Metabolomic Responses of Arabidopsis Suspension Cells to Bicarbonate under Light and Dark Conditions

Global CO2 level presently recorded at 400 ppm is expected to reach 550 ppm in 2050, an increment likely to impact plant growth and productivity. Using targeted LC-MS and GC-MS platforms we quantified 229 and 29 metabolites, respectively in a time-course study to reveal short-term responses to different concentrations (1, 3, and 10 mM) of bicarbonate (HCO3−) under light and dark conditions. Results indicate that HCO3− treatment responsive metabolomic changes depend on the HCO3− concentration, time of treatment, and light/dark. Interestingly, 3 mM HCO3− concentration treatment induced more significantly changed metabolites than either lower or higher concentrations used. Flavonoid biosynthesis and glutathione metabolism were common to both light and dark-mediated responses in addition to showing concentration-dependent changes. Our metabolomics results provide insights into short-term plant cellular responses to elevated HCO3− concentrations as a result of ambient increases in CO2 under light and dark.

Zostera marina, the diurnal metabolomic changes were recorded in anoxic conditions 14 . Following an integrative approach of combining transcriptomic and metabolomic datasets into in silico metabolic models in rice, light-specific metabolic changes were observed 15 . Although HCO 3 − -responsive guard and mesophyll cell responses were cataloged in canola leaves 16 , nothing is yet known about the temporal metabolomic changes associated with differential HCO 3 − levels or under light and dark conditions. The use of single-cell types eliminates the 'averaging effect' of metabolomes that occurs with tissues, organs or whole plants 17 . In one such study using Chlamydomonas reinhardtii cells, 128 metabolites with significant differences between high-and low-CO 2 treated cells were detected, of which 82 included amino acids, lipids, and carbohydrates 18 . Recently, C. reinhardtii transcriptome and proteome were studied under varying CO 2 concentrations. Furthermore, it was observed that varying CO 2 concentrations had an effect on 25% of the transcriptome in C. reinhardtii 19 . Proteomic studies revealed the role of 22 extracellular proteins, which were only expressed under low CO 2 conditions 20 . In addition, transcriptional coordination of physiological responses in Nannochloropsis oceanica CCMP1779, a marine unicellular alga was recorded under light/dark cycles 12 . In another unicellular dinoflagellate (Symbiodinium sps.), the transcriptomic changes of cells in response to light intensity under different trophic conditions were investigated 21 . Furthermore, the diurnal transcriptome of C. reinhardtii undergoing cellular and metabolomic differentiation was captured 22 . Here we chose NaHCO 3 treatment as a CO 2 enrichment model system to investigate its effect on Arabidopsis cellular metabolome under varying concentration, duration of exposure, and light and dark conditions.

Results
The dark and light responsive metabolomes and their responses to HCO 3 − treatment. Our targeted high performance liquid chromatography (HPLC)-multiple reaction monitoring mass spectrometry (MRM-MS) approach allowed us to detect and quantify a total of 229 metabolites in A. thaliana suspension cells at various time-points across four biological replicates (Supplementary Tables S1-3). The polar to semi-polar metabolites covered various pathways of secondary metabolites, amino acids, purines, flavonoids, 2-oxocarboxylic acids, central carbon metabolites, glutathione, and beta-alanine metabolism that were spread over the KEGG metabolic map (Supplementary Figure S1). Another 29 metabolites were identified and quantified using the gas chromatography (GC) platform to include cyanoamino acid, pantothenate, fatty acids, beta-alanine, galactose, glycolysis, branched-chain amino acid, arginine, and proline metabolism among others (Supplementary Table S4). The HCO 3 − treated suspension cell metabolomes showed quantitative variations based on HCO 3 − concentrations (1, 3, and 10 mM), light and dark, and the time of treatment. Across the treatments, the shared metabolites showed differential levels of changes (fold change cut-offs of <0.8 and > 1.2 at p < 0.05) (Supplementary Table S3). The significantly changing metabolites revealed 156 changing metabolites (either increase or decrease), which are common to both light and dark conditions (Fig. 1A). These common metabolites are enriched in flavones, flavonol, and flavonoid biosynthesis, amino acid (Arg, Pro, Tyr, Ala, Asp and Glu) metabolism, alkaloid biosynthesis and glutathione metabolism. On the other hand, 28 significantly changed metabolites are enriched in pyridine biosynthesis and 32 in one carbon pool by folate, folate biosynthesis and amino acid metabolism. Concentration dependent metabolomic changes revealed shared pathways (i.e., flavonoid, alkaloid, glutathione metabolism), with 3 mM treatment induced the highest number (176) of significant changes enriched in pyrimidine and flavonoid metabolism (Fig. 1B). For 10 mM treatment, beta-alanine, Arg and Pro and Cys and Met metabolism showed enrichment. In addition, the commonly enriched pathways based on 79 commonly shared metabolites showed concentration-independent changes belong to flavone, flavonol and other flavonoid biosynthesis, purine metabolism, glutathione metabolism and amino acid metabolism (Ala, Asp, Glu, Cys, and Met) (Fig. 1B).

Grouped metabolomic changes in HCO 3
− treated cells. Within subject-ANOVA analysis using the four factors including treatment (control vs. HCO 3 − ), light conditions (light and dark), concentration (1, 3, and 10 mM), time-course (0-120 min) and interactions indicated significant changes (p < 0.05) in 89, 123, 228, 112 and 28 metabolites, respectively (Supplementary Table S5). Taurine metabolism was significantly affected by time, whereas flavonoid biosynthesis, glutathione and phenylalanine metabolism were associated with light dependent changes. Furthermore, both taurine metabolism and flavonoid metabolism were associated with HCO 3 − treatment-specific changes, whereas glyoxylate and dicarboxylate metabolism, Ala, Asp and Glu metabolism and taurine metabolism were responsive to interactions of all factors (Table 1). However, the highest numbers of metabolites were enriched in flavonoid biosynthesis, glutathione metabolism, beta-Ala metabolism, purine and pyrimidine metabolism, Phe and Tyr metabolism.
Hierarchical clustering analysis (HCA), performed to classify all metabolites based on treatments, light conditions, concentrations, and time-points revealed three distinct clusters (Fig. 2). The first cluster mostly reflected the increased metabolites under the dark conditions, while the last cluster reflected increased metabolites across all the concentrations and conditions. Similarly, when all samples were clustered by HCA as a function of metabolite responses, they were clustered into several regions of high and low correlations (Spearman) (Fig. 3A). For 3 mM treatment, all the samples under dark and light conditions correlated well among each other (R 2 < 0.9) in a condition specific manner, i.e., light-independent HCO 3 − response, whereas for 1 mM and 10 mM treatments, the correlations were low (Fig. 3A). Similarly, metabolite pair-wise (Spearman) correlation map indicated strong correlation among sugar and nucleotide phosphates and amino acids during the time-course study (Fig. 3B).  Table S6). Pathways enriched in the dark across the three concentrations include purine and pyrimidine metabolism, flavonoid biosynthesis, beta-alanine metabolism, tyrosine metabolism and alkaloid metabolism. This is completely opposite to what was observed under the light conditions, where 80, 65 and 63 metabolites (enriched with flavone, flavonol and other flavonoid biosynthesis, alkaloid metabolism, and tyrosine metabolism) showed increasing patterns with 1, 3, and 10 mM HCO 3 − , respectively. In contrast, the number of metabolites for the other two-patterns (i.e., bell-shaped with initial increase followed by decrease in later time-points and two-peaked patterns) showed consistent decreases. The bell-shaped pattern showed concentration-dependent increases in the number of metabolites i.e., 45, 48, and 64 metabolites for 1, 3, and 10 mM HCO 3 − , respectively, usually enriched with beta-alanine metabolism (Supplementary Table S6). While a few of the models demonstrated clear and linear patterns of changes, a majority of the metabolites showed mostly biphasic patterns with two peaks and two troughs.

Global metabolic and metabolite-specific responses to HCO 3
− treatment. In OPLS-DA, the supervised multivariate linear models revealed grouped and differential responses of the suspension cells to HCO 3 − with an interpretable visualization in addition to providing the important metabolites, which helped in differentiating the various treatments (groups). Upon OPLS-DA evaluation, the resulting score plots indicated great separation of control and HCO 3 − treatment (Fig. 5A), slight separation between dark and light-treated conditions (Fig. 5B), 3 mM standing out of the three concentrations (Fig. 5C), and a clear separation at 0 time point controls and the other time-points (Fig. 5D). In addition, flavonoids such as neohesperidin, isoliquiritigenin, and luteolin showed treatment and light-dependent changes in the time-course study ( Fig. 5A-D). GC-MS based profiling of primary metabolites revealed the contribution of individual metabolites in the loading plots, i.e., sugars (glucose, galactose, trehalose and melezitose), amino acids (Val, Ala, Gly), and fatty acids (stearic and palmitic acid) among others (Fig. 6A). OPLS-DA analyses based variable importance in projections (VIP) scores indicated the roles of sugars (glucose, galactose, trehalose), amino acids (Ala, beta-Ala, Gly, pyroglutamic acid), lactic acid, ethanolamine, boric and phosphoric acids in the HCO 3 − responsiveness (Fig. 6B). To further validate the models, we performed random permutation tests with the OPLS-DA model. In accordance with the OPLS-DA models, all the components and validation with 1000 random permutation tests generated intercepts with acceptable (> 0.9) variance (R 2 ) and the cross-validated variance (Q 2 ) values (> 0.85).

Discussion
Increases in atmospheric CO 2 concentrations provide additional resources for photosynthesis, which leads to high accumulation of photosynthate 10,23 . Carbonic anhydrase (CA) converts CO 2 into HCO 3 − for incorporation into cellular metabolism and acts as an upstream regulator of CO 2 24 . CA catalyzes the reaction , where rapid equilibration leads to increased concentration of CO 2 around RuBisCO. In addition, the fundamental function of PEPCase is to catalyze the conversion of HCO 3 − and phosphoenolpyruvate (PEP) to oxaloacetate. In dicots, PEPCases are activated by glucose 6-phosphate (G6P) and inhibited by malate or Asp 25 . Furthermore, the dark fixation of CO 2 /HCO 3 − is important for plant growth 26 . Our results showed wide-spread shifts in cellular metabolism at all four levels, i.e., in response to treatment (control and HCO 3 − ) (Fig. 6A), light conditions (dark and light) (Fig. 1A), concentrations (1, 3, and 10 mM) (Fig. 1B) and time-course (0, 5, 15, 30, 60, 120 min) ( Table 1). We observed large-scale HCO 3 − concentration-dependent metabolomic changes in the photomixotrophic A. thaliana suspension cells. In addition, transcriptomic reprogramming in response to modulation of light can occur in a matter of seconds in A. thaliana 13 . We observed more synchronized and comparable responses at 3 mM HCO 3 − as compared to low (1 mM) or high (10 Mm) HCO 3 − , possibly because those concentrations are either too low to elicit a cellular response or too high that they trigger stress responses and suppress other metabolic processes. Evidently, saturated photosynthesis takes place in soybean leaf mesophyll cells under light at 10 mM HCO 3 − 27 . Moreover, photosynthetic activities in photosystem II (PSII) are known to increase when treated with 1 mM HCO 3 − , but not when treated with 12 mM HCO 3 − , indicating that higher concentration may not be activating PSII, but only playing a protective role against excessive damage 28 . In addition, dose-dependency of HCO 3 − (increments from 2 mM to 5 mM HCO 3 − ) in inducing malate and citrate synthesis in plant cells is also known 29 . Thus, in this study the observed metabolomic responses not only reflect HCO 3 − dependent responses but also species-specific responses. Recently, microarray analyses conducted in moss (P. patens) gametophytes revealed that the expression levels of 814 genes were affected under elevated CO 2 , where most transcriptional reprogramming occurred in photosynthetic regulation, carbon metabolism, and stress responses 9 . In our metabolomic data sets, pyrimidine metabolism and folate metabolism showed unique responses to light and dark conditions, respectively, whereas flavonoid metabolism, glutathione, and alkaloid metabolism were enriched in both dark and light conditions. In fact, light induced phenylpropanoid metabolism and flavonoid biosynthesis in A. thaliana roots, as a high-irradiance response is well documented 30 . Increased folate metabolism in dark conditions indicates stress response as reiterated from the involvement of folate in seedling establishment of A. thaliana in dark 31 . Interestingly, lactic acid, a biomarker for concentration-dependent response to HCO 3 − , displayed the highest VIP score derived from PLSDA analyses. Lactic acid is a known indicator of hypoxia stress in maize roots 32 . Hence, lactic acid production is also indicative of anoxic and stressful conditions leading to anaerobic respiration in photosynthetic suspension cells in response to HCO 3 − . Similarly, induction of secondary metabolites, e.g., alkaloids by elevated CO 2 is known in many alkaloid producing plants 33 . Furthermore, in cyanobacterial model Cyanothece sp. PCC 7822, proteomics and transcriptomics indicated that proteins for nitrogenase and the pentose phosphate pathway were increased in the dark, whereas glycolysis and TCA cycle were more prominent in the light 34 . Furthermore, Glu, a proteogenic amino acid and indicator of protein abundance in plant cells displayed HCO 3 − concentration dependent accumulation patterns. We observed the levels of benzoic acid, a phenylpropanoid metabolism derivative, accumulated only in the dark, where its levels were depleted by HCO 3 − under light (Supplementary Table S4). Thus, not only does HCO 3 − perturbs light-dependent secondary metabolism, but also results in shifts in respiration, transcription (light-induced changes in pyrimidine metabolism and concentration-dependent changes in purine metabolism), amino acid (Ala and pyroglutamic acid with VIP > 1), sugar (glucose, galactose, trehalose with VIP > 1) and folate metabolism.
Significant changes in glutathione metabolism under light ( Fig. 2A) at both 1 mM and 10 mM HCO 3 − (Fig. 2B) indicate the involvement of redox regulation in presence of HCO 3 − (Table 1). Diurnal and light-mediated regulation of sulfur and glutathione metabolism in A. thaliana is known 35 . Upon HCO 3 − treatment, glutathione metabolism, an integral part of redox regulation showed changes in both light and concentration-dependent manner. Increased oxidative stress in C3 plants 36 and decreased ascorbate by high CO 2 11 further supports redox associated changes upon HCO 3 − treatment. Changes in taurine and flavonoid metabolism (Fig. 5, Table 1) with time further substantiate the role of antioxidants in amelioration of elevated HCO 3 − induced oxidative responses. In a recent study, up-regulation of genes encoding enzymes involved in oxidative signaling and ROS scavenging such as glutathione-S-transferase, peroxidases, catalases, Cys and thioredoxin-pathway were reported under elevated CO 2 9 . Thus, elevated CO 2 can enhance maintenance of the redox potential due to elevated rates of CO 2 assimilation and electron transport, as well as low photorespiration 37 . In a nutshell, cellular redox regulation is intricately associated with HCO 3 − responses, and eventually with elevated CO 2 conditions. In addition, using HCO 3 − feeding/enrichment as an alternative to elevated CO 2 , we show the usefulness of this system in A. thaliana suspension cell cultures which can be extended to other cell-types and plant cultures.
Increased levels of metabolites from glycolysis/gluconeogenesis, galactose, starch and sucrose metabolism along the time-course (Supplementary Table S3) are consistent with previous results of CO 2 induced accumulation of sugars and starch 38 . Although maintenance of a C/N balance in plants is complex and mediated by multiple mechanisms 39 , we observed metabolic flow in favor of primary metabolism with increased HCO 3 − . In contrast, metabolites showing decreases throughout the time-course belonged to fatty acid, vitamin B6, pantothenate metabolism among others (p < 0.1). Phe is involved in biosynthesis of secondary metabolites (phenylpropanoids and flavonoids). Its increase possibly indicated increased pathway fluxes to generate more defense-related metabolites. Furthermore, elevated CO 2 is known to associate with increased secondary metabolites, e.g., phenolics, flavonoids, alkaloids and terpenoids 38 . Increased secondary metabolism is a characteristic of light signal, which induce systems-wide changes in rice transcriptome 15 . Thus, HCO 3 − not only regulates primary metabolism, but also modulates secondary metabolism in plant cells indicating a genome-wide metabolic shifts in light-, concentration-, and time-dependent manner.
As time-resolved single-cell type metabolomic studies are rare, the concentration, light, and time-dependent HCO 3 − induced metabolomic changes reported here in A. thaliana suspension cells would help us understand CO 2 responsive metabolic changes. Typical cellular HCO 3 − induced responses are characterized by significant changes in flavonoid, glutathione, and Phe metabolism in light-and concentration-dependent manner, indicating functional and temporal behavior in the cells. Validation of these regulated metabolic pathways by means of traditional and omics approaches is essential for a systems biology perspective on HCO 3 − response. The results of this study have shown the utility of metabolomics tools towards improved understanding of the plant CO 2 sensing, utilization, and response mechanisms under light and dark conditions.

Materials and Methods
Plant materials and culture conditions. A. thaliana var. Landsberg erecta (LER) suspension cell cultures were obtained from Dr. Joshua Heazlewood (University of Western Australia, Australia) and maintained in 250 ml flasks under illumination of 130 μ mol m −2 s −1 on an orbital shaker with constant shaking at 120 rpm, 22 °C. They were sub-cultured weekly by tenfold dilution into fresh Murashige and Skoog (MS) medium 40 supplemented with 3% sucrose, kinetin (0.05 mg/L) and α -naphthalene acetic acid (NAA) (0.5 mg/L).

HCO 3
− treatment of suspension cells. The suspension cells were treated with NaHCO 3 at a final concentration of 1, 3, or 10 mM. For controls, NaNO 3 (NO 3 − is a macronutrient in MS media) was added in same concentration to nullify the effect of Na + during the treatment. Furthermore, as HCO 3 − induced significant pH changes of the MS media (pH 5.8), we added 50 mM (for 1 mM and 3 mM HCO 3 − ) and 100 mM (for 10 mM HCO 3 − ) MES buffer (pH 5.8) to the MS media, and the pH-stabilized MS media were used for both control and HCO 3 − treatments. In addition, the buffered MS media were sonicated to remove atmospheric gases prior to addition of HCO 3 − . Cells were incubated at 25 °C for 0, 5, 15, 30, 60, and 120 min on a shaker, and four replicates (n = 4) were generated for each data point after treatment with HCO 3 − (treatment) and control. After treatment, 100 ml cells at a concentration of 3 × 10 6 ml −1 were filtered using filter paper mounted on funnels, quickly blotted dry using Kim wipes and immediately snap-frozen in liquid nitrogen and stored in a − 80 °C freezer until metabolite extraction. Targeted profiling by HPLC-MRM-MS. The targeted metabolite profiling of the samples were performed using HPLC-MRM-MS as described 16 . An Agilent 1100 HPLC system (Agilent Technologies, Santa Clara, CA, USA) was used with an autosampler (Agilent, Santa Clara, CA, USA) to inject the samples. A C18-reverse phase analytical column (Gemini 5 μ; 150 × 2.0 mm, Phenomenex, Torrance, CA, USA) with the following gradient program helped separation of the compounds using two solvents i.e., 0.1% formic acid in H 2 O (solvent A) and 0.1% formic acid in acetonitrile (solvent B) for 1:99, v/v, at 0 min; 1:99, v/v, at 0.2 min; 99.5:0.5, v/v at 31 min; 99.5:0.5, v/v at 34 min; 1:99, v/v, at 34.2 min; 1:99 at 60 min at room temperature at a flow rate of 500 μ L/min. The HPLC was coupled to a hybrid triple quadrupole-ion trap 4000 Q-TRAP system equipped with a TurboIonSpray (TIS) interface (AB Sciex Inc., Foster City, CA, USA). The electrospray ionization (ESI) parameters were as previously described 16 , while the optimized MRM assay parameters are in Supplementary Table S1. QQQ scans were acquired as MRM experiments, where we monitored 59 (5.8 min), 40 (9.9 min), 45 (7.9 min), 31 (12.5 min), and 18 (23.8 min) MRM transitions over five periods (segments) in positive ionization, and 66 and 11 MRM transitions for two periods in negative ionization 42 . Each MRM transition was obtained with a dwell times specific for compounds, and a total cycle time was 1-1.5 s to ensure at least 10 data points across the peak. Only water served as blank solutions, while ddH 2 O was injected during washes. Pooled samples from the entire study served as quality control for monitoring the chromatography and MS conditions. Data imported using Analyst TM software version 1.5.1 (AB Sciex, Foster City, CA, USA) while peak areas were integrated and quantified using the IntelliQuan algorithm of the MultiQuant TM software version 2.1 (AB Sciex, Foster City, CA, USA).
Profiling of metabolomes by GC-MS. GC-MS was performed following Lisec et al. 43 , where 20 μ l of aliquoted extracts were dried and were sequentially derivatised with 100 μ l of 20 mg ml −1 methoxyamine hydrochloride (MeOX) in pyridine, followed by shaking at 55 °C for 90 min, then overnight incubation, which was followed by trimethylsilylation by adding 100 μ l of N-methyl-N-trimethylsilyl-trifluoroacetamide (MSTFA) to the mixture and incubation for 30 min at 37 °C, 30 min at 60 °C, and finally for 120 min at 100 °C to obtain optimized derivatization Samples were injected at a split ratio of 1:25 (250 °C) with a He carrier gas flow of 2 ml/min in a GC-MS system consisting of an autosampler, a TRACE TM 1310 gas chromatograph (Thermo Scientific, San Jose, CA, USA) in-line with a TSQ8000 Triple Quadrupole mass spectrometer (Thermo Scientific, San Jose, CA, USA) equipped with an electron ionization (EI) source. A TR-5MS capillary column (30 m × 0.25 mm × 0.25 μ m, Thermo Scientific, Waltham, MA) was used and the temperature program was essentially as described 16 . A standard n-alkane mixture (C8-C40) was injected at the beginning and end of the analysis for tentative identification and to monitor any shifts in retention indices (RI). The GC-MS data obtained from  Xcalibur ™ software version 2.0 (Thermo Finnigan, Austin, TX, USA) in raw format were aligned and processed as described 44 . Filtered raw GC-MS data sets comprised of data from four biological replicates (n = 4) and 29 curated analytes with less than 10% missing values, which were imputed using 1/3 rd of minimum values in each sample. Metabolite identification and quantification were performed as described 16 using MSRI spectral libraries of Golm Metabolome Database available from Max-Planck-Institute for Plant Physiology, Golm, Germany (http://csbdb.mpimp-golm.mpg.de/csbdb/gmd/gmd.html) 45 , Automated Mass Spectral Deconvolution and Identification System (AMDIS, National Institute of Standards and Technology, USA) and NIST Mass Spectral Reference Library (NIST11/2011; National Institute of Standards and Technology, USA) library.
Statistical analysis. The individual MS data were assembled into a concatenated data set into a single. csv sheet in Microsoft Excel (Microsoft Corp., Seattle, WA, USA) before statistical analyses using statistical software R (Version 2.9.1, R Development Core Team 2007, https://www.r-project.org/) 46 Logarithmic and median transformations of metabolite data were performed using DeviumWeb 47 and MetaboAnalyst 48 , and data were transformed, scaled, missing values imputed, and outlier removed before proceeding to uni-and multi-variate statistical analysis of the data. Raw processed data are available in Supplementary Table S2. ANOVA was performed using DeviumWeb 47 . p-values were adjusted by Benjamini-Hochberg correction (BH) 49 . Probability level, α was set to 0.05, adjusted for multiple hypotheses testing using BH to allow for a maximum 5% probability (q = 0.05) in false positive detection (FDR). Metabolite fold changes with cut-offs of 1.2 and 0.8, and significant changes (p < 0.05) were calculated (Supplementary Table S3). Principal component analysis (PCA) and orthogonal partial least square discriminant analysis (OPLS-DA) were performed using DeviumWeb 47 , where output consisted of score plots to visualize the contrast between different samples and loading plots to explain the cluster separation. The data files were scaled with unit variance without any transformations.