P. tricornutum (Pt4) (UTEX 646) was cultivated in an f/2 medium (Guillard and Ryther, 1962) in a photobioreactor (PBR) (FM150, Photon System Instruments, Czech Republic) operated in turbidostat mode (Table 1) using the company software (Version 0.7.14). The PBR was equipped with sensors for light transmission, chlorophyll (Chl) fluorescence, temperature-pH (Mettler Toledo InPro3253SG/120/PT1000), dissolved inorganic carbon (DIC) (Mettler Toledo InPro5000/12/120), and dissolved oxygen (Mettler Toledo InPro6800/12/120). The PBR setup is presented in Supplementary Figure 1. The culture was continuously supplied with 1.2 mL min⁻¹ of a mixture of air and CO₂ (2,000 ppm CO₂ total). After 5 days of stable growing conditions, the PBR was fed with f/2 medium containing 0.15 mM NaNO₃. We will refer to the day after the medium was changed as the Day from the Reduction of Nitrogen Supply (DRNS). The pH, temperature, DIC, and O₂ were continuously monitored. The (Chl) fluorescence yield measurements were carried on in a light-adapted state and in a dark-adapted state of each light-dark cycle (Table 1) (for the definition of the fluorescence parameters, refer to Supplementary Table 2).

The cultivation in turbidostat mode requires that the culture turbidity remains constant. Turbidity (OD735) was measured at 735 nm to avoid pigment contribution. To maintain the OD735 constant, a peristaltic pump (Photon Systems Instruments PP500) was actioned by the PBR central controller when OD735 increased over a threshold set at 0.151 to maintain a cell density of 4 10⁶ cells⁻¹. The medium in excess was pushed in a container and collected as an overflow (14 in Supplementary Figure 1). The daily volume of overflow allows for the estimation of the dilution rate. In fact, the time course of the cell density (x) is a function of the cell division rate (μ_s) and also of the dilution rate (D) (Equation 1) (Gresham and Hong, 2015).

In turbidostat mode, dxdt equals zero and the Equation 1 reduces to Equation 2.

The decrease in limiting nutrients induces a decrease in division rate that the automatic dilution compensates accordingly. Thus, the changes in the dilution rate reflect the division rate.

Cells were counted daily using a Neubauer hemacytometer according to Heydarizadeh et al. (2019).

Samples were sterilely collected (Supplementary Figure 1). For carbohydrate, protein, lipid, and pigment extraction, cells were centrifuged (3,000 × g, 10 min, 4°C). Pellets were washed 3 times with a phosphate-buffered solution (PBS) before storage at −80°C. The samples for RNA extraction and metabolite extraction were collected on glass filters (Whatman® glass microfiber filters, Grade GF/A, Merck, Germany) under a gentle vacuum (50 mmHg), immediately frozen in liquid nitrogen, washed 3 times with a PBS solution, and stored at −80°C. Samples for carbon and nitrogen determination were harvested on fiberglass filters under gentle vacuum (50 mmHg), washed 3 times with a PBS solution, and dried at 70°C for 48 h. The sampling volumes and quantities are reported in Supplementary Table 3.

The nitrate concentration in the medium was determined according to Carvalho et al. (1998). Briefly, the absorbance at 220 nm was measured on a centrifuged culture supernatant (Lambda-25, Perkin-Elmer, USA). A sodium nitrate (NaNO₃) aqueous solution was used for calibration.

Internal Nitrogen (N-int) and Carbon (C-int) pools were determined using the C7N elemental analyzer (Eager 300, Thermo Scientific, Massachusetts, USA) according to Heydarizadeh et al. (2019).

The total RNA was extracted from samples with the Spectrum™ Plant Total RNA Kit (Sigma-Aldrich, Missouri, USA). The purified total RNA extracts were processed with an RNAstable® LD kit (Sigma-Aldrich). Paired-end next-generation sequencing has been performed by the Annoroad company (Corporate HQ, Beijing, China) with an Illumina NextSeq 550AR Sequencer. The sequencing results and quality are reported in Supplementary Dataset 1. The raw data were processed with the Trimmomatic tool (version 0.36, Bolger et al., 2014) to remove adapters. A total of 45.9 million high-quality clean reads (96.15% clean reads) were aligned with the Hisat alignment software (version 2.1.0, Kim et al., 2019) on the high-quality reference transcriptome (NCBI assembly accession number: GCF_000150955.2). Expression abundances were quantified using Htseq-count (version 0.11.2, (Anders et al., 2015) and differentially expressed genes (FDR ≤ 0.001; |Log2(FC)| ≥ 1) identified using the R package DESeq (Bioconductor version 3.11, Love et al., 2014). Functional annotations of the transcriptome were carried out using Gene Ontology (GO) terms and enriched with the bioinformatics tool Blast2Go and a manual cross-reference between published data and online public databases (https://protists.ensembl.org/; https://mycocosm.jgi.doe.gov/; https://www.ncbi.nlm.nih.gov/). Heatmaps were generated with an ad hoc Python script based on the seaborn repository (gplots version 0.11.0, https://seaborn.pydata.org/index.html). The scripts can be found in the following repository: https://github.com/IlBarbe/SimpleSmallPythonScipts.

Neutral lipids were estimated by Nile red fluorescence according to Huang et al. (2019). Pellets were resuspended in dimethyl sulfoxide (DMSO) 25% and stained with 1.5 μL Nile red (Sigma-Aldrich, Saint-Quentin Fallavier, France) solution 1 mg mL⁻¹ in DMSO (100%). The 298 K fluorescence at 580 nm was excited at 530 nm (slits 5 nm) (Perkin Elmer LS-55, Villebon sur Yvette, France). DMSO diluted glyceryl trioleate (Sigma-Aldrich T7140-500MG, purity ≥99%) was used for calibration (Supplementary Figure 2).

Protein extraction was performed by mechanical lysis with the FastPrep-24™ 5G Homogenizer (MP Biomed, France). Sample pellets were mixed with glass beads (Merck, USA, 425-600 μm, acid-washed) and 50 μL ultrapure water for mechanical lysis (4 m s⁻¹ for 30 s). The sample was chilled in ice for 1 min. After a second lysis cycle, 950 μL of ultrapure water were added to the lysate and homogenized for 30 s. The lysate was centrifuged at 16,000 × g for 20 min at 4°C. Protein concentrations were determined using the Bradford method (Bradford, 1976) using a double-beam spectrophotometer (Perkin Elmer Lambda-25). Bovine serum albumin (Sigma-Aldrich, purity ≥ 98%) was used for protein quantification calibration (Supplementary Figure 2).

Carbohydrate quantification was performed according to a slightly modified Dubois' protocol (Dubois et al., 1956). Briefly, the extraction was performed by mechanical lysis with the FastPrep-24™ 5G Homogenizer (MP Biomed, France). Sample pellets were mixed with glass beads (Merck, 425–600 μm, acid-washed) and 1.5 mL of 0.05 M sulphuric acid-water solution. Extract-containing tubes were incubated at 60°C for 30 min, vortexing every 10 min, then chilled for 10 min on ice. The lysate was centrifuged at 16,000 × g for 20 min at 4°C. In glass tubes, 2.5 mL of H₂SO₄ (>95%) were added to 1 mL of extract supernatant with the subsequent addition of 250 μL of 3%, phenol-water solution (99%, Merck). The glass tubes incubated in a water bath at 95°C for 30 min and chilled on ice for 10 min. Colorimetric quantification has been performed reading the absorbance at 485 nm using a double-beam spectrophotometer (Perkin Elmer Lambda-25). Glucose (Merck, USA, ≥ 99.5%) was used for total carbohydrate quantification calibration (Supplementary Figure 2).

The energy stored in organic molecules (E_a) was obtained by converting the different energy storage fractions (i.e., neutral lipids, proteins, and carbohydrates) into energetic equivalents using their combustion energy (i.e., Lipids: 39,500 mJ mg⁻¹, Proteins: 24,000 mJ mg⁻¹, Carbohydrates: 17,500 mJ mg⁻¹) (Gnaiger, 1983; Aderemi et al., 2018).

The available energy is thus:

where E_{carbohydrates}, E_lipids, and E_proteins were calculated by multiplying the respective combustion energy by the quantity of corresponding macromolecules per 10⁶ cells.

Pigment extraction was performed by mechanical lysis with the FastPrep-24™ 5G Homogenizer (MP Biomed, France) under green ambient light (Schoefs, 2004). Sample pellets were mixed with glass beads (Merck, 425-600 μm, acid-washed) and 100 μL of pure acetone (Merck, ≥ 99.5%) for mechanical lysis (4 ms⁻¹ for 30 s). Samples were centrifuged at 16,000 × g for 5 min at 4°C and supernatants were collected in a new microtube. Lysis cycles were repeated until the pellets fade to white. Absorbance spectra of the pooled supernatants were recorded between 400 nm and 800 nm (Perkin Elmer Lambda-25). The Chl a, Chl c, and total carotenoids concentrations were calculated according to Heydarizadeh et al. (2017) using the following equations:

where v is the pigment extraction volume, V is the sampling volume, and l is the optical path.

Metabolomic analyses were performed on samples (i.e., CTRL 4, 5, 6, 7, 8 DNRS, and 10/11 DNRS) grouped on single sample. Metabolite extractions were performed according to Courant et al. (2013). Briefly, samples were plunged into 5 mL of a boiling mixture (ethanol-water, 75:25 v/v, 90–95°C) for 1 min, vigorously shaken for 3 s, and re-incubated for 1 min. The extraction process was then quenched by plunging tubes for 3 min in ethanol water solution (75:25 v/v) at −80°C. After filter removal, the tubes were centrifuged (4,500 × g, 5 min, 20°C). The supernatants were filtered on precombusted glass wool and collected in glass vials. The organic solvent phase was evaporated using a rotavapor (Rotavapor R-205 Evaporation System, Buchi; heating bath B-490, Buchi, Vacuum pump PC500, Vacuubrand) and lyophilized (Lyophilizer Christ Alpha 1-2 Plus, Christ; Vacuum pump E2M5, Edwards). The lyophilized extract was then resuspended with 1 mL methanol water solution (20:80 v/v) and stored at −80°C prior to filtration (mesh 0.2 μm). Liquid chromatography-high resolution tandem mass spectrometry (LC-HRMS/MS) analyses were performed with an Agilent 1290 Infinity II LC system coupled to a high-resolution time-of-flight mass spectrometer (Q-Tof 6550 iFunnel, Agilent technologies, CA, USA) equipped with a Dual Jet Stream® electrospray ionization source (positive mode). Detailed information on analyses parameters is reported in Supplementary Information 1. The Agilent Mass Hunter Workstation software (version B.07) was used to process the raw MS data, including the extraction of molecular features (MFs), generation of molecular formulas, library searching, and database searching. The identification of the most relevant entities was supported by MS information and searches in the METLIN database (http://metlin.scripps.edu) and in the Dictionary of Marine Natural Products (DMNP) library (Blunt et al., 2008; Wolfender et al., 2015).

The metabolomic analyses resulted in the detection of 4,267 compounds ranging from 0.1 and 2.8 kDa in size. To reduce the number of compounds to elucidate, the change in compound detected quantity as compared to the change in transcript abundance during the time course of the experiment. As a preliminary step, only compounds and transcripts which significantly changed in abundance, as compared to those under controlled conditions, at least at a time point, were retained. After this first screening, each compound pattern has been compared with each gene expression pattern. To be as stringent as possible, only the combinations which displayed a significant direct correlation were included (|slope|=1, from 0.95 to 1.05; p < 0.05). The results showed 419 compound-gene combinations which included 115 compounds. The 115 compounds were kept for further analyses and elucidation.

ANOVA was performed using the online freeware (https://acetabulum.dk/anova.html). Principal component analyses (PCA) were performed with a compiled python script exploiting pandas (version 1.2.1), numpy (version 1.19.5), matplotlib (version 3.3.3), and scikit-learn (version 0.24.1) packages. Propagation of uncertainties in data processing was carried out following the mathematical rules (Arras, 1998). Linear and nonlinear fitting analyses were performed with the free version of CurveExpert software available for download (Hyams, D. G., CurveExpert software, http://www.curveexpert.net, 2010).

Carbohydrates, lipids, and proteins are potential substrates for respiration and represent the three major macromolecule classes used to E_a. The kinetic of the E_a shows two increasing phases separated by a transient pause (ANOVA 95%, p = 0.8) (Figure 1C). The first increase depends on the transition phase and consists in the accumulation of both carbohydrates and neutral lipids (Figure 1D). A pause occurred at the onset of the NS phase with no significant E_a increase (ANOVA 95%, p = 0.8; Figure 1C). E_a accumulation resumed with the only energy accumulation in the form of neutral lipids (Figure 1D).

Soluble proteins cell quotas decreased gradually from the onset of the transition phase, negatively contributing to the energy storage (Figure 1D). In the present experiment, the major source of carbon as building blocks for these high-energy-containing molecules should have been DIC. However, DIC increased rapidly during the transition phase and has continued to increase in NS with a reduced rate (Figure 1E), while C-int displayed a small increase (9.3%), not coherent with the high lipid accumulation (Figures 1C,D). These observations suggest that the metabolic reorientation relies on carbon reallocation mechanisms rather than carbon fixation. To confirm the hypothesis of reduction in carbon fixation and bring evidence on the carbon reallocation theory, the photosynthetic activity and respiration activities were assessed.

In NRep, the cellular pigment quota remained stable and reduced to 37% of the maximum in NS conditions (Table 2). The reduction of the pigment cellular quota has direct effects on light energy harvesting and it can be a direct response to the changes in photosynthetic efficiency. Chlorophyll fluorescence was used as a proxy of photosynthetic efficiency (Roháček et al., 2008). Both the dark-adapted state (Φ_P0) and light-adapted state (Φ_II) decreased during the transition phase, reaching an average of 0.48 and 0.23 respectively (Figure 2A), indicating a reduced photochemical utilization of the light energy. The photochemical quenching (qP) that indicates the actual photochemical capacity of PSII in the light-adapted state decreased from 100% to 60–70%, suggesting a decrease in the capacity of reopening the PSII after excitation (Figure 2B). This generated a need to dissipate the excess of the absorbed energy as heat as revealed by the transient increase of the non-photochemical quenching (qN) from 5 to 15 DNRS (Figure 2C). A pattern similar to Φ_P0 was observed for the relative electron transport rate (rETR) (Figure 2D).

The daily average dissolved oxygen rapidly decreased at the onset of the transition phase (Supplementary Figure 4A). The decrease in dissolved oxygen has a negative linear correlation with the increase in DIC (Supplementary Figure 4B), suggesting that the factor influencing the DIC affects also the dissolved oxygen. Since a negative linear correlation was observed between DIC and photosynthetic performance (Supplementary Figures 5A,B), the same correlation analysis was performed between the dissolved oxygen and the same two photosynthesis parameters showing a positive relation in both cases (Supplementary Figures 5C,D). A decrease in dissolved oxygen is thus due to the decrease in its production by the reduced photosynthetic activity and its consumption by the respiratory activity.

To get information on the capability of the cell to recover from NS, a nitrogen pulse was applied to the culture after 31 DRNS to reach a final nitrate concentration of 0.35 mM. Supplementary Figure 6 shows the fast response of the cells to the sudden increase in N-ext. Within 24 h, cells remobilized the energy stored in lipids to transiently synthesized proteins and pigments, resuming cell division after a lag phase of at least 3 days as well.

On the mindset of investigating the effects and mechanisms underlying the NRep-NS transition, transcriptomic analyses were performed on 11 time points. To observe the changes induced by the reduction of nitrogen supply (RNS), the changes in gene transcript availability were expressed compared with the control NRep. Overall, 2,560 genes were differentially expressed in at least one DNRS (Supplementary Dataset 1). The PCA based on transcriptomic data (Figure 3A) exhibits a pattern very similar to the PCA based on the biochemical and physiological data (Figure 1A). However, they differ by the positioning of 15 DNRS that is hereby detached from the NS phase group, suggesting that important changes in gene expression occurred at this time point. For the sake of simplicity in the figure representation, five phases will be considered: (i) NRep, (ii-iii) 5 and 6 DRNS, (iv) NS, and (v) 15 DRNS. In the case of NRep and NS, the average transcript levels across the corresponding DNRS were calculated for each gene for heatmaps creation (Figures 4–6). To strengthen the information given by the transcriptomic analysis, metabolic profiling analysis was performed through LC-MS on samples corresponding to 4, 5, 6, 8, 10, and 11 DNRS, which all allowed the detection of 4,267 compounds (p-value ≤ 0.05) (Supplementary Dataset 2). Individual compound cell quota change kinetics showed that 2,867 compounds displayed a significant variation (|log2FC| ≥ 2) compared to controlled conditions (before DRNS) for at least one time point. A PCA analysis was performed taking into account these 2,867 compounds (Figure 3B), thus, resulting in a similar pattern as observed on a transcriptomic level. A little shift was observed corresponding to the translation of the transcript changes at the metabolic level, and a great change observed was also observed on days 10 and 11.

In nature, microalgae grow in very variable conditions and they may be submitted to reduced N availability that may eventually evolve into N starvation. As described above, the only modulated parameter is the N availability in the form of nitrate while all the other parameters are kept constant. The multidisciplinary approach allowed us to have a comprehensive view of the transition from NRep to NS. The broad picture of the metabolic reorientation, given by the PCA of transcriptomics, metabolomic and biochemical-physiological responses (Figures 1, 3), was crucial to highlight that the DRNS induces a shift from an equilibrium to another. As expected, an equilibrium is observed in NRep conditions since cells are not subjected to limitation. Interestingly, a second equilibrium is observed when NS is reached. Cell metabolism has been reoriented and no qualitative changes were observed. Independently from the timepoint, cells in the NS phase possess the same metabolic instruction set. The observed changes during the NS phase are, thus, only quantitative (e.g., increased neutral lipid content), while qualitative changes (i.e., metabolic reorientation) have occurred during the transition phases.

The N-int appears to be the driving force that guides the shift between NRep and NS. In fact, cells remained in NRep phase until the minimum N-ext was reached and cells were forced to utilize the N-int. Division rate and N-int are closely interconnected (Figure 1B). In fact, the pool of N-int is split between the two daughter cells during cell division and the lower N-int will affect the cell division rate (Alipanah et al., 2015). An equilibrium is reached when the actual N availability is enough to maintain a minimum level of cell division rate. Even if the parameter guiding the NRep to NS transition is N-int, cells are not unaware of the changing environmental conditions before the trigger point. Indeed, 69 genes were differentially regulated at 4 DRNS including PLAAOx and CSAP encoding genes and one transcription factor of the heat shock family (HSF) (Supplementary Dataset 1).

The time shift of the decay of N-int compared with N-ext (Figure 1) is due to the presence of an intracellular N storage pool (Shoman, 2015). Aside from organic N, present in proteins and other N-containing molecules, nitrate can be directly stored in the cells. Mccarthy et al. (2017) reported that cells possess a remarkable ability to rapidly transport, assimilate, and also, presumably, store NO3- in vacuoles. The transient decrease of NR encoding transcript during the transition phase in contrast with increased transcript levels of genes encoding NO3- transporters of both plasmalemma and tonoplast, confirms the hypothesis of the potential storage of NO3- in the vacuoles (Figure 4). A temporary decrease in NR has been also observed by Longworth et al. (2016) within 72 h of NS.

Nitrate was the only N source in the medium, however, the effort of the cell to scavenge N from other N-containing molecules, such as free amino acids, is suggested by the upregulation of the gene encoding the PLAAOx. This enzyme oxidizes the aminoacidic groups to release ammonium which can be readily exploited by the cell as a source of N. The early overexpression of the PLAAOx and CSAP encoding genes (Figure 4), which has also been observed in T. lutea (Garnier et al., 2016), suggests a potential involvement of PLAAOx and CSAP in N sensing. Moreover, an alternative source of ammonium can be photorespiration which has already been observed as taking place in N-limited conditions (Huang et al., 2015). The increase in transcripts encoding, not only mitochondrial GOGAT but also serine/alanine:glyoxylate aminotransferase and glycine decarboxylase, leads to the conclusion that in NS, a cellular effort in ammonia assimilation is observed following what was introduced by Remmers et al. (2018). Amidase activity is also contributing to ammonia scavenging, facing N deprivation (Alipanah et al., 2015).

The passage to NS induces exposes cells to an energy imbalance between light availability and demand from cellular metabolism and growth (Remmers et al., 2018). The gradual development of the chlorotic phenotype is consistent with the decrease in transcript abundance of the majority of genes encoding LHCs, which is also reflected in the optimization of antenna size. On the other hand, the increased LHCX4 and LHCR10 transcript abundance suggests the need to increase the dissipation of excess absorbed energy. These two proteins belong to distinct families of LHCs, reported playing important roles in photoprotection (Lepetit et al., 2012; Collier Valle et al., 2014; Taddei et al., 2016; Buck et al., 2019). Both LHCX4 and LHCR10 encoding genes have been reported to be upregulated under nitrogen deprivation (Yang et al., 2013; Alipanah et al., 2015). While LHCR10 has been shown to be involved in photoprotection in response to the low light-to-high light transition of P. tricornutum (Collier Valle et al., 2014), LHCX4 does not appear to be involved in the fluorescence quenching photoprotection (Buck et al., 2019, 2021). Buck et al. (2021) suggested that one gene alone cannot contain all required regulatory cis-elements needed to respond to the multitude of environmental triggers and, thus, LHCX4 protein may play an integrative role between N starvation and photoprotection. Hence, the optimization of antenna size resulted in a decrease of the energy flow through the photosynthetic reaction centers, limiting photodamages and NADPH and ATP productions, the two key cofactors of the Calvin-Benson-Bassham cycle. In NS conditions, ribulose cannot be transformed into 3-phosphoglycerate but could enter the photorespiratory cycle (Sharkey et al., 1988; Bailleul et al., 2015). This is demonstrated by the increase of DIC (also observed by Valenzuela et al., 2012), the lack of increase in cell carbon quota (Figure 1C), the changes in expression of genes encoding Calvin-Benson-Bassham, and photorespiration components (Supplementary Dataset 1). This interpretation is further supported by the upregulation of the genes coding the enzymes involved in the photorespiratory pathway such as the alanine:glyoxylate aminotransferase (J49601, Supplementary Dataset 1).

A reduction of the plastidial activities follows a reduction of its energy content. Despite being exposed to low nitrogen availability for longer than 25 days, P. tricornutum was capable of a fast (within 24 hours) recovery of cell activities when a N source was reintroduced (Supplementary Figure 6). This fast recovery is incompatible with a complete shutdown and degradation of the chloroplast. These results confirmed previous observations made by Valenzuela et al. (2012). In short, NS turns chloroplast metabolism into a stand-by mode and, hence, the cell metabolism. In this scenario, the energy available in the chloroplast is not enough to keep it functioning and the mechanisms of organelle crosstalk need to be established as demonstrated by Bailleul et al. (2015). Indeed, the increase in the transcription levels for the gene encoding components of the malate valve (2-oxoglutarate:malate translocator and malic enzyme) suggests a connection between the different cellular pyruvate hubs and the energy transfer from the TCA to the plastid. Interestingly, the activation of the pyruvate hub (PEPCK, PEPC1, and PYC1) has been described as a default answer to the energy unbalance (Heydarizadeh et al., 2019), allowing the chloroplast stand-by mode as described in this work. This stand-by mode is reminiscent of the quiescent state described in Chlamydomonas reinhardtii (Tsai et al., 2018) and Nannochloropsis oceanica (Zienkiewicz et al., 2020). Regardless of their denomination, quiescent state, or standby mode, these mechanisms seemed highly conserved among microalgae and are crucial for cell survival in unfavorable conditions, such as those that may occur in nature. Deciphering these mechanisms may contribute to speeding up the development of microalgae as cell factories for the production of metabolites.

Central carbon metabolism plays an important role in carbon partitioning mainly toward the synthesis of the major organic macromolecule classes (i.e., carbohydrates, proteins, and lipids), and plays a crucial role in response to environmental changes (Launay et al., 2020).

Different from TAG, the carbohydrate cellular quota is higher under NS but no continuous increase is observed (Figure 1D). In fact, Caballero et al. (2016) observed that P. tricornutum accumulates only 6% of cellular carbon in the form of chrysolaminarin in NS. In general, microalgae tend to break down storage carbohydrates in order to redirect carbon toward lipid biosynthesis (Hockin et al., 2012; Yang et al., 2013). Physiology and transcriptomic analyses showed a decrease in photosynthetic activity and carbon fixation through the Calvin-Benson-Bassham cycle, implying a decrease in de novo glucose generation and, thus, a reduced substrate for storage carbohydrates. The increase in carbohydrate cell quota, already reported in the literature (Caballero et al., 2016; Remmers et al., 2018), may suggest an increase in energy storage in the form of chrysolaminarin. However, according to the transcriptomic analysis, a decrease in chrysolaminarin biosynthesis and an increase in its breakdown could have occurred (Figure 6). The observed carbohydrate quota increase could be the effect of the active central metabolism suggested by the increase in transcript abundance of cytosolic glycolysis and TCA. The observed increase in carbohydrate quota is not continuous and occurred mostly during the transition phase. Thus, this increase is not due to an effort in carbohydrates storage but a carbon backbones remobilization and a central metabolism overactivity. The increase in transcript levels encoding two enzymes involved in gluconeogenesis, PEP carboxykinase (PEPCK) and pyruvate carboxylase 1 (PYC1), could suggest the neosynthesis of simple sugar and could explain the increase in carbohydrate cell quota. This hypothesis is corroborated by the observation of Matthijs et al. (2017) as an increase in maltotriose, maltose, and glucose in nitrogen-depleted P. tricornutum cells.

Remmers et al. (2018) stated that the central carbon metabolism response to NS is an intricate regulation of three oxidative pathways and gluconeogenesis. An increase in the transcript levels of the gene encoding components cytosolic classic glycolytic pathway is observed, suggesting an increase in glycolytic activity and confirming what was observed by Remmers et al. (2018) on both the transcriptomic and proteomic levels. Organelle counterparts are, on the other hand, downregulated. The increase in transcripts levels encoding components of the PPP, which produce NADPH (Kroth et al., 2008), has been observed. This pathway, together with chrysolaminarin degradation and glycolysis, suggested an increased effort of N starved cells to provide carbon fluxes through the TCA and balanced the reduced NADPH production from photosynthesis, as reported previously by Alipanah et al. (2015). The upregulation of genes encoding components of the TCA cycle also suggests an overactivity of answering the need of cells to compensate for the energy loss due to the reduced photosynthetic activity. Furthermore, TCA serves as a recycling of the carbon backbones derived from proteins and amino acids (Hockin et al., 2012). These catabolic activities produce oxaloacetate which can be converted in PEP by the activity of PEPCK or enter the malate shunt, contributing to the subcellular compartment energy balance (Selinski and Scheibe, 2019). Flügge et al. (2011) highlighted the crucial role of the PEP supply to plastids and its restriction, which may be eventually fatal for the plastid performance. Both the accumulation of TAG and carbohydrates, as well as protein degradation, are energy-requiring activities (Peth et al., 2013; Subramanian et al., 2013). Accelerating the central carbon metabolism allows the cells to both generate energy molecules and recycle carbon backbones during NS.

The gradual accumulation of lipids and carbohydrates starts at the onset of the transition phase (Figure 1D). To explain this result, two alternative and non-exclusive hypotheses can be proposed: either the carbon required for the production of these molecules comes from the photosynthetic carbon fixation or/and from internal carbon reallocation. The data presented in this manuscript are in favor of the second hypothesis because all the genes involved in the FAS cycle are drastically downregulated, which would result in a strong reduction of the carbon flux through the de novo lipid biosynthesis pathway. The reduction in de novo FA biosynthesis has been previously observed in literature which supports the hypothesis of a lipid reallocation rather than biosynthesis (Valenzuela et al., 2012; Alipanah et al., 2015; Remmers et al., 2018). This reduction is attested by the strong increase in DIC, together with a decrease in photosynthesis (Φ_II and qP). Further proof of the reallocation of the carbon, rather than a fixation, is also explained by the very small percentage increase in cell carbon quota (Figure 1C), which does not fit with a de novo lipid biosynthesis deriving from carbon fixation. Moreover, the de novo fatty acid biosynthesis requires the acyl carrier protein (ACP), which expression is downregulated. Thus, in the absence of active carbon fixation, cells have no other choice but to convert pre-existing membrane lipids such as PC and PE. The choline and ethanolamine groups are removed by the action of the PLD, of which the majority of genes are upregulated in NS. The upregulation of genes encoding for the DGAT fits with an increase in TAG synthesis utilizing acyl groups, mostly generated by the breakdown of the phospholipids by the activity of the PLA, which encoding genes are upregulated. On the same frame, PDAT enzymes directly transfer an FA from the phospholipids to the DAG, while DGAT attaches a free FA group to DAG molecules. The same happens to the MGDG and DGDG, which are also a source of DAG and FA. Despite no gene encoding galactolipases have been so far identified in P. tricornutum, both transcriptomic and metabolomic data indicate an increase in galactolipid catabolism, thus, releasing fatty acids for TAG synthesis through lipid recycling.