulphuraria ACUF 064 was cultivated axenically in repeated batch in three different experimental conditions: short, medium, and long hydraulic retention time (HRT) simulation of a tubular reactor under OBM. At each condition, several consecutive batches were performed.
The first batch started similarly in the three scenarios as an adaptation to the new conditions. First, the reactor was inoculated with an autotrophic culture at an initial biomass concentration (C x ) of 0.8 g L −1 .
Then, the reactor was operated autotrophically for 3 days, followed by a mixotrophic adaptation phase of 2 days. During the latter, a glucose solution of 100 g L −1 was fed at a fixed rate of 1.0 ± 0.3 g solution h −1 .
After this phase, aeration was switched off and DO was controlled at 90% air saturation by addition of 100 g L −1 glucose solution for 2 more days (i.e., adaptation to OBM). At this point, the cultures were diluted with fresh medium and the first batch finished. Thereafter, glucose was supplied in pulses to the reactor from the second batch until the last batch of each condition. In addition, these batches started always at a biomass concentration of 3 g L −1 and were carried for 4 days until the next dilution. These starting concentrations and batch duration were selected as optimal to stay within the linear growth range based on previous work (Abiusi et al., 2021). In total, five batches were carried for the long simulation and eight for the short and medium simulations, including the initial adaptation batch.

| Glucose pulse feeding
Glucose was supplied discontinuously to mimic the oxygen and substrate gradients experienced by the cells in a tubular reactor operated under OBM. The glucose pump was activated at its maximum rate (147 g h −1 ) for a certain period of time (i.e., blockwise) when DO increased to 60% air saturation. This glucose pulse triggered an almost immediate increase in glucose concentration that drove oxygen below the setpoint of 60%, as depicted in Figure 1. This setpoint was selected based on preliminary trials that focused on maximizing response stability after the pulse and preventing setpoint deviation from air saturation levels (data not shown). The sudden increase in glucose concentration represents the cells passing by the beginning of the tube, where the organic substrate is fed. As the cells oxidized the glucose, oxygen decreased (i.e., cells moved through the simulated tube). When glucose was depleted, DO rose again triggering a new glucose pulse. The length of the glucose pulse determined the time between consecutive DO peaks. This time equals the HRT within the simulated tubular reactor, assuming (1) a single glucose feeding point at the beginning and (2) mixotrophic growth throughout the whole tube. The pumping times required for the long, medium, and short simulations were determined empirically by gradual adjustment of the pumping time at the beginning of batch 2. Once stable cycles of similar duration to the desired HRT were obtained, a fixed pumping time was kept during the rest of each experimental condition. The calculation of the HRT is explained in the next section. Occasionally, right after batch dilution, there was a DO spike (>100% air saturation) caused by the sudden increase in photon supply (Abiusi et al., 2021) that could not be controlled by pulse feeding. During these periods, glucose solution was fed at a fixed rate of 2.1 ± 0.1 g h −1 to maintain cell adaptation to mixotrophy. Once DO decreased below 100% air saturation, pulse feeding was resumed.

| Retention time calculation
As a reference for the retention time simulation, we selected a tubular photobioreactor with a solar collector loop (L) of 300 m and an internal diameter (d) of 0.06 m. These are common dimensions for pilot scale tubular photobioreactors, such as the GemTube MK-1 1500 s tubular photobioreactor (Lgem) located at our AlgaePARC F I G U R E 1 Schematic representation of the glucose pulse feeding strategy. Dissolved oxygen concentration (blue) and glucose pumping rates (red) are depicted in time. The relative size of the glucose pump rate has been exaggerated for the sake of visual representation.
MOÑINO FERNÁNDEZ ET AL. | 1571 greenhouse facility (Bennekom, the Netherlands), described elsewhere (Guimarães et al., 2021). In this study, three different HRT were simulated for such reactor: short, medium, and long. The long simulation aimed to reproduce the maximum HRT attainable in such system. That is, the system operated at the minimum liquid velocity possible. Under OBM, there is no oxygen accumulation, and thus the liquid velocity (v L ) is only constrained by ensuring turbulent flow for proper mixing. In a system that resembles flow through a pipe, turbulent flow is achieved when the Reynolds number (Re) is above values of 3000-4000 (Acién Fernández et al., 2001;Carlozzi & Torzillo, 1996). By applying this condition, the minimum v L can be then solved: where ρ L stands for the liquid density (kg m −3 ) and μ L the liquid dynamic viscosity (Pa s) of water at 37°C. This calculation results in a v L of 0.05 m s −1 . Then, the corresponding HRT can be derived: Accordingly, the minimum v L results in an HRT of 100 min for the long simulation. For the short simulation, we selected 0.20 m s −1 and hence a retention time of 25 min. This velocity lies at the lower limit of the range of velocities commonly employed in tubular photobioreactors operated autotrophically (Belohlav et al., 2021;Hall et al., 2003;Wongluang et al., 2013). Therefore, it can be considered at the upper limit of the desired velocities for OBM. Finally, for the medium simulation, we arbitrarily selected a retention time of 50 min, a value that falls in between the short and long retention times.

| Mixotrophic yield on substrate calculation
First, the total carbon (TC)-based amount of biomass produced in every batch was calculated (M x , C-g). Because daily sampling did not match perfectly the volume increase due to glucose and base addition, the volume change was also taken into account: where C x f and C x i are the biomass concentrations at the end and beginning of the batch, respectively (g L −1 ). V PBR is the working volume at the beginning of the batch (L) and ΔV PBR is the net volume change at the end of the batch (L). C % f and C % i are the biomass carbon concentrations at the end and beginning of the batch (% w C w x −1 ), respectively. In the last term, C x sample represents the biomass concentration of every monitoring sample (g L −1 ), V sample the volume of the sample, and C % avg the batch average biomass carbon concentration. Then, the TC-based amount of glucose consumed by the cells (M s , C-g) was determined for every batch as follows: where M T (g) represents the total amount of 100 g L −1 glucose solution provided to the reactor and C g sol the glucose concentration in the glucose solution (g L −1 ). C g f is the glucose concentration at the end of the batch (g L −1 ) and C g sample is the glucose concentration of every sample (g L −1 ). Finally, C % g indicates the carbon fraction of glucose (% w/w). The mixotrophic biomass yield on substrate (Y mixo x/s , ) was then derived for every batch: x s mixo x s / 2.9 | Dry weight concentration C x was estimated by biomass dry weight (DW, g L −1 ) determination in technical duplicates. Fresh aliquots of culture (1-5 mL) were diluted to 30 mL with demineralized water and filtered over preweighed Whatman GF/F glass microfiber filters (diameter of 55 mm, pore size 0.7 μm). The filters were washed with 30 mL of deionized water and dried at 100°C for at least 3 h.

| Average absorption cross-section
Average absorption cross section (a x , m 2 kg x −1 ) in the photosynthetically active radiation region (400-700 nm) of the spectrum was determined as explained in detail in de Mooij et al. (2015). In short, fresh samples from the reactor were diluted to the range of 0.5-2 g L −1 DW and transferred to two cuvettes with an optical path of 2 mm. The absorbance was measured with a UV-VIS/double beam spectrophotometer (Shimadzu) equipped with an integrating sphere (ISR-2600).

| Photosystem II quantum yield (QY)
Fresh biomass samples were diluted to an optical density (OD) at 750 nm between 0.3 and 0.8 and incubated in darkness at 35°C for 20 min in duplicate. The dark-adapted photosystem II maximum QY of photochemistry (QY, F v /F m ) was measured at 455 nm with an AquaPen-C AP-C 100 (Photon Systems Instruments).

| TC determination
Aliquots of culture were sampled at the end of every batch and centrifuged for 10 min at >20,000 relative centrifugal force (RCF).
The supernatant was stored at −20°C. The pellets were washed twice with deionized water and stored at −20°C. Before measurement, both pellets and supernatants were thawed at room temperature and analyzed for TC (g L −1 ) in duplicate with a TOC-L analyzer (Shimadzu).
The biomass carbon content (C % , %w C w x −1 ) was calculated by dividing TC by the DW of the same sample.

| Coproporphyrin III (COPROIII) monitoring
Aliquots of culture were sampled from the reactor at different timepoints during all treatment batches and centrifuged for 10 min at >20,000 RCF. The supernatant was immediately used to measure OD at 400 nm in duplicate.
2.14 | COPROIII quantification COPROIII identity and concentration were determined via an Agilent 1290 Infinity high performance liquid chromatography (HPLC) system (Agilent). The system consisted of a binary pump, an autosampler, an oven set at 25°C, and a fluorescence detector operated at 404 nm for excitation and 620 nm for emission.

| Glucose quantification
Aliquots of culture were sampled from the reactor at different timepoints during all treatment batches and centrifuged for 10 min at >20,000 RCF. The supernatant was used to measure glucose concentration with a YSI 2950 Biochemistry Analyzer (YSI Life Sciences).

| Organic acids, alcohols, and trehalose quantification
Aliquots of culture were sampled at the end of every batch and centrifuged for 10 min at >20,000 RCF. The supernatant was stored at −20°C. After thawing at room temperature, the liquid samples were analyzed in duplicate with respect to ethanol, glycerol, citric acid, lactic acid, acetic acid, succinic acid, pyruvic acid, formic acid, and trehalose using an Agilent 1290 Infinity HPLC system (Agilent), with an Agilent 1290 Infinity Binary Pump, Agilent 1290 Infinity Autosampler, Agilent 1290 Infinity diode array detector operated at 210 nm and an Agilent 1260 Infinity RI detector operated at 45°C.
The HPLC was operated with a Rezex ROA-Organic Acid H + 300 × 7.8 mm (Phenomenex) column and a SecurityGuard guard column Carbo-H-4 × 3.0 mm (Phenomenex) at 60°C and 0.008 mM H 2 SO 4 as mobile phase at 0.8 mL min −1 as flow rate.

| Carbohydrates quantification
Carbohydrates in the supernatant were determined by the colorimetric method proposed by Dubois et al. (1956). Aliquots of culture were sampled at the end of every batch and centrifuged for 10 min at >20,000 RCF. The supernatant was stored at −20°C. Before the analysis, samples were thawed at room temperature. 50 μL of the sample and 450 μL of medium were added to glass tubes in duplicate.
Then, 500 μL of 5% w/w phenol solution were added to the tubes and subsequently 2.5 mL of concentrated sulfuric acid were also added without active mixing. The samples were incubated first at room temperature for 10 min and then at 35°C for 30 min. During the latter samples were vortexed every 5 min. After the incubation, absorbance was measured at 483 nm. Glucose was used as a reference for the standard curve in a range of 0-0.1 g L −1 .

| Soluble protein quantification
Aliquots of culture were sampled at the end of every batch and centrifuged for 10 min at >20,000 RCF. The supernatant was stored at −20°C and thawed at room temperature when used for analysis. Soluble protein concentration in the supernatant was determined in duplicate with the DC Protein Assay kit from Bio-Rad, following the principle of Lowry's method. Absorbance was determined at 750 nm and bovine serum albumin was used as standard in a range of 0 -1.4 mg L −1 .  (6): Propagation of errors for multiplication operations was calculated according to Equation (7): where σ x is the standard deviation associated with the value x and so on.  COPROIII results from the spontaneous oxidation of COPROGEN, an intermediary in the synthesis pathway of chlorophyll, phycobilin, and the group heme (Hansson & Hederstedt, 1994). Under aerobic conditions, COPROGEN is oxidized to protoporphyrinogen IX by HemF. This enzyme has been putatively identified as Gasu_19740 in the genome of G.
sulphuraria, which also contains a putative isoform, Gasu_14610 (Caspi et al., 2014). Under the two conditions tested in this study in which oxygen was cyclically limiting, we observed a remarkable increase of COPROIII in the supernatant of the cultures. In these two scenarios, the activity of HemF might have not proceeded at normal rate due to the lack of oxygen or its expression might have been downregulated (Goto et al., 2010;Kim et al., 2015). Consequently, COPROGEN accumulated in the cell and was either released and oxidized to COPROIII extracellularly or oxidized in the cell and released as COPROIII. Its excretion, possibly at the expense of energy (Krishnamurthy et al., 2007), might be beneficial as the light activity of the tetrapyrrole ring might be a source of damaging ROS within the cell (Sułek et al., 2020).
Interestingly, the genome of G. sulphuraria also contains a putative HemN (Gasu_49660), an oxygen-independent enzyme that catalyzes the conversion of COPROGEN to protoporphyrinogen IX employing S-adenosyl methionine (Caspi et al., 2014 Sarian et al. (2016). In that work, G. sulphuraria was cultivated mixotrophically without aeration and the oxygen remained at 0% air saturation for days. Similarly to our study, the cultures were photosynthetically active and, therefore, there was oxygen production inside the cells, which might have also interfered with the activation of HemN.

| Oxygen limitation resulted in reduced pigment content
The average absorption cross section (a x , m 2 kg x −1 ) was measured  (Abiusi et al., 2021). Interestingly, in that study, DO was maintained stable at 90% air saturation throughout the whole reactor run, suggesting that the oxygen disturbances of the short condition had no noticeable effect on the pigment content. This adaptation also happened in the long and medium conditions, however, the drastic effect of the anoxic cycles covered the small reduction caused by adaptation to mixotrophy.

| Substrate utilization
Besides oxygen availability, the other important factor to compare among the different conditions tested in this study is the efficiency of substrate utilization. In this section, the focus is given to the mixotrophic yield on substrate (Y mixo x/s , C-g C-g −1 ), whereas productivities and growth rates are not discussed. We centered our analysis in this parameter because maximizing the degree of substrate conversion is one of the goals in OBM. Furthermore, the three experimental conditions tested in this study, which we have called long, medium, and short tube retention times, were obtained providing glucose pulses of different lengths. As the control system was based on DO levels, the number of pulses was not fixed and varied between batches and conditions. Because of this, the total amount of glucose provided to the reactor is not exactly the same from batch to batch and from condition to condition. For this reason, the mixotrophic productivities of the three runs are not directly comparable. To take into account the different amounts of glucose supplied, we can better compare substrate utilization by looking at Y mixo x/s . Y mixo x/s at the end of every batch is reported in Table 1. We obtained yields ranging between 0.70 and 0.86 C-g C-g For example, 0.92 and 0.89 C-g C-g −1 have been achieved in repeated batch (Abiusi et al., 2021) and in chemostat modes with G. sulphuraria ACUF 064, respectively (Abiusi et al., 2022). In the former publication, it was observed that mixotrophic metabolism requires a period of adaptation (i.e., several batches) to reach high yields on the substrate. Nevertheless, in that case, 0.92 C-g C-g −1 was obtained already in the third batch, while in this study, such value could not be reached even after seven batches.
Changes in biomass composition, maybe triggered by the conditions of this study, could affect the maximum theoretical yield of OBM as it is determined stoichiometrically (Abiusi et al., 2020a).

T A B L E 1 Measured and recalculated mixotrophic biomass yield on substrate (Y mixo
x/s , C-g·C-g −1 ) at the end of every batch from the long, medium and short tube retention time simulations. However, considerable changes in biomass composition are unlikely and even if present, cannot explain by themselves a decrease as pronounced as 22%. A decrease in yield might also indicate carbon redirection toward a certain product, such as the aforesaid COPROIII.
From a broader perspective, a reduction in carbon utilization efficiency is a common phenomenon in the scale-up of aerobic heterotrophic processes (Lara et al., 2006). We assumed that EDTA stayed stable throughout the experiments and that the cells did not degrade it. In addition, we monitored glucose concentration along the experiments, including the concentration at the end of the batches (Table 2, Glucose). Accordingly, by subtracting the carbon present in EDTA and glucose from the TC measurements, we can determine the amount of carbon resulting from biological activity. Overall, with this correction, we obtained concentrations between 0.16 and 0.42 C-g L −1 (Table 2, Corrected TC). In Table 1, we recalculated Y mixo x/s by adding the carbon present in the supernatant as if it were part of the biomass. The addition of excreted carbon to the biomass would result in yields ranging from 0.78 to 0.96 C-g C-g −1 , with all batches but one obtaining results ≥0.82. With this increase, the recalculated values are either similar or closer to the maximum Y mixo x/s obtained in previous studies (Abiusi et al., 2021(Abiusi et al., , 2022, closing the carbon gap. However, despite the recalculation, there is still some deviation among the results. This might be a result of the combined effects of the inherent variability of batch processes and the large deviation obtained in some of the TC measurements.
We cannot completely exclude that Y mixo x/s could have also been partially affected by increased cellular maintenance requirements, and thus, a higher fraction of carbon lost as CO 2 . In fact, this phenomenon has already been noticed in G. sulphuraria (Graverholt & Eriksen, 2007). In our study, oxygen and glucose changes were in the order of tens of minutes, sufficient for continuous cycles of regulation at the protein level. These would have resulted in increased energy demand, consequently channeling more substrate into energy production. In principle, during our mixotrophic process, the autotrophic metabolism is also active and recycling the CO 2 from respiration back into biomass. However, momentary peaks of CO 2 production might have overloaded the carbon fixing capacity, perhaps also affected by the pigment disruption caused by oxygen limitation. Regrettably, this could not be validated due to the lack of active aeration during OBM. Gas production derived from algal metabolism by itself was too low to be detected by our gas analyzer. Nonetheless, based on the TC measurements, we can still conclude that the decrease in Y mixo x/s was mainly caused by carbon leakage into the supernatant.

| Cells produced extracellular carbohydrates and proteins
When looking at the possible causes that explain the presence of carbon in the supernatant, the first suspect is the accumulation of COPROIII in the medium, described in previous sections. This An alternative explanation for the decrease in Y mixo x/s might come from the aforementioned excretion of fermentation products triggered by glucose and oxygen fluctuations. To validate this hypothesis, we scrutinized the supernatant of the experiments by means of HPLC. In total, we monitored citric acid, lactic acid, acetic acid, succinic acid, pyruvic acid, formic acid, trehalose, ethanol, and glycerol. None of these compounds were detected in the supernatant. In a prior study with heterotrophic cultures of G. sulphuraria, the algal cells were subject of similar cycles of glucose availability and depletion (Graverholt & Eriksen, 2007). The authors did not detect any excreted molecule as a result. The maximum glucose concentration reached in that study was 0.5 g L −1 , while in our work it was 0.6, 0.2, and 0.1 g L −1 for the long, medium, and short conditions, respectively (data not shown). As maximum concentrations were in the same range, this is an additional indication to dismiss the fermentation pathway transient activation hypothesis.
Interestingly, the carbon excretion was identified as a mixture of carbohydrates and proteins by means of the Dubois and Lowry methods ( Table 2). Carbohydrates were estimated in the order of 0.25-0.49 g L −1 , with the highest values corresponding to the long tube retention time simulation, as only in this run concentrations surpassed 0.40 g L −1 . On the other hand, protein concentration was in the range of 0.14-0.24 g L −1 , with no clear differences among conditions. These results are in agreement with the pattern of TC measurements, as higher values were also found for the long condition. The combination of carbohydrates and proteins, with a larger fraction of the former, suggests that the cells were producing extracellular polymeric substances (EPSs). EPSs are common and have been extensively studied in some species of cyanobacteria and microalgae, including red microalgae such as Porphyridium purpureum (Cruz et al., 2020;Pierre et al., 2019). Despite having a rigid cell wall, G. sulphuraria has been reported to retain the ability to form an extracellular mucilage matrix, although possibly at very low amounts and not always (Gaignard et al., 2019;Lang et al., 2020;Oesterhelt et al., 2008;Vis & Necchi, 2021;Weber et al., 2007). In fact, EPSs have been only recently measured in this species and their functions have not been investigated yet (Sun et al., 2021;Zhu et al., 2022). In other photosynthetic microorganisms, EPSs serve a plethora of functions, including providing structural support, external nutrient storage, or protection against environmental stresses (Flemming & Wingender, 2010).
T A B L E 2 Raw total carbon (Raw TC, C-g L −1 ), glucose (mg L −1 ), corrected total carbon (Corrected TC, C-g L −1 ), coproporphyrin III (COPROIII, mg L −1 ), carbohydrates (g L −1 ), proteins (g L −1 ) and calculated carbon content in the extracellular polymeric substance (EPS TC, C-g L −1 ) measured in the supernatant at the end of every batch from the long, medium and short tube retention time simulations. We did not observe more biofilm formation (data not shown) than former studies with the same strain in the same system (Abiusi et al., 2022). Additionally, no special treatment had to be performed to separate the EPS from cells rather than a simple centrifugation step, indicating that the produced EPS are released rather than cell-bound.  (Miyatake et al., 2014;Stuart et al., 2016). An external carbon and energy storage might help G. sulphuraria to endure periods of light scarcity, frequent in the environments that they colonize (Gross & Oesterhelt, 1999). In the dynamic environment of our scale-down study, cells underwent cycles of glucose feast and famine which could have triggered a nutrient storing response. Perhaps, rather than a monocausal origin, it is the combined the effect of glucose and oxygen fluctuations what triggers production of EPS. Further insight into EPS production in G.
sulphuraria is required to establish their specific function in this context.
Assuming an average carbon content of 44% for the carbohydrates and 53% for the proteins (Rouwenhorst et al., 1991), we can estimate carbon excreted in the form of these two macromolecules by simple addition. The results of the calculation are in the range of 0.18-0.29 Cg L −1 (  (1), special attention should be placed on keeping DO concentrations above limiting levels. This implies that a simplistic and straightforward adaptation of our current lab-scale DO control strategy, explained elsewhere (Abiusi et al., 2020a), might not be sufficient to achieve so. In the case of (2), carbon excretion is most likely triggered by glucose and oxygen fluctuations. In practice, this suggests that glucose and oxygen profiles within the reactor have to be kept as flat as possible. Hence, glucose feeding has to be tightly regulated and the control strategy should be fast enough to react to small changes in oxygen concentration.
No scale-down study is perfect, and thus it is important to acknowledge its constraints. Due to technical limitations in the automation of our setup, the employed pulse glucose feeding strategy was only implemented with constant and continuous illumination during the experiments. When scaling up the process outdoors, however, the microalgae will deal with day/night cycles.
During nighttime, there will not be photosynthetic oxygen production and therefore fresh air will need to be provided to compensate for respiration. The repercussion of nightly aeration and of day/night cycles per se in the mechanisms described here is uncertain. For Conceptualization, supervision, writing review and editing. recovery from ultrahigh-NH4+ industrial effluent with coproduction of high-protein biomass by photo-fermentation.

SUPPORTING INFORMATION
Additional supporting information can be found online in the Supporting Information section at the end of this article.