Identification of a transporter Slr0982 involved in ethanol tolerance in cyanobacterium Synechocystis sp. PCC 6803

Cyanobacteria have been engineered to produce ethanol through recent synthetic biology efforts. However, one major challenge to the cyanobacterial systems for high-efficiency ethanol production is their low tolerance to the ethanol toxicity. With a major goal to identify novel transporters involved in ethanol tolerance, we constructed gene knockout mutants for 58 transporter-encoding genes of Synechocystis sp. PCC 6803 and screened their tolerance change under ethanol stress. The efforts allowed discovery of a mutant of slr0982 gene encoding an ATP-binding cassette transporter which grew poorly in BG11 medium supplemented with 1.5% (v/v) ethanol when compared with the wild type, and the growth loss could be recovered by complementing slr0982 in the Δslr0982 mutant, suggesting that slr0982 is involved in ethanol tolerance in Synechocystis. To decipher the tolerance mechanism involved, a comparative metabolomic and network-based analysis of the wild type and the ethanol-sensitive Δslr0982 mutant was performed. The analysis allowed the identification of four metabolic modules related to slr0982 deletion in the Δslr0982 mutant, among which metabolites like sucrose and L-pyroglutamic acid which might be involved in ethanol tolerance, were found important for slr0982 deletion in the Δslr0982 mutant. This study reports on the first transporter related to ethanol tolerance in Synechocystis, which could be a useful target for further tolerance engineering. In addition, metabolomic and network analysis provides important findings for better understanding of the tolerance mechanism to ethanol stress in Synechocystis.


Introduction
Bioethanol production through microbiological processes has drawn greater attention in recent years due to increasing cost of non-renewable fossil fuels and environmental concerns related to over-utilization of fossil fuels (Hahn-Hagerdal et al., 2006). Various native or engineered microorganisms, such as Saccharomyces cerevisiae (Eiadpum et al., 2012;Yu et al., 2012), Zymomonas mobiles (Hayashi et al., 2012;Letti et al., 2012) and even Escherichia coli (Zhou et al., 2008;Manow et al., 2012) have been employed for ethanol production. As an alternative, photosynthetic cyanobacteria have been recently engineered by various synthetic biology tools into an "autotrophic microbial cell factory" to produce biofuels and fine chemicals directly from CO 2 using solar energy (Angermayr et al., 2009;Ducat et al., 2011;Ruffing, 2011;Machado and Atsumi, 2012;Oliver and Atsumi, 2014), which provides a complementary approach to the above heterotrophic microorganisms. In a recent study, by systematic evaluation and selection of alcohol dehydrogenase (adh) genes from different cyanobacterial sources and optimization of culturing conditions, Gao et al. (2012) engineered a synthetic Synechocystis strain able to produce ethanol levels of 212 mg/L per day and 5.50 g/L in 26 days (Gao et al., 2012). Although significant progress has been made, current ethanol productivity in cyanobacterial systems is still largely lagging behind the yeast systems (Antoni et al., 2007). Recent studies have suggested that one of the crucial factors responsible for the low ethanol productivity of cyanobacterium cells could be their very low tolerance toward ethanol (Dunlop, 2011;Jin et al., 2014). For example, in the model cyanobacterium Synechocystis sp. PCC 6803 (hereafter Synechocystis), the cell growth was arrested by up to 50%, and started to aggregate under 1.5% (v/v) ethanol (Qiao et al., 2012). In another study, Kamarainen et al. (2012) also evaluated tolerance of the cyanobacterial hosts to various biofuels, and found that the Synechocystis growth was less than 50% of the control in BG11 medium supplemented with 2.0 g/L ethanol (Kamarainen et al., 2012). The studies suggested that it is necessary to adopt engineering strategies to ethanol tolerance in cyanobacterial systems.
Many organic solvents are toxic to microorganisms. The toxicity of organic solvents depends on its log Pow (the partition coefficient between n-octanol and water) and final concentration of accumulation in the cell which contribute to increase permeability of the membrane. Readily water-miscible organic solvents such as ethanol, toxicity correlates directly with hydrophobicity (Ingram, 1981). In addition, ethanol has been shown to affect the proton motive force (Cartwright et al., 1987), and to increase leakage of metabolites from cells.
A membrane transport protein (or simply transporter) is a membrane protein involved in the movement of ions, small molecules, or macromolecules, such as another protein, across a biological membrane. Protein transporters for small molecules have been suggested as an important mechanism against ethanol toxicity (Ding et al., 2009;Stanley et al., 2010;Foo et al., 2014). For example, studies showed that the FPS1 gene, encoding the plasma membrane aquaglyceroporin that can facilitate transmembrane transport of small-uncharged molecules like polyols and urea, was up-regulated in response to ethanol exposure in yeast (Aguilera et al., 2006;Teixeira et al., 2009). Similarly, a recent functional genomics study with E. coli under exogenous nbutanol stress also found that transporters were among the most regulated functional components (Rutherford et al., 2010). Another recent study, involving heterologous expressing of 43 transporters in E. coli, showed that selected transporter could improve cell survival under stress of biofuels such as geraniol and limonene; however, none of the transporters improved tolerance for n-butanol and isopentanol . Nevertheless, these early studies suggested that utilization of transporters could be one of the approaches for engineering tolerance and biofuels production strains (Ding et al., 2009;Stanley et al., 2010;Dunlop et al., 2011).
Transporters have been reported for roles against many types of environmental stresses in cyanobacteria, such as arsenate resistance in Anabaena variabilis (Thiel, 1988), Cu 2+ resistance in Nostoc calcicola (Verma and Singh, 1991), salinity stress in Synechococcus sp. PCC 7942 and Synechocystis sp. PCC 6803 (Nomura et al., 1995;Mikkat et al., 1996), acid stress in Synechocystis sp. PCC 6803 (Tahara et al., 2012), and heavy metals in filamentous Oscillatoria brevis (Tong et al., 2002). Among all transporters in Synechocystis, ATP binding cassette (ABC) transporters residing in the inner membrane that are involved in the transport of a wide variety of substrates at the expense of ATP hydrolysis are the most common and well-characterized type (Davidson et al., 2008). For example, the slr1295, slr0513, slr0327, and sll1878 genes encoding polypeptides of an ABC-type ferric iron transporter that played a major role in iron acquisition have been identified (Katoh et al., 2001). However, so far no transporter involved in ethanol resistance has ever been reported in any cyanobacterial species.
In this study, to discover transporters related to ethanol tolerance, we constructed a mutant library of transporter genes in Synechocystis, and then screened them for tolerance changes under ethanol stress. The efforts led to the discovery of the slr0982 gene that may be involved in tolerance against ethanol stress in Synechocystis. In addition, to further explore the ethanol tolerance mechanism mediated by transporter Slr0982, we applied a metabolomics approach to comparatively analyze the differential cellular responses between the Synechocystis wild type strain and the slr0982 mutant (Wang et al., 2013). The study provides important findings for better understanding of the tolerance mechanism to ethanol stress in Synechocystis.

Bacterial Growth Conditions and Ethanol Treatment
Synechocystis sp. PCC 6803 and the knockout mutants constructed in this study were grown in BG11 medium (pH 7.5) under a light intensity of approximately 50 µmol photons m −2 s −1 in an illuminating incubator of 130 rpm at 30 • C (HNY-211B Illuminating Shaker, Honour, China) (Qiao et al., 2012). Cell density was measured on a UV-1750 spectrophotometer (Shimadzu, Japan) at OD 730 or on an ELx808 Absorbance Microplate Reader (BioTek, Winooski, VT, USA) at OD 630 . For growth and ethanol treatment, 40 µL fresh cells at OD 630 of 0.2 collected by centrifugation and were then inoculated into 200 µL of BG11 liquid medium in 96-well cultivation plates. Ethanol at a final concentration was added at the beginning of cultivation. The 96-well cultivation plates were fixed in the shaker and measured directly on ELx808 Absorbance Microplate Reader at OD 630 every 12 h. To ensure accuracy of finding, the mutants with differential growth patterns under ethanol stress were then confirmed by growth in 250-mL flasks, in which 10 mL fresh cells collected by centrifugation and were inoculated into 50 mL of BG11 liquid medium in a 250 mL flask. Culture samples (1 mL) were taken and measured at both OD 730 and OD 630 every 12 h. Growth experiments were repeated at least three times to confirm the growth patterns.

Construction and Analysis of Knockout Mutants
A fusion PCR based method was employed for the construction of gene knockout fragments (Wang et al., 2002). Briefly, for the gene target selected, three sets of primers were designed to amplify a linear DNA fragment containing the chloramphenicol resistance cassette (amplified from a plasmid pACYC184) with two flanking arms of DNA upstream and downstream of the target gene. The PCR strategy of mutant construction was shown in Supplementary Figure S1. The linear fused PCR amplicon was used directly for transformation into Synechocystis by natural transformation. The chloramphenicol-resistant transformants were obtained and passed several times on fresh BG11 plates supplemented with 10 µg/mL chloramphenicol to achieve complete chromosome segregation. The successful knockout mutants were confirmed by colony PCR and sequencing analysis. PCR primers for mutant construction and validation were listed in Supplementary Table S1. Comparative growth analysis of the wild-type Synechocystis and the mutants were performed in 250 mL flasks each with 50 mL BG11 medium of normal or ethanol.

Flow Cytometric Analysis
To reveal cell morphology differences, flow cytometric analysis was performed on a Calibur fluorescence-activated cell sorting (FACS) cytometer (Becton Dickinson) with the following settings: forward scatter (FCS), E00 log; side scatter, 400 V. Control and ethanol-treated cells were harvested at 24, 48, and 72 h, respectively, washed twice with phosphate buffer (pH 7.2) (Sigma-Aldrich), and then resuspended in the same phosphate buffer to a final OD 580 of 0.3 (approximately 1.5 × 10 7 cells mL −1 ). A total of 5 × 10 4 cells were used for each analysis according to the method by Marbouty et al. (2009). Data analysis was conducted using the CellQuest software, version 3.1 (Becton Dickinson).

Complementation of slr0982 in the slr0982 Mutant
Gene expressing vector pXT37b was kindly provided by Dr. Xuefeng Lu of Qingdao Institute of Bioenergy and Bioprocess Technology of Chinese Academy of Sciences (Tan et al., 2011). The promoter of plastocyanin (P petE ) was first replaced by that of phycocyanin beta chain (P cpcB ) in this study. The ORF of slr0982 was subcloned into NdeI/XhoI site of the modified pXT37b, resulting in pXT-slr0982. The primers sequences for slr0982 cloning were 5 ′ -CGCCATATGATGTCTGATACAGT CATTCGAGTGG-3 ′ and 5 ′ -CCGCTCGAGTCATGCAATTTT CTCCACATTCCAG-3 ′ . The pXT-slr0982 was introduced back into the slr0982 mutant by natural transformation. The slr0982complementation strain was named slr0982/pXT-slr0982. To achieve complete chromosome segregation, slr0982/pXT-slr0982 was passed several times on fresh BG11 plates supplemented with 10 µg/mL spectinomycin. Homologous integration of the expressing cassette and complete segregation were confirmed by PCR analysis.

LC-MS Based Metabolomics Analysis
(i) Sample quenching, extraction, and preparation: All chemicals used for LC-MS metabolomic analyses were obtained from Sigma-Aldrich (Taufkirchen, Germany). Cells were collected by centrifugation at 8000 × g for 8 min at room temperature (Eppendorf 5430R, Hamburg, Germany). The cell samples were quenched and extracted rapidly with 900 µL of 80:20 methanol/H 2 O (−80 • C) and then frozen in liquid nitrogen. The samples were then frozen-thawed three times to release metabolites from the cells. The supernatant was collected after centrifugation at 15,000 × g for 5 min at −4 • C and then stored at −80 • C. The remaining cell pellets were re-suspended in 500 µL of 80:20 methanol/H 2 O (−80 • C) and the above extraction process was repeated. The supernatant from the second extraction was pooled with that from the first extraction and stored at −80 • C until LC-MS analysis (Bennette et al., 2011); (ii) LC-MS analysis: The chromatographic separation was achieved with a SYnergi Hydro-RP (C18) 150 mm × 2.0 mm I.D., 4 µm 80 Å particles column (Phenomenex, Torrance, CA, USA) at 40 • C. Mobile phase A (MPA) is an aqueous 10 mM tributylamine solution with pH 4.95 adjusted with acetic acid and Mobile phase B (MPB) is 100% methanol of HPLC grade (Darmstadt, Germany). The optimized gradient profile was determined as follows: 0 min (0% B), 8 min (35% B), 18 min (35% B), 24 min (90% B), 28 min (90% B), 30 min (50% B), 31 min (0% B). A 14-min post-time equilibration was employed, bringing total run-time to 45 min. Flow rate was set as a constant 0.2 mL/min (Park et al., 2011). LC-MS analysis was conducted on an Agilent 1260 series binary HPLC system (Agilent Technologies, Waldbronn, Germany) coupled to an Agilent 6410 triple quadrupole mass analyser equipped with an electrospray ionization (ESI) source. Injected sample volume for all cases was 10 µL; capillary voltage was 4000 V; and nebulizer gas flow rate and pressure were 10 L/min and 50 psi, respectively. Nitrogen nebulizer gas temperature was 300 • C. The MS was operated in negative mode for multiple reaction monitoring (MRM) development, method optimization, and sample analysis. Data were acquired using Agilent Mass Hunter workstation LC/QQQ acquisition software (version B.04.01) and chromatographic peaks were subsequently integrated via Agilent Qualitative Analysis software (version B.04.00); (iii) Targeted metabolite analysis: a total of 24 metabolites were selected for LC-MS based targeted metabolite analysis in this study as described previously (Su et al., 2014). The standard compounds for these 24 metabolites were purchased from Sigma, and their MS and MS/MS experimental parameters were optimized with the mix standard solution. All metabolomics profile data was first normalized by the internal control and the cell numbers of the samples, standardized by average, log2 transformed and then subjected to Principal Component Analysis (PCA) using software SIMCA-P 11.5 (Laiakis et al., 2010). Heatmap were created using MultiExperiment Viewer software available publically at http:// www.tm4.org/.

GC-MS based Metabolomics Analysis
All chemicals used for metabolome isolation and GC/MS analyses were obtained from Sigma-Aldrich (Taufkirchen, Germany). For metabolomic analysis, cells of the wild type and the slr0982 mutant were collected from control and ethanol-stressed cultures at 24, 48, and 72 h, respectively. For each sample, cells equivalent to ∼10 8 cells, were collected by centrifugation at 8000 × g for 10 min at 4 • C (Eppendorf, Hamburg, Germany). The cell pellets were immediately frozen in liquid nitrogen and then stored at −80 • C before use. The metabolomic analysis protocol included: (i) Metabolome extraction: cells were resuspended in 1.0 mL cold 10:3:1 (v/v/v) methanol: chloroform: H 2 O solution (MCW), and frozen in liquid nitrogen and thawed for five times. Supernatants were collected by centrifugation at 14000 × g for 3 min at 4 • C. To normalize variation across samples, an internal standard (IS) solution (100 µg/mL U-13Csorbitol, 10 µL) was added to 100 µL supernatant in a 1.5-mL microtube before it was dried by vacuum centrifugation for 2-3 h (4 • C). (ii) Sample derivatization: derivatization was conducted according to the two-stage technique described by Roessner et al. (2001). The samples were dissolved in 10 µL methoxyamine hydrochloride (40 mg/mL in pyridine) and shaken at 30 • C for 90 min, then were added with 90 µL N-methyl-N-(trimethylsilyl) trifluoroacetamide (MSTFA) and incubated at 37 • C for 30 min to trimethylsilylate the polar functional groups. The derivate samples were collected by centrifugation at 14000 × g for 3 min before GC/MS analysis. (iii) GC-MS analysis: analysis was performed on a Agilent 7890A Gas Chromatograph (GC) coupled to a Agilent MSD 5975 system (Agilent Technologies, Inc., Santa Clara, CA, USA) equipped with a HP-5MS capillary column (30 m × 250 mm id). 2 µL derivatized sample was injected in splitless mode at 230 • C injector temperature. The GC was operated at constant flow of 1.0 mL/min helium. The temperature program started isocratic at 45 • C for 2 min, followed by temperature ramping of 5 • C/min to a final temperature of 280 • C, and then held constant for additional 2 min. The range of mass scan was m/z 38-650. iv) Data processing and statistical analysis: The mass fragmentation spectrum was analyzed using the Automated Mass Spectral Deconvolution and Identification System (AMDIS) (Stein, 1999) to identify the compounds by matching the data with Fiehn Library (Fiehn, 2002) and the mass spectral library of the National Institute of Standards and Technology (NIST). Peak areas of all identified metabolites were normalized against the internal standard and the acquired relative abundances for each identified metabolite were used for future data analysis. All metabolomics profile data was first normalized by the internal control and the cell numbers of the samples, log2 transformed, and then subjected to PCA using software SIMCA-P 11.5 (Laiakis et al., 2010).

Quantitative Real-time RT-PCR Analysis
The identical cultures used for metabolomics analysis were also used for RNA isolation and quantitative real-time RT-PCR (RT-qPCR) analysis of gene expression. RT-qPCR analysis was performed as described previously (Kloft et al., 2005;Wang et al., 2012). Quantification of gene expression was determined according to standard process of RT-qPCR that used serial dilutions of known concentration of chromosome DNA as template to make a standard curve. The 16s rRNA was used as an internal control. Three technical replicates were performed for each gene. Data analysis was carried out using the StepOnePlus analytical software (Applied Biosystems, Foster City, CA). Briefly, the amount of relative gene transcript was normalized by that of 16s rRNA in each sample (mutant or wild type), and the data presented were ratios of the amount of normalized transcript in the treatment between the mutant and the wild type. The gene ID and their related primer sequences used for real-time RT-qPCR analysis were also listed in Supplementary Table S1.

WGCNA Correlation Network Construction
Correlation network was created from the metabolomic data set, first by calculating weighted Pearson correlation matrices corresponding to metabolite abundance, and then by following the standard procedure of WGCNA to create the networks (Zhang and Horvath, 2005;Langfelder and Horvath, 2008;Wang et al., 2013). Briefly, weighted correlation matrices were transformed into matrices of connection strengths using a power function. These connection strengths were then used to calculate topological overlap (TO), a robust and biologically meaningful measurement that encapsulates the similarity of two metabolites' correlation relationships with all other metabolites in the network. Hierarchical clustering based on TO was used to group metabolites with highly similar correlation relationships into modules. Metabolite dendrograms were obtained by average linkage hierarchical clustering, while the color row underneath the dendgram showed the module assignment determined by the Dynamic Tree Cut of WGCNA. The network for each module was generated with the minimum spanning tree with dissimilarity matrix from WGCNA. The modules with correlation r > 0.5 or r < −0.5, and p-value less than 0.05 were extracted for further investigation. Hub metabolites were screened by high connectivity with other metabolites (≥5) in the modules strongly associated with genotypes based on correlation coefficient r > 0.5 or r < −0.5.

Pathway Enrichment Analysis
Metabolic pathway enrichment analysis of the responsive metabolites was conducted using the information from the KEGG (Kyoto Encyclopedia of Genes and Genomes) database according to the following formula (Su et al., 2014): N is the number of all metabolites within all KEGG pathways, M is the number of metabolites within a given KEGG pathway, n is the number of the responsive metabolites within all KEGG pathways, m is the number of the responsive metabolites within a given KEGG pathway. A p-value less than 0.05 was used as a cutoff for enriched KEGG pathways.

Screening for Ethanol Sensitivity in Transporter Mutant Library
A survey showed that at least 387 putative transporter-encoding genes were present in the Synechocystis genome (Kaneko et al., 1996). In this study, to uncover transporter genes involved in ethanol tolerance in Synechocystis, we excluded all transporter genes with well-characterized functions, such as bicarbonate or phosphate transporting. To this end, a total of 58 knockout mutants of transporter genes were constructed and the gene deletion was confirmed by PCR (Supplementary  Table S1). To achieve complete segregation, positive colonies were passaged through several cultivations on solid BG11 plates supplied with gradually increased chloramphenicol, and the full integration into all chromosomes was also confirmed by PCR (Supplementary Figures S1A-M). After 12 passages under selective pressure, full segregation was achieved for 54 transporter genes except for the slr0949, slr1224, slr1248, and sll1453 genes, suggesting that these transporter genes may carry essential roles for the cellular growth under the testing conditions. The mutants, in parallel with the wildtype Synechocystis, were then monitored for growth under various ethanol concentrations (i.e., 1.5, 1.8, and 2.0%, v/v) in 96-well cultivation plates. To ensure accuracy of the growth measurement, differential growth patterns under ethanol stress were also confirmed by growth in 250-mL flasks. The screening allowed the identification of slr0982 mutant, which was more sensitive to ethanol stress than the wild type. No significant tolerance change was found for other mutants under the tested condition. The full integration of slr0982 was shown in Supplementary Figure S1. As shown in Figure 1, in the normal BG11 medium, the slr0982 mutant grew equally well as the wild-type control, suggesting the knockout of slr0982 gene didn't affect cell growth under normal growth condition. However, in the BG11 medium supplemented with ethanol, although growth of both the wild-type strain and the slr0982 mutant was reduced, growth of the slr0982 mutant was reduced more significantly when compared with its growth in the normal BG11 medium (Figures 1A-C). In addition, the results showed that the ethanol sensitivity of the slr0982 mutant seemed concentration-dependent, as more growth reduction was observed in the medium supplemented with 1.8% than with 1.5% ethanol (Figures 1A,B). Taken together, the results showed that Frontiers in Microbiology | www.frontiersin.org the slr0982 mutant was more sensitive to ethanol stress than the wild-type cells, and the slr0982 gene may be involved in ethanol tolerance in Synechocystis.
To further confirm the involvement of gene slr0982 in ethanol tolerance, we constructed a slr0982-complementation strain in the slr0982 mutant, and the effect of the complementation was confirmed by comparatively measuring growth time-curve of relative strains. The results showed that in the normal BG11 medium, three strains grew almost equally; while in the BG11 medium with ethanol, the growth of slr0982/pXT-slr0982 was almost fully recovered to that of wild type, and significantly better than the slr0982 mutant under the same ethanol stress, demonstrating that slr0982 was involved in ethanol tolerance in Synechocystis (Figure 1).
The slr0982 gene, also designated as rfbB gene, was annotated as O-antigen (OAg) transport gene in the Cyanobase genome database (Nakao et al., 2010). In a recent study, Fisher et al. (2013) studied the possible function of slr0982 in Synechocystis, and the results showed no difference between OAg purified from the wild-type and the deletion mutant slr0982, although the bioinformatics inspection suggested the gene product slr0982 appeared to function in OAg transport; furthermore, the results showed that exopolysaccharides (EPS) purified from the slr0982 mutant was altered in composition when compared to the wild type, suggesting that the slr0982 gene may be involved in surface modification. In addition, analysis of the slr0982 mutant showed that it was deficient in EPS export compared to wild-type, suggesting the Slr0982 transporter may be involved in exporting of EPS (Fisher et al., 2013). The results were consistent with early studies in yeast and Z. mobilis that the composition of cell membrane and cell wall can significantly influence ethanol tolerance (Hermans et al., 1991;Ding et al., 2009). Interestingly, slr0982 seemed located within a putative operon, in which three hypothetical gene, slr0978, slr0980, and slr0981 without any known function were located upstream, where several genes involved in polysaccharide metabolism, slr0983 (glucose-1-phosphate cytidylyltransferase), slr0984 (CDP-glucose 4,6-dehydratase), and slr0985 (dTDP-4-dehydrorhamnose 3,5-epimerase) were located downstream of slr0982, respectively (Fisher et al., 2013). The results were also consistent with our previous RNA-seq transcriptomics analysis which showed that expression of slr0983 was differentially regulated by ethanol exposure , further suggesting that the slr0982 gene and the slr0978-slr0985 gene cluster might be involved in ethanol tolerance.

Analysing Cell Morphology of the Wild-Type and the slr0982 Strain
Cell morphology of the wild-type Synechocystis and the slr0982 mutant under both control and ethanol-spiked conditions was also compared using microscopic inspection and flow cytometric analyses. Although no difference was observed with microscope, the flow cytometric analyses showed that the cell size of the slr0982 mutant was slightly increased under ethanol stress when compared with wild type. The size increase became more obvious after longer stress treatment at 48 and 72 h when compared with 24 h after ethanol stress (Figure 2). Many types of microbial cells are surrounded by thick exopolysaccharide chains, which are part of the network maintaining cell size, as it has been reported that altered EPS production occurred (i.e., less large EPS band while more small EPS band in the slr0982 mutant) (Fisher et al., 2013), so it is probably not surprising that the change of exopolysaccharides synthesis will affect cell size and cellular response to stress. For example, a glycogen excess E. coli mutant TR1-5 was found to exhibit effects on cell size and surface (adherence) properties (Romeo et al., 1993). surface area, which is directly related to cell size. And Y-axis was normalized pixel count. WT-C and WT-E: wild type grown with or without 1.5% (v/v) ethanol stress. slr0982-C and slr0982-E: the slr0982 mutant grown with or without 1.5% (v/v) ethanol stress.
Targeted LC-MS Metabolomic Analysis of the Wild-Type and the slr0982 Strain LC-MS based metabolomics analysis has been recently applied to studies of cyanobacterial metabolism (Bennette et al., 2011;Schwarz et al., 2013), due to its advantages toward chemically unstable metabolites, such as the redox active nucleotides (NADPH, NADH) and the hydrolytically unstable nucleotides (ATP, GTP, cAMP, PEP) that are crucial for all major metabolic pathways. Using an optimized protocol of sampling, chromatography and mass spectral detection (Wang et al., 2014b), the LC-MS metabolomic analysis allowed successful monitoring of the intracellular levels of 24 metabolites in Synechocystis, including intermediates in central carbon metabolism, cellular energy charge and redox poise (Bennette et al., 2011). We eventually established reproducible analyses for 24 key metabolites in the central metabolism in Synechocystis, including AcCOA, ADP, ADP-GCS, AKG AMP, ATP, COA, DHAP, FBP, F6P, FUM, GAP, G6P, GLU, NAD, NADH, NADP, NADPH, UDP-GCS, OXA, PEP, 3PG, R5P, and RiBP involved in the key metabolic pathways. Using them as references, a semi-quantitative characterization of all 24 metabolites from all Synechocystis cell samples is achieved. With the optimized LC-MS protocol, we collected the cells of the wild type and the slr0982 mutant grown in BG11 media (control) and ethanol 1.5% (v/v) at three time points (i.e., 24, 48, and 72 h). Each sample consisted of three biological replicates (Supplementary  Table S2). To comparatively evaluate the effects of gene knockout on Synechocystis metabolism, we generated PCA score plots of metabolomic profiles of the wild-type strain and the slr0982 mutant through the growth time course under ethanol stress (Supplementary Figures S2A-C). Analysis of the score plots revealed obvious clustering patterns of three biological replicates for each sample, suggesting good analytical quality. In addition, the metabolomic profiles of the mutant can be visibly separated from those of the wild type at both control and ethanol stress conditions, suggesting good analytical resolution which can distinguish different metabolic status in the cells. Moreover, a slightly decreasing difference between the wild type and the slr0982 mutant under ethanol stress condition was observed, probably due to the fact that the cells started aging after a longer time of ethanol-stress treatment, so that metabolomic profiles of cells were more a reflection of cell aging rather than that of ethanol stress response.
Heat maps were created for all 36 LC-MS metabolomic profiles from three time points (i.e., 24, 48, 72 h) (Figure 3). In the analysis, the ratio of a given metabolites was calculated between the concentration of the metabolite under a given condition and the average concentration of the metabolite in all samples at each time point. At 24 h, under control condition without ethanol stress, very similar metabolite abundance patterns were found for almost all metabolites between the wild type and the slr0982 mutant, consistent with their similar growth pattern in the normal BG11 media. However, when the wild type grown under control and ethanol stress was compared, the results showed that a majority of metabolites were found up-regulated under the ethanol stress condition, more significantly for AcCOA, NADPH, NADP, NADH, NAD, ADP-GCS, ATP, ADP, G6P, F6P, and R5P, which were key FIGURE 3 | Heat map analysis of LC-MS metabolomics. The row displays metabolite and the column represents the samples. Metabolites significantly decreased were displayed in green, while metabolites significantly increased were displayed in red. The brightness of each color corresponded to the magnitude of the difference when compared with average value. WT-C and WT-E: wild type grown with or without 1.5% (v/v) ethanol stress. slr0982-C and slr0982-E: the slr0982 mutant grown with or without 1.5% (v/v) ethanol stress. co-enzyme or bioenergetics molecules participating in many important metabolic pathways, such as glycolysis, pentose phosphate pathway and TCA cycle, and well-known to have crucial roles in stress response. As an example, it has been reported that NADPH as an important coenzyme participates in ethanol tolerance in yeast, where overexpression of NADPHdependent alcohol dehydrogenase (ADH6) significantly increased the 5-hydroxymethylfurfural tolerance (Petersson et al., 2006). Glucose-6-phosphate dehydrogenase that catalyzes the conversion from G6P and NADP to 6-phospho-D-glucono-1,5-lactone and NADPH was responsive to radiation stress in Synechococcus lividus (Conter et al., 1987). Ribose-5phosphate isomerase that catalyzes the conversion between ribose-5-phosphate (R5P) and ribulose-5-phosphate (Ru5P) was found differentially regulated under oxidative stress conditions in photosynthetic green alga Chlamydomonas reinhardtii (Zaffagnini et al., 2012). Sucrose-phosphate synthase (SpsA, Sll0045) that uses F6P as substrate to form sucrose 6-phosphate is a key enzyme to synthesize one major compatible solute, sucrose, against salt stress in Synechocystis (Curatti et al., 1998;Desplats et al., 2005;Klahn and Hagemann, 2011).
At 48 and 72 h, under control condition without ethanol stress, differences in terms of metabolite abundance patterns were observed between the wild type and the slr0982 mutant, suggesting that after a long time of ethanol stress treatment, cellular metabolism started to diverge even though their growth patterns were similar. In addition, such a difference seemed more obvious at 72 h than 48 h (Figure 3). In the wild type between control and ethanol stress conditions, similar patterns of increased abundances of key molecules were found in the cells under ethanol stress condition, although the increases seemed less significantly, especially at 72 h, which was probably due to the accelerated cell aging and death after longer time of ethanol stress. However, in the slr0982 mutant cells at 48 and 72 h, the abundances of AcCOA, ADP-GCS, ATP, ADP, G6P, F6P, and R5P were enhanced when compared with 24 h (Figure 3). Taken together, the LC-MS targeted metabolomic analysis suggested that the slr0982 deletion could affect or delay the up-regulation of several key metabolites responsive to ethanol stress, which may eventually contribute to the tolerance loss in the slr0982 mutant (Figure 1).
To validate results from LC-MS analysis, 6 genes were selected for quantitative RT-PCR analysis; they are glucose-6phosphate isomerase (slr1349), 6-phosphofructokinase (sll0745), phosphoribulokinase (sll1525), phosphopyruvate hydratase (slr0752), citrate synthase (sll0401) and glutamate dehydrogenase (slr0710). The genes were selected because their enzymatic substrates or products were differentially regulated between the slr0982 mutant and the control under ethanol stress condition, as revealed by the metabolomic analysis (Figure 3). The gene ID and their related primer sequences used for real-time RT-qPCR analysis were listed in Supplementary Table S1. Comparative RT-qPCR analysis showed these genes were down-regulated 1.5-2.6 folds under ethanol stress in the mutant when compared with the wild type (Table 1). In general, good agreement between the gene expression analysis and the LC-MS metabolomic analysis was observed: down-regulation of slr1349 and sll0745 genes was consistent with the decreased abundance of FBP, while down-regulation of sll1525, slr0752, and slr0710 was consistent with the decreased abundances of RiBP, PEP and Glu, respectively.

Untargeted GC-MS Metabolomic Analysis of the Wild-Type and the slr0982 Strain
Following the optimized extraction protocol established previously (Krall et al., 2009), the cells of the wild type control and the slr0982 mutant grown in BG11 media with and without 1.5% (v/v) ethanol stress were collected at three time points (i.e., 24, 48, and 72 h) and the further MS analysis allowed the chemical classification of a total 48 metabolites from Synechocystis, including various amino acids, sugars, and organic acids detected in almost all replicate samples (Supplementary  Table S3). Score plots of PCA were generated to compare the mutant and the wild type in different conditions at three time points (i.e., 24, 48, and 72 h). The time-series metabolomic data showed that: (i) clustering of the biological replicates for each sample was clearly observed, suggesting overall good quality of the GC-MS analysis; (ii) interestingly, although no difference was observed for the wild type and the slr0982 mutant in terms of cell growth in the control media, clustering analysis based on their intracellular metabolites showed that they were clearly separated even at 24 h (Figure 4), indicating that the metabolic difference between the wild type and the slr0982 mutant, consistent with the above results from the LC-MS metabolomics (Figure 3); (iii) through the time course from 24 to 72 h under ethanol stress condition, metabolomic profiles of the slr0982 mutant were gradually moving more close to those of the wild type, probably due to the increasing cell aging and death after long ethanol stress (Figure 4).

WGCNA Analysis of Metabolomic Profiles Associated with the Ethanol Tolerance
To identify metabolic modules and hub metabolites related to ethanol tolerance in the slr0982 mutant, a correlationbased WGCNA analysis was applied to analysis to the GC-MS metabolomic datasets (Zhang and Horvath, 2005;Langfelder  and Horvath, 2008). The network analysis was not applied to LC-MS metabolomic data due to their relatively small data size. In this study, we constructed unsigned networks using a GC-MS metabolomic data set consisted of 48 metabolites, and then localized the correlated metabolites into various metabolic modules. In addition, the association of each distinguished metabolic module with mutant or ethanol stress treatment was determined. Setting a minimal number of metabolites in any module greater than 3, our WGCNA analysis showed that 6, 8, and 7 distinct metabolic modules can be detected within the metabolic networks at 24, 48, and 72 h, respectively (Supplementary Figure S3). Using a cutoff of correlation coefficient (r value) greater than 0.5 or less than −0.5, and their statistical confidence (p-values) less than 0.05, the analysis showed that a total of 4, 4, and 2 distinguished metabolic modules were highly associated with mutant or ethanol stress condition at 24, 48, and 72 h, respectively (Supplementary Figures S4A-C). At each time point, six modules (i.e., M1, M2, M3, M6, M8, and M9) were associated only with ethanol stress, three modules (i.e., M4, M5, and M7) were only associated with the knockout of the slr0982 gene, and M10 module was associated with both the ethanol stress and the knockout of the slr0982 gene, respectively. Metabolites included in each of the highly associated modules were presented in Table 2.
Using a criterion that at least two of its member metabolites identified from any given pathway, a pathway enrichment analysis was conducted and the results showed eight pathways were enriched in various distinct modules ( Table 3). Among them, the results showed that "Tryptophan metabolism" (Map00380) was enriched with statistical significance p-values less than 0.05 in ethanol-associated module M2. Enrichment of "Tryptophan metabolism" in ethanol-associated modules was consistent with early studies in E. coli and Saccharomyces cerevisiae, in which tryptophan biosynthetic pathway was previously found up-regulated in the ethanol-tolerant E. coli strains by a microarray analysis, and supplementation of tryptophan to the culture medium increased the specific growth rate of E. coli under ethanol stress (Horinouchi et al., 2010), and overexpressing tryptophan biosynthesis genes in yeast resulted in a stress tolerance to 5% ethanol (Hirasawa et al., 2007). In addition, "Biosynthesis of unsaturated fatty acids" (Map01040) was enriched in ethanol-associated module M3, and "Fatty acid biosynthesis" (Map00061) was enriched in both ethanol-associated module M3 and M6, consistent with the welldefined roles of fatty acid, especially unsaturated fatty acids in membrane modification against ethanol stress in various microbes (Heipieper and De Bont, 1994;Chiang et al., 2008). One well-described change is the shift from cis to trans unsaturated fatty acids to decrease membrane fluidity, resulting in a corresponding increase in solvent tolerance (Dunlop, 2011). In a previous study with cyanobacteria, the acyl-lipid desaturase (desA) gene from Synechocystis was expressed in prokaryotic E. coli and eukaryotic Solanum tuberosum cells, which led to an enhanced cold tolerance due to increased unsaturated fatty acid concentration in their lipids (Amiri et al., 2010). Our previous RNA-seq transcriptomics analysis also showed that slr1350 encoding acyl-lipid desaturase was upregulated by ethanol exposure in Synechocystis .
Enrichment analysis showed that "Biosynthesis of secondary metabolites" (Map01110), "Chloroalkane and chloroalkene degradation" (Map00625), "Benzoate degradation" (Map00362) and "Aminobenzoate degradation"(Map00627) were enriched with statistical significance p-values less than 0.05 in module M4, M9, M10, and M10, respectively ( Table 2). The "Biosynthesis of secondary metabolites" pathway contains many metabolites known to be related to stress conditions, such as D(+) trehalose involved in salt stress in cyanobacteria (Hagemann, 2011), ethanol resistance in E. coli (Wang et al., 2013) and S. cerevisiae (Zheng et al., 2013;Wang et al., 2014a), and Lisoleucine and L-glutamic acid involved in ethanol tolerance in E. coli (Horinouchi et al., 2010;Wang et al., 2013). "Benzoate degradation" (Map00362) and "Aminobenzoate degradation" (Map00627) were enriched due to the same two metabolites, benzoic acid and benzene-1,2,4-triol. However, their roles in ethanol tolerance are still unclear.  Within metabolic networks, hub metabolites that are involved in a high number of reactions are typically biologically important (Pfeiffer et al., 2005). Assuming a cutoff of connectivity greater than 5 in the networks as hub metabolites, a few hub metabolites were identified from the metabolic network constructed by the WGCNA, including sucrose and L-pyroglutamic acid in the module M4 at 24 h and stearic acid in the module M6 ( Figure 5). While M6 was associated with only ethanol stress, the role of stearic acid along with several connected fatty acid may be associated with membrane modification against ethanol stress (Dunlop, 2011). M4 was associated with the slr0982 knockout event ( Table 2). As discussed above, sucrose was well studied as compatible solutes against salt stress in cyanobacteria (Hagemann, 2011). As osmoprotectants, pyroglutamic acid was found to accumulate in response to salt stress and function as an osmoprotectant along with sucrose, in the halotolerant methanotroph Methylobacter alkaliphilum (Trotsenko Iu and Khelenina, 2002). In addition, these two compouds were also found to be involved in ethanol tolerance in Synechocystis (Zhu et al., 2015).

Conclusions
By constructing and screening a mutant library of 58 transporterencoding genes, a gene encoding an ABC transporter, slr0982, was identified to be involved in ethanol tolerance in Synechocystis. As the first ethanol-tolerance related transporter found in cyanobacteria, the gene could be valuable as a target of ethanol tolerance engineering for high-efficiency production in cyanobacterial systems. Although no evidence showed that Slr0982 is directly involved in the export of ethanol, the analysis suggested that it may affect transporting of other molecules important to the function and maintance of cytoplasmic membrane or extracellular structure.
To further decipher the ethanol tolerance mechanism mediated by transporter Slr0982, comparative LC-MS and GC-MS based metabolomic analyses were applied to determine the variation of intracellular intermediates in the wild type and the slr0982 mutant grown in BG 11 media with or without ethanol stress. The PCA and WGCNA analyses were used for analyzing the metabolomic data set in order to identify specific metabolites responsive to ethanol stress between mutant and wild type. The results showed that metabolites such as AcCOA, ADP-GCS, ATP, ADP, G6P, F6P, and R5P, and metabolic modules such as "Tryptophan metabolism, " "Biosynthesis of unsaturated fatty acids" and "Biosynthesis of secondary metabolites" might be related to ethanol tolerance in cyanobacterium Synechocystis, and further analysis the mechanisms that these metabolites and modules may represent could be valuable for tolerance in Synechocystis. Taken together, this study provided new insights into ethanol tolerance in cyanobacterium Synechocystis.

Author Contributions
LC and WZ conceived of the study. YZ, LC, and WZ drafted the manuscript. YZ carried out the mutant construction, overexpression, phenotypic and RT-PCR analyses. GP carried out the metabolomic data analysis. XN and XZ carried out the LC-MS analysis. MS and XZ carried out the GC-MS analysis. All authors read and approved the final manuscript.