ARTICLE. Omega-3 Production by Fermentation of Yarrowia lipolytica: From Fed-Batch to Continuous. Introduction

Transcription

1 ARTICLE Omega-3 Production by Fermentation of Yarrowia lipolytica: From Fed-Batch to Continuous Dongming Xie,,2 Edward Miller, Pamela Sharpe, Ethel Jackson, Quinn Zhu DuPont Industrial Biosciences, Experimental Station, Wilmington, Delaware 9880; telephone: þ ; fax: þ ; 2 Department of Chemical Engineering, University of Massachusetts-Lowell, One University Avenue, Lowell, Massachusetts 0854 ABSTRACT: The omega-3 fatty acid, cis-5,8,,4,7-eicosapentaenoic acid (C20:5; EPA) has wide-ranging benefits in improving heart health, immune function, and mental health. A sustainable source of EPA production through fermentation of metabolically engineered Yarrowia lipolytica has been developed. In this paper, key fed-batch fermentation conditions were identified to achieve 25% EPA in the yeast biomass, which is so far the highest EPA titer reported in the literature. Dynamic models of the EPA fermentation process were established for analyzing, optimizing, and scaling up the fermentation process. In addition, model simulations were used to develop a two-stage continuous process and compare to singlestage continuous and fed- batch processes. The two stage continuous process, which is equipped with a smaller growth fermentor (Stage ) and a larger production fermentor (Stage 2), was found to be a superior process to achieve high titer, rate, and yield of EPA. A two-stage continuous fermentation experiment with Y. lipolytica strain Z7334 was designed using the model simulation and then tested in a 2 L and 5 L fermentation system for,008 h. Compared with the standard 2 L fed-batch process, the two-stage continuous fermentation process improved the overall EPA productivity by 80% and EPA concentration in the fermenter by 40% while achieving comparable EPA titer in biomass and similar conversion yield from glucose. During the long-term experiment it was also found that the Y. lipolytica strain evolved to reduce byproduct and increase lipid production. This is one of the few continuous fermentation examples that demonstrated improved productivity and concentration of a final product with similar conversion yield compared with a fed-batch process. This paper suggests the two-stage continuous fermentation could be an effective process to achieve improved production of omega-3 and other fermentation products where non-growth or partially growth associated kinetics characterize the process. Biotechnol. Bioeng. 206;9999: 5. ß 206 Wiley Periodicals, Inc. KEYWORDS: Yarrowia lipolytica; omega-3 fatty acids; fed-batch fermentation; continuous fermentation Current address of Dongming Xie is Department of Chemical Engineering, University of Massachusetts-Lowell, One University Avenue, Lowell, MA 0854, USA. Correspondence to: D. Xie Received September 206; Revision received 28 October 206; Accepted 0 November 206 Accepted manuscript online xx Month 206; Article first published online in Wiley Online Library (wileyonlinelibrary.com). DOI 0.002/bit.2626 Introduction Omega-3 fatty acids refer to the long-chain polyunsaturated fatty acids (LCPUFA) with the first C55C double bond at the n-3 position, that is, the third carbon from the methyl end of the carbon chain. There have been many clinical studies showing a wide range of health benefits from the omega-3 LCPUFAs, especially eicosapentaenoic acid (C20:5; EPA) and docosahexaenoic acid (C22:6; DHA) (Chacon-Lee and Gonzalez-Marino, 200; Kapoor and Patil, 20; Martins et al., 203; Oikawa et al., 2009; Swanson et al., 202; Yokoyama et al., 2007). In human nutrition EPA and DHA are largely obtained from the diet, especially cold-water oceanic fishes (Martins et al., 203). EPA and DHA are synthesized de novo in marine microorganisms and phytoplankton. Some marine fishes (e.g., wild salmon, Pacific sardine) can accumulate significant amounts of EPA and DHA by consuming microalgae cells. Fish oil is the main source of EPA and DHA in the human diet, but its availability and sustainability have been questioned due to overfishing and probable contamination in the ocean environment. To overcome this limitation, biotechnology industries started to produce DHA directly from microalgae in large-scale fermentation processes (Kyle, 200). However, there is no large scale land-based EPA production from wild type organisms, because EPA productivity is too low to be economically commercialized. Consequently, DuPont initiated a research program to develop a land-based, sustainable EPA source by metabolic engineering of Yarrowia lipolytica (Xie et al., 205). Desaturase and elongase genes have been introduced into the oleaginous yeast to synthesize omega-3 fatty acids under aerobic fermentation conditions (Xue et al., 203). Though EPA has been the focus of this research program, the metabolic engineering of Y. lipolytica provides a platform technology to produce either tailored omega-3 (EPA, DHA) or omega-6 (ARA, GLA) fatty acids for a variety of applications (Zhu and Jackson, 205). Triglyceride production in wild type Y. lipolytica is developmentally regulated as a global response to nitrogen starvation in the presence of excess carbon source (Papanikolaou and Aggelis, 20; Ratledge, 2002). Published fed-batch fermentation processes utilizing this organism have retained this two-stage developmental feature of cell growth followed nitrogen limitation to induce triglyceride production (Xie et al., 205). Most industrial ß 206 Wiley Periodicals, Inc. Biotechnology and Bioengineering, ol. 9999, No. xxx, 206

2 fermentation processes are batch or fed-batch fermentation operations due to ease of operation, minimized risk of contamination, high product titer, flexibility for end-product choice, and/or sophisticated scale-up experience (Dombek and Ingram, 987; Xie et al., 200a, b). Although batch and fed-batch fermentation technology was mainly developed in the middle of last century, they typically have significantly lower volumetric productivities when compared to more traditional reaction engineering technologies. Continuous biomanufacturing processes have attracted more attention as a means to improve volumetric productivities (Li et al., 20; Menzel et al., 997; Ni and Sun, 2009; Taylor et al., 995). The productivities in most reported continuous fermentation processes were improved, but typically at the expense of product concentration, conversion yield, or both (Ethier et al., 20; Mutschlechner et al., 2000). The disadvantages in product concentration and/or yield, plus the increased effort to avoid contamination, led to very few successful applications of continuous fermentation in large scale production. Continuous fermentation can be either single-stage or multi-stage depending on target molecule production kinetics. Single-stage continuous fermentation is suitable for growth associated bio-production processes, that is, the product is biomass or a non-biomass product with production rate proportional to growth rate (Shuler and Kargi, 2002). However, most fermentation products including the production of secondary metabolites, and omega-3 fatty acids are only partially growth associated or even non-growth associated. Continuous processes for non-growth associated kinetics typically employ multiple fermenters to partition production into discrete segments such that the conditions for cell growth and product formation can be separately optimized. Though progress has been made in the research of overflow multi-stage continuous fermentation systems to improve productivities, decreases in product titers and/or conversion yields were still seen when compared to the standard batch or fed-batch results (Chang et al., 20; Gapes et al., 996; Koller and Braunegg, 205; Mutschlechner et al., 2000). In this paper, a two stage continuous fermentation process for the production of EPA by recombinant strains of Y. lipolytica was developed and compared against a standard fed-batch process. Dynamic models of EPA production using fed-batch and continuous fermentation data were established and used to analyze, optimize, design, and scale-up both the fed-batch and two stage continuous processes. This modeling approach developed in this paper was an important means to optimize performance beyond extrapolation from limited experimental data and addresses one of the fundamental limitations to continuous process development observed in the literature. A two-stage continuous fermentation process was developed based on model simulation results to significantly improve EPA productivity without losses in EPA titer and conversion yield, which was different from what has been seen in most literature of continuous fermentation, that is, the gain of productivity by a continuous operation was typically accompanied by losses in either product titer, yield, or both. Finally, a representative two-stage continuous fermentation experiment for EPA production by Y. lipolytica was conducted at 2 5 L scale for 6 weeks and successfully validated the model predictions. Materials and Methods Strains The Y. lipolytica strains Z609 and Z7334 in this study were derived from the wild type strain ATCC #20362 (Mat A). Both strains included cloned desaturases and elongases from various host organisms for the synthesis of EPA (C20:5), and mutation in the peroxisome biogenesis gene for further improved lipid and EPA production (Zhu et al., 204). Fed-Batch Fermentation Protocols Seed Culture The seed vials of Y. lipolytica strains Z609 and Z7334 stored at 80 C were thawed for 5 min at room temperature. Inocula were prepared by transferring ml vial solution to a 250 ml shake flask containing 50 ml seed culture medium, which consisted of Difco TM yeast nitrogen base without amino acid (6.7 g/l), Bacto TM yeast extract (5 g/l), KH 2 PO 4 (6.0 g/l), Na 2 HPO 4 2H 2 O (3.3 g/l), D-Glucose (20.0 g/l), MgSO 4 7H 2 O (.5 g/l), and thiamin HCl (.5 mg/l). The seed cells were grown in shake flasks for 8 24 h at 30 C, 300 rpm in an Innova Shaker Incubator until an OD 600 of 2 5 was reached. The seed culture was used to inoculate the 2 5 L fermentors at 3 5% (v/v). Fed-batch Fermentation The shake-flask seed culture (50 ml, OD 600 ¼ 2 5) was transferred to a 2 L fermentor (Biostat B, Sartorius BBI, Germany) to initiate the fermentation (t ¼ 0 h). The initial fermentation medium was.0 L and contained Bacto TM yeast extract (30 g/l), (NH 4 ) 2 SO 4 (22 g/l), KH 2 PO 4 (6.0 g/l), Na 2 HPO 4 2H 2 O (3.3 g/l), D-Glucose (50.0 g/l), MgSO 4 7H 2 O (.5 g/l), thiamin HCl (.5 mg/l), trace metals (00) (2.0 ml/l), and Antifoam 204 (Sigma,.0 ml/l). The trace metal (00) stock solution contained citric acid (5 g/l), CaCl 2 2H 2 O (.5 g/l), FeSO 4 7H 2 O (0.0 g/l), ZnSO 4 7H 2 O (0.39 g/l), CuSO 4 5H 2 O (0.38 g/l), CoCl 2 6H 2 O (0.20 g/l), and MnCl 2 4H 2 O (0.30 g/l). It was filtersterilized through 0.22 mm sterile membrane and stored at 4 C. The po 2 level of the fermentation experiments was set at 25% of air saturation by cascade controls of agitation speed between 300 and,200 rpm and pure oxygen enrichment (if needed). The aeration rate was fixed at 0.5 L/min or otherwise specified. The temperature was maintained at 30 C throughout the run. The ph value was controlled at 5.5 during 0 24 h and then increased to 7.0 and maintained at 7.0 in the remainder of the run by feeding KOH (56% w/v). Glucose (700 g/l) feeding commenced when its concentration in fermentation medium decreased below 20 g/l. Glucose concentrations were maintained at about 20 g/l during the entire process by adjusting glucose feed rate based on off-line glucose measurements. Two-Stage Continuous Fermentation Protocols A schematic diagram of the two-stage continuous fermentation system is shown in Figure. Seed culture and the sampling 2 Biotechnology and Bioengineering, ol. 9999, No. xxx, 206

3 Figure. Two-stage continuous fermentation for omega-3 production by Y. lipolytica. protocols were the same as for the fed-batch fermentation experiments. The two-stage continuous experiment was typically run for 6 weeks by using the following described protocols: Step : Operating Stage-2 Fermentation in Fed- Batch Mode (0 20 h) The shake-flask seed culture (3 50 ml, OD 600 ¼ 2 5) was transferred to a 5 L fermentor (Biostat B-Plus, Sartorius BBI, Germany) to initiate the fermentation (t ¼ 0 h). The initial working volume was 3.0 L with a medium containing 20 g/l Bacto TM yeast extract and 5 g/l (NH 4 ) 2 SO 4. All other medium components are exactly the same as described in the fed-batch protocols. The po 2 level of the fermentation experiments was set at 25% of air saturation by cascade controls of agitation speed between 300 and,200 rpm and pure oxygen enrichment (if needed). The aeration rate was fixed at.5 L/min or otherwise specified. The temperature was maintained at 30 C during 0 72 h and then lowered to 28 C after 72 h. The ph value was controlled at 5.5 for t ¼ 0 24 h and then at 7.0 for t > 24 h by feeding KOH (56% w/v). Glucose (700 g/l) feeding commenced when its concentration in fermentation medium decreased below 20 g/l. Glucose concentrations were maintained at 2 5 g/l during the entire process by adjusting glucose feed rate based on off- line glucose measurements. Step 2: Operating Stage- Fermentation in Fed- Batch Mode (72 20 h) Three days after the Stage-2 fermentation being started (t ¼ 72 h), the Stage- fermentation (2 L, Biostat B-Plus, Sartorius BBI) was initiated by inoculation with one flask of seed culture (50 ml, OD 600 ¼ 2 5). The initial working volume was.0 Lwith the same medium as described in the fed-batch protocols. The po 2 level was set at 25% of air saturation by cascade controls of agitation speed between 300 and,500 rpm and pure oxygen enrichment (if needed). The aeration rate was fixed at 0.5 L/min. The temperature was maintained at 30 C. The ph value was controlled at 5.5 during h (based on the Stage-2 elapsed time) and then increased to 7.0 in the rest of run by feeding KOH (56%w/v, or 0 N). The Stage- glucose feed contained 600 g/l glucose, 30 g/l Amberex 003 (yeast extract), 22 g/l (NH 4 ) 2 SO 4, 6 g/l KH 2 PO 4, 3.3 g/l Na 2 HPO 4 2 H 2 O, 3 ml trace metals (00), 7.5 mm MgSO 4, and 2.3 mg/l thiamin HCl. The Stage- glucose feeding commenced when the residual concentration in Stage- fermentor decreased below 20 g/l. Glucose concentrations were maintained within 2 5 g/l by adjusting the feeding rate. Step 3: Operating the Two-Stage Continuous Fermentation (20,008 h) After 20 h in Stage 2, the two- stage continuous fermentation was started by continuously pumping broth from Stages to 2 and pumping out the broth product from Stage 2 vessel to a storage tank (Fig. ). Temperature, ph and po 2 continued to be controlled at the same values in the rest of the run. The working volume was controlled at.5 L for Stage- and 4.0 L for Stage 2, respectively. The working volume for each stage of fermentor was maintained by fast running the withdrawing pump while carefully placing the end of withdrawing tube right above the liquid level. The glucose feeding rate was controlled at about 30 ml/min for Stage and 25 ml/min for Stage 2, respectively. After 95 h in Stage 2, the feed for Stage was also changed to 700 g/l glucose to consume all the residual nitrogen in Stage- fermentor. The feeding rates for both stages were also frequently adjusted to maintain the residual glucose between 2 and 5 g/l. The continuous operation was continued until t ¼,008 h. The fermentation broth from both Stages and 2 tanks were harvested as part of final product. Analyses of Fermentation Samples Fermentation samples (5 ml for each) were removed two times a day (8 6 h intervals). In general,.0 ml fresh broth sample was quickly frozen and stored at 80 C for lipid analysis, 0. ml was used for OD 600nm analysis after a :,000 dilution, 5 mlwas used for determination of dry cell weight (DCW, converted to biomass unit or unit/l), and 2 ml was used for glucose and possible byproduct (e.g., organic acids) analysis after collecting the supernatant by centrifugation (4,300g, 4 C, 5 min). The YSI-2900 Biochemistry Analyzer was used for glucose assay. The protocols for DCW and fatty acid analyses were described previously (Christie, 2003; Hong et al., 202; Zhang et al., 20). Xie et al.: Two-Stage Continuous Fermentation for Omega-3 3 Biotechnology and Bioengineering

4 Isolation of High-Lipid Producers From the Continuous Experiment The fresh samples from the vessel 2 of the two-stage continuous experiment were streaked out on Y. lipolytica minimal plates to obtain single colonies. Dozens of single colonies were selected based on possible growth phenotypes and regrown in test-tube cultures each containing 5 ml Y. lipolytica minimal medium (YMM). The YMM contained Difco TM yeast nitrogen base without amino acid (6.7 g/l), D-Glucose (20.0 g/l), and thiamin HCl (.5 mg/l). Cells in the test tubes were grown at 30 C, 300 rpm in the Innova 40R shaken incubator for about 24 h until the OD 600nm reached About ml culture solutionwas taken from eachtest tube to make a glycerol stock vial by mixing with ml sterilized glycerol solution (50%w/w), which was then stored in a 80 C freezer. The remaining culture solution of each test tube was used to inoculate a 25 ml shake flask containing 25 ml flask fermentation medium (FFM). The FFM contained Bacto TM yeast extract (5 g/l), D-Glucose (20.0 g/l), MgSO 4 7H 2 O (.5 g/l), KH 2 PO 4 (6.0 g/l), Na 2 HPO 4 2H 2 O (3.3 g/l), and thiamin HCl (.5 mg/l). Cells in the flask experiments were grown at 30 C, 300 rpm in the shaken incubator (Innova 40R) for 8 24 h until the OD 600nm reached 0 or above. After that, cells from each flask were harvested by centrifugation (Eppendorf 580) at 4 C, 4,300g for 5 min. Cells were then resuspended in 25 ml high-glucose medium (HGM) and incubated at 30 C, 300 rpm (Innova 40R) for additional three days for lipid and EPA production. The HGM contained D-Glucose (80.0 g/l), KH 2 PO 4 (6.3 g/l), and K 2 HPO 4 (27 g/l). Samples from the end of flask experiments were taken for final OD 600nm, glucose, and lipid analyses, as described in the fermentation protocols. Results and Discussion Omega-3 Production by Fed-batch Fermentation The omega-3 fed-batch fermentation process was a typical two-phase process, which started as growth phase for the first h and then followed by the oleaginous phase (or production phase) for the remaining period (Xie et al., 205). Nitrogen for cell growth was provided from both yeast extract (organic nitrogen source) and (NH 4 ) 2 SO 4 batched inthe initial medium. After the provided nitrogen was consumed, fermentation was transitioned to the oleaginous or production phase. Two metabolically engineered Y. lipolytica strains, Z609 and Z7334, were used in the fed-batch fermentation study. Under the typical fed-batch fermentation conditions as described in the experimental protocols, both strains had a very similar maximum specific growthrate(m max ¼ 0.26 h ), and produced very similar amounts of lipid and EPA in the production phase (Fig. 2). The experiments employed a 2 L Sartorius BBI fermentor system. To meet the high oxygen uptake requirement under high cell density conditions, this fermentation system was equipped with a maximum Figure 2. Comparison between experimental data and model predictions of the fed-batch fermentation results for Y. lipolytica strain Z609 and Z Biotechnology and Bioengineering, ol. 9999, No. xxx, 206

5 ,200 rpm agitation and pure oxygen enrichment of up to 50% of total aeration, which successfully provided a high oxygen transfer rate (OTR) capacity to support the maximum oxygen uptake rate (OUR) of nearly 300 mmol/l/h. With this high-otr capacity, the dissolved oxygen level was well-maintained at about 25% during this fed-batch fermentation of Y. lipolytica. Both Z609 and Z7334 strains were then able to produce lipid as 50% (w/w) of biomass and 50% (w/w) of total lipid as EPA inthe end of the fed-batch fermentation (Fig. 2). The EPA content in biomass (25%, w/w) was so far the highest among all reported in the literature. Modeling of Fed-Batch Fermentation The fed-batch experimental data were further used to build a mathematical fermentation model (see Appendix A for more details), which includes the rate equations for the eight state variables (cell density X, glucose S, nitrogen N, dissolved oxygen O, byproduct C, lipid L, EPAE, and working volume ), one control variable (feed rate for glucose F S ), and one derived variable (oxygen uptake rate OUR) of the fermentation. The kinetic parameters in the model equations (Table I) were obtained by the method of least squares. This was achieved by using the data analysis tool, Solver, in Excel to minimize the difference between all the measured experimental data and the model predicted data. The BA (isual Basic for Applications) tool in Excel was used to compose all the codes required for integrating the differential equations and performing the model simulations. Critical factors such as nitrogen content, dissolved oxygen level, glucose concentration, and cell density effect were considered in the model to describe the dynamics of cell growth, lipid accumulation, EPA production, and byproduct formation. As shown in Figure 2, the model fits well to the experimental data. The dynamic fermentation model has become a powerful tool to analyze the omega-3 production under various fermentation conditions and to guide the fermentation optimization and process scale-up. Model Simulation of Single-Stage and Two-Stage Continuous Fermentation The fed-batch fermentation model (Appendix A) was further modified to simulate a two-stage continuous fermentation process by considering the continuous feeding-in (substrate, nutrients, and Table I. Model Parameters for Y. lipolytica Z609 and Z7334. Parameter Unit alue Parameter Unit alue K ic g/g.5 r E 0.3 K in g/l r L 0.78 K ix unit/l 52 Y C/S unit/g 0.89 K N g/l Y L/S unit/g 0.47 K O % Air 0.65 Y X/N unit/g 27.0 K O2 % Air 5.50 Y X/S unit/g.52 K S g/l a E K SE h a L 0.07 K SL h 0.02 b C,max h m S g/unit/h 0.02 m max h 0.26 base) and the withdrawing of broth products in each stage of the fermentation. More details of the continuous fermentation model are described in Appendix B. The goal of model simulation of the continuous EPA production was to find opportunities for significantly improving EPA productivity without sacrificing EPA titer in lipid and EPA conversion yield. With the help of the fed-batch and continuous fermentation models, the EPA production by Y. lipolytica Z609 (or Z7334) under various fed-batch, singlestage continuous, and two-stage continuous fermentation conditions were compared by simulation results without the necessity of conducting so many time-consuming and laborintensive fermentation experiments. The simulation results and the comparison are shown in Table II. For the fed-batch simulation, only the standard medium and glucose feed were considered, as described in the fed-batch experiment protocols. For the simulation of all single-stage and two-stage continuous fermentation cases, nitrogen content in glucose feed was the growth limiting factor and it varied in case studies by changing the C/N mass ratio. However, no nitrogen source (yeast extract and ammonium sulfate) was included in the glucose feed for either fed-batch fermentation or the second stage of the two-stage continuous fermentation since EPA production is more favored under extremely nitrogen-limited conditions. The residual glucose concentrations for all the simulated continuous cases were set at 5 g/l or lower to minimize the waste of glucose and make the best use of the feed. Though only the two-stage continuous model was described in Appendix B, the single-stage continuous cases were simulated by considering only the Stage- in the two-stage model. All the continuous data shown in Table II represent steady-state values. Two different levels of OUR capacity were considered in the fedbatch simulation (Table II, Entry and 2). The fed-batch case # represents a standard control for all the simulation cases, and was validated by experiment, as previously discussed in the fed-batch results (Fig. 2). Fed-Batch Case # had a maximum OUR capacity of 300 mmol/l/h with the fermentor equipped with pure oxygen enrichment in the aeration system. Case #2 had a much lower OUR capacity (75 mmol/l/h) for using air as the sole oxygen source. Both fed-batch cases used the same medium and glucose feed, and reached similar cell density in the simulation. However, the actual dissolved oxygen (DO) level in case #2 was lower than the set point (25% air saturation) during h due to the lack of pure oxygen enrichment in aeration. The model predicted significantly decreased EPA production (titer, rate, and yield) in case #2 due to the loss of DO control and was confirmed in other previous fedbatch experiments (data not shown). Usually, to avoid the loss of DO control when the fermentor has a lower oxygen transfer capacity, a lower cell density level is suggested for the fermentation by charging less growth nutrients in the initial medium. Eight cases of single-stage continuous fermentation were simulated to compare with the standard fed-batch fermentation (Table II, Entry ) for EPA production. The operating conditions listed in Table II, Entry 3 0 cover the reasonable ranges of operation in a potential single-stage continuous fermentation process for EPA production, which include glucose concentrations in feed between 450 and 700 g/l, C/N ratios in feed between 24 and 68 g/g, and dilution rates between and h. It is noted in Xie et al.: Two-Stage Continuous Fermentation for Omega-3 5 Biotechnology and Bioengineering

6 Table II. Model simulation results of omega-3 EPA production by the metabolically engineered Y. lipolytica Z609/Z7334 under fed-batch, single-stage continuous, and two-stage continuous fermentation conditions. Fermentation vessel- Fermentation vessel-2 Entry no. Fermentation operation mode Effective ferment time (EFT) a Glucose feed Conc. C/N Dilution rate Glucose feed Residual glucose OUR max Conc. C/N Dilution rate h g/l g/g h g/l mm/h g/l g/g h g/l mm/h Overall EPA production Residual glucose OUR max / 2 Titer Rate b Yield c % biomass Unit/L/h % Std. Fed-batch # N/A d N/A N/A N/A N/A N/A N/A N/A Fed-batch # N/A N/A N/A N/A N/A N/A N/A N/A Single-stage N/A N/A N/A N/A N/A N/A contn. # 4 Single-stage N/A N/A N/A N/A N/A N/A contn. #2 5 Single-stage N/A N/A N/A N/A N/A N/A contn. #3 6 Single-stage N/A N/A N/A N/A N/A N/A contn. #4 7 Single-stage N/A N/A N/A N/A N/A N/A contn. #5 8 Single-stage N/A N/A N/A N/A N/A N/A contn. #6 9 Single-stage N/A N/A N/A N/A N/A N/A contn. #7 0 Single-stage N/A N/A N/A N/A N/A N/A contn. #8 Two-stage N/A / contn. # 2 Two-stage N/A / contn. #2 3 Two-stage N/A / contn. #3 4 Two-stage N/A / contn. #4 5 Two-stage N/A / contn. #5 6 Two-stage N/A / contn. #6 7 Two-stage N/A / contn. #7 8 Two-stage contn. # N/A / a For the fed-batch fermentation, the effective fermentation time (EFT) ¼ elapsed fermentation time; for the single-stage continuous fermentation, EFT ¼ residence time ¼ /F ¼ /D; For the two-stage continuous fermentation, EFT ¼ the overall residence time ¼ ( þ 2)/F2 ¼ ( þ /2)/D2. b Rate refers to volumetric productivity. c Only the relative yield is given in this paper by using the standard fed-batch fermentation (Entry no. ) as the control. d N/A stands for data not applicable. Table II that the residual glucose concentrations in the continuous fermentor are controlled within 5 g/l to avoid both glucose limitation and too much excess of glucose (waste of carbon in downstream process). Therefore, dilution rate, glucose concentration and C/N ratio in feed cannot be randomly picked and have to be carefully designed for the continuous fermentation by using the model simulation. For example, if a higher glucose concentration is used in feed, then a lower C/N ratio (or more nitrogen source) in feed should be considered to grow more cells to consume the fed glucose so that the residual glucose concentrations in fermentor are still within 5 g/l, or a lower dilution rate should be used so that cells in the fermentor have longer residence time to consume the fed glucose to 5 g/l residual concentrations. All eight cases of the single-stage continuous fermentation improved the EPA productivity (or referred to rate in the paper) by 0 200%, which is consistent with most of the previously reported continuous fermentation results in the literature. However, much less improvement in EPA productivity were predicted by the model for those single-stage continuous processes with longer residence times (Table II, Entry 3 5) to achieve comparable EPA titer as seen 6 Biotechnology and Bioengineering, ol. 9999, No. xxx, 206

7 in fed-batch fermentation (22.2% in biomass). Also, a significantly higher C/N ratio in the feed is suggested to achieve a higher EPA titer in biomass. Though the use of more concentrated glucose feed requires longer residence time in the fermentor to consume glucose to low concentrations ( 5 g/l), it still improved the overall EPA productivity (Entry 3 5). Including more nitrogen source in glucose feed (lower C/N ratio) stimulates more cell growth in the fermentor, so then the fermentation requires less residence time to consume the glucose. However, less residence time also leads to lower EPA titers even though the overall EPA productivities are greatly improved (see Entry 3, 6, and 8 or Entry 5, 7, and 9 in Table II). A more than twofold higher EPA productivity can be achieved if the continuous fermentor is operated at the highest dilution rate of h (Entry 0), but this also leads to the lowest EPA titer (0.6% in biomass). While the benefit of improved EPA productivities was predicted by model for all cases of the single-stage continuous fermentation, it was also found that EPA conversion yield decreased by 42 47% in all cases as compared to the standard fed-batch fermentation. Therefore, the model suggests the gain of EPA productivity in single-stage continuous fermentation is at the cost of significant losses in EPA yield and/or EPA titer in biomass, which makes the single-stage continuous operation not an attractive option for EPA production. To overcome the reduction of EPA titer and yield in the singlestage continuous fermentation, a two-stage continuous fermentation strategy was proposed (Fig. ). The feed for the Stage- vessel contains glucose and some other growth nutrients (e.g., nitrogen source) since the main focus in Stage is biomass formation. The feed for the Stage 2 vessel contains only glucose in order to provide nitrogen limiting conditions necessary for lipid and EPA production. Eight cases of the two-stage continuous process were simulated and compared in Table II (Entry 8). Our goal is to find the conditions that achieve significantly improved EPA productivities without sacrificing titers and yields. All eight cases were simulated under various dilution rates and vessel capacities for each stage to achieve the same 22.2% EPA titer in biomass, as compared to the standard fed-batch fermentation at EFT ¼ 20 h (Table II, Entry ). Based on the simulation results, the glucose concentration in Stage feed has a minor effect on the overall EPA productivity. Increasing the glucose concentration from 500 to 600 g/l improves the overall EPA productivity by about 7% (Table II, Entry 8). Also, the ratio of the two fermentor vessels working volumes ( / 2 ) has very little impact on the EPA rate. For example, when / 2 decreases from / (Table II, Entry 2) to /4 (Table II, Entry 7 8), the model predicts no significant difference in EPA productivities, but the requirements for Stage OUR capacity increases from to mmol/l/h. It seems the vessel for the cases of lower / 2 requires more nitrogen source (lower C/ N ratio) in its feed to grow more cells to consume the fed glucose (Table II, Entry 7 8), which leads to a higher specific growth rate, a higher dilution rate, and a higher OUR requirement. However, the maximum OUR requirement for vessel (Table II, Entry 8) is still no higher than that for the standard fed-batch fermentation (275 mmol/l/h, Entry in Table II). In addition, the OUR requirements (95 07 mmol/l/h) for the production fermentor, vessel 2, is predicted to be much lower than that for the fed-batch fermentation. The most interesting aspect of the two-stage continuous fermentation system is the overall conversion yield of EPA, which is about 95 98% of what is achieved in standard fedbatch fermentation. This in contrast to what is reported in the literature regarding productivity gains in continuous systems at the expense of reduction in continuous fermentation product titers/concentrations and/or conversion yields (Ethier et al., 20; Mutschlechner et al., 2000). Model simulation predicts that the two-stage continuous fermentation is able to improve EPA productivity by about 70% with very similar EPA titers in biomass and EPA yields as in the fed-batch process. Since the cell densities in all the two-stage continuous cases are 30 40% higher than in the standard fed-batch fermentation (data not shown in the table), the absolute EPA concentrations (biomass concentrations EPA titer in biomass) in the broth are also improved by 30 40%. Experimental Study of Two-Stage Continuous Fermentation The two-stage continuous fermentation case #6 (Entry 6 in Table II), which has a low / 2 ratio for potential high-rate EPA production at large scale, was chosen for a further experimental validation. Table III shows more detailed steady-state operating conditions and results from the computer simulation. The working volume was controlled at.5 L for vessel and 4.0 L for vessel 2, respectively, by fixing the bottom end of each withdrawing tube end at the calibrated position. The average residence time was about 45 h for vessel ( /F )and65hfor vessel 2 ( 2 /F 2 ), which led to an overall residence time of about 90 h (( 2 þ )/F 2 ). In addition, vessel 2 was inoculated and started first as a fed-batch fermentation, followed by vessel at Table III. Computer simulation data of a representative case of EPA production by the two-stage continuous fermentation of Y. lipolytica Z7334 (Entry 6, Table II), which was further used for an experimental validation. Model simulation Conditions and results Stage Stage 2 Overall Operating conditions essel working volume, - L N/A Glucose feed rate, F S - ml/h N/A Glucose concentration in feed, S FS - g/l N/A Nitrogen concentration in feed, N FS - g/l N/A Base (KOH, 560 g/l) feed rate, F B - ml/h N/A Dilution rate, D -h Residence time - h Predicted steady-state results Cell density, X - Unit/L Residual glucose concentration, S - g/l Residual nitrogen concentration, N - g/l Lipid content in biomass, L/X - % biomass EPA content in lipid, E/L - % lipid EPA titer in biomass, E/X - % biomass EPA productivity (rate) - Unit/L/h EPA conversion yield - % Std. fed-batch Xie et al.: Two-Stage Continuous Fermentation for Omega-3 7 Biotechnology and Bioengineering

8 t ¼ 72 h, also in fed-batch mode. The continuous operation started at t ¼ 20 h by continuously pumping broth from vessels to 2 and from vessel 2 to a harvesting container at the predetermined rates. At t ¼ 95 h, the glucose feed for vessel was also changed to the same as vessel 2, that is, 700 g/l glucose with no nitrogen. This was to consume all the residual nitrogen in vessel and accumulate higher EPA titer before the whole fermentation was shut down at t ¼ 008 h. The experimental results are shown in Figure 3. Due to the coexistence of two cascading fermentation vessels, each operated at different residence times, and as such the time needed to reach a steady state was different for each vessel. During the first 20 h, EPA titer in biomass, EPA conversion yield, and byproduct yield for vessel 2 were very similar to what we observed in the previous fed-batch experiments since it was operated in fedbatch mode (see both Figs. 2 and 5). When vessel was inoculated at t ¼ 72 h and then the two-stage continuous operation started at t ¼ 20 h, the cell densities in both stages increased quickly Figure 3. Experimental results of EPA (C20:5) production by two-stage continuous fermentation of Y. lipolytica Z7334 with process conditions predetermined by computer simulation (see Entry 6 in Tables II and III). The symbols stand for real experimental data and dash lines represent the steady-state values predicted by the model. 8 Biotechnology and Bioengineering, ol. 9999, No. xxx, 206

9 (Fig. 3A). The Stage cell density became stabilized at around 30 Unit/L and EPA titer became stabilized too at 6% biomass after t ¼ 300 h, which suggests it took about four tank turnover in vessel ( 45 h) to reach a steady state. essel 2 had a longer residence time (65 h), and both cell density and EPA titer were stabilized at 270 Unit/L and 23% biomass respectively after t ¼ 600 h. A much longer period (more than sevenfold of vessel 2 s residence time or fivefold of the overall residence time [90 h]) was needed to reach a steady state in vessel 2. Since vessel 2 was continuously fed with cells from vessel, a much longer stabilizing time was needed for vessel 2 because it had to wait until vessel reached a steady state before its own fermentation state became stabilized. Another possible reason causing the longer stabilizing time needed for vessel 2 was the 5 0% fluctuations in the liquid levels in both vessels caused by agitation and some unstable foaming conditions during the experiment, and this also led to significant oscillations in the residual glucose concentrations (Fig. 3B). The second interesting aspect of the two-stage continuous fermentation system is the strain stability for EPA production during the long process. The Y. lipolytica Z7334 had grown for about 60 generations during the elapsed,008 h of the experiment. No significant losses in EPA titer and yield were seen in the two-stage continuous experiment while the overall productivity was improved by 80% (Fig. 3 and Table I). This suggests that the strain suffered no loss in genetic stability, which is a common concern for most continuous fermentation processes employing metabolically engineered strains (Shuler and Kargi, 2002). The cell densities, EPA titers, EPA rates, EPA yields, and lipid titers after 600 h in both stages were similar to or even higher than the corresponding model predictions. Compared with the standard fed-batch run, the Table I. Comparison between experimental results of the standard fed-batch fermentation and two-stage continuous fermentation of Y. lipolytica Z7334. Fermentation parameters Fed-batch fermentation Two-stage continuous fermentation (average) Percentage of increase (%) EPA titer in biomass (% biomass) EPA rate a (Unit/L/h) EPA yield b (%std. fed batch) EPA concentration (Unit/L) Biomass concentration (Unit/L) Lipid titer in biomass (%biomass) EPA in lipid (%lipid) Effective fermentation time c (h) Maximum OUR requirement (mmol/ L/h) (vessel ), 00 (vessel 2) N/A a All rates refer to accumulated volumetric productivities. b Yield refers to the relative yield as compared to the standard fed-batch fermentation. c For the two-stage continuous fermentation, the effective fermentation time (EFT) ¼ the overall residence time ¼ ( þ 2)/F2 ¼ ( þ /2)/D2. two-stage continuous fermentation improved the final EPA productivity by 80% with very similar EPA titer in biomass and EPA conversion yield (Table I). The EPA titer in this paper refers to the percentage of EPA weight in whole biomass (E/(E þ L E þ X f ) 00%); it is a specific titer. The EPA concentration (EPA titer in biomass biomass concentration) was actually improved by 40% in the two-stage continuous fermentation since the biomass concentration was 40% higher (Table I). This is one of the very few continuous fermentation examples in the published literature that successfully demonstrated significant improvements in both volumetric productivity and concentration of the product with the conversion yield wellmaintained. The consistently improving EPA production during the long period of continuous fermentation was also accompanied by the continuous evolution of the Y. lipolytica Z7334 strain. From the data it can be seen that the strain produced more and more lipid in biomass (Fig. 3C) and less and less byproduct (Fig. 3G) over the time. The lipid titer in biomass increased by about 20% while the total byproduct decreased by 20% in both stages during the entire fermentation period. The overall EPA conversion yield was also kept in a continuous increasing trend (Fig. 3H). This interesting evolution also led to underestimations of experimental data by model predictions, especially in the late stage of the continuous fermentation (Fig. 3C, E H). It seemed the strain evolved to reduce byproduct (mainly organic acids) formation during the continuous fermentation to alleviate any possible inhibition in cell growth. The reduction in byproduct formation helped to introduce more carbon flow to the lipid synthesis pathway and further led to improvement in EPA production (Fig. 3F). As a result, OUR (oxygen uptake rate) increased gradually over time (Fig. 3E). This increase may be the result of increased lipid and EPA production, which have highly oxygen-dependent desaturation steps in the biochemical synthesis pathway (Meesapyodsuk and Qiu, 202). Also, a higher CO 2 evolution rate (CER) was observed as the OUR increased, which gave a consistent respiratory quotient (RQ ¼ CER/OUR) close to during the entire continuous fermentation process. The strain evolution for more product accumulation and less byproduct formation was an unanticipated feature that we discovered during the omega-3 production by the two-stage continuous fermentation. The glucose feed for vessel was changed to a glucose-only feed solution after t ¼ 95 h. This was done to mimic a real production situation since the broth from both vessels and 2 should be harvested with high EPA titers in biomass. During 95,008 h, lipid and EPA titers in vessel was further significantly increased due to the depletion of the residual nitrogen (Fig. 3C and D). This helped guarantee the high EPA titers in the final harvested biomass (>23%, w/w) from both vessels and 2. Isolation of High-Lipid Producers From the Continuous Experiment The discovery of strain evolution for more lipid production and less byproduct formation during the two-stage continuous fermentation experiment provided an opportunity of isolating high-lipid producing strains from the samples near the end of the run. Fresh broth samples from vessel 2 of the continuous experiment at Xie et al.: Two-Stage Continuous Fermentation for Omega-3 9 Biotechnology and Bioengineering

10 t ¼ 95 h were streaked out on Y. lipolytica minimal plates to obtain single colonies. Dozens of single colonies were selected based on difference in growth phenotypes and then tested for the ability of improved lipid and EPA production in shake- flask experiments. To identify the top lipid producers, the original Y. lipolytica Z7334 cells from the stock vial and the evolved Z7334 cell mixture from the sample at t ¼ 95 h were used as the two controls for the comparison with the isolated evolved Y. lipolytica Z7334 variants. The results are shown in Figure 4. Under the shake-flask fermentation conditions, the original Y. lipolytica Z7334 cells produced 62% lipid in biomass and 45% EPA in lipid (28% EPA in biomass), but the cells from 95 h of the twostage continuous experiment (mixture of all evolved cells) produced 70% lipid in biomass and 39% EPA in lipid (28% EPA in biomass). This further confirmed the observation during the twostage continuous fermentation experiment that the Y. lipolytica Z7334 cells evolved to produce more lipid and similar EPA content in biomass. Further examination of the screening results from a pool of resolved single-colony Y. lipolytica Z7334 isolates show their properties are quite different from each other. Some isolates produced more than 70% lipid in biomass and nearly 40% EPA in lipid (e.g., isolate # and #5 in Fig. 4), and their overall EPA titers (%biomass) were even higher than that of the original Y. lipolytica Z7334 strain. However, so far the Y. lipolytica strain s evolution for more lipid and less byproduct was still based on observations in phenotype changes. More studies will be needed in future to figure out whether there were true genotype changes during the long period of continuous fermentation. Liu et al. (205a) found that the lipid content of Y. lipolytica strains was able to be significantly improved by an evolutionary metabolic engineering approach. Though rational design of lipid synthesis pathway and the followed metabolic engineering were still the most commonly used approaches to build high lipid-producing strains (Blazeck et al., 204; Liu et al., 205b), the evolution approaches either by continuous fermentation experiments as shown in this study or by evolutionary metabolic engineering approach as demonstrated by Liu et al. (205a) provide additional opportunities to find hypeproducers of lipid or lipid-derived products in future. Figure 4. Comparison of lipid titer (%biomass) and EPA content in lipid (%lipid) between the flask fermentation of original Y. lipolytica Z7334, the mixture of the evolved Z7334, and single isolates of the evolved Z7334. The evolved Y. lipolytica Z7334 cells were obtained from the 95 h sample of the two-stage continuous fermentation experiment. Advantages of the Two-Stage Continuous Fermentation Like most other fermentation programs, the research on omega-3 fermentation was primarily focused on batch and fed-batch processes, especially for the purposes of strain screening and optimization of basic fermentation conditions (Xie et al., 205). However, current batch or fed-batch fermentation technologies are still based on processes developed more than 70 years ago. As significant progress has been achieved today in microbiology, metabolic engineering, protein engineering, and advanced genetic tools, the fermentation technologies from the middle of the last century are far behind of the fast pace of advances in modern biotechnology. Urgent needs from the current fermentation industry include novel bioreactors with high capacities and low cost, fully automatic bioprocess monitoring and control system, and advanced bioprocess integration for high productivities. One big challenge in batch/fed- batch fermentation processes is the low productivity, which makes it difficult to compete with most chemical or petrochemical processes, which are typically operated continuously for higher productivities. In this paper, we compared omega-3 production under fedbatch, single-stage continuous, and two-stage continuous fermentation conditions, first by model analysis, and then by experimental validation. The model simulation results showed that the two-stage continuous fermentation process in general gives significant higher EPA titer and conversion yield than the single-stage continuous process though both of them improve EPA productivity over the standard fed-batch process. The advantages of the two-stage continuous fermentation were predicted by model and then validated by a 2 L (Stage ) and 5 L (Stage 2) continuous experiment. There were four major reasons accounting for the success of the two-stage continuous fermentation for EPA production by engineered Y. lipolytica:. Significantly less effective fermentation time (residence time) was required to achieve the same EPA titer due to the savings of both turnaround time and initial growth time by continuous operation, which helped improve EPA rate (productivity). 2. Two separate fermentation vessels were used for independent and optimal control of cell growth and EPA production, which helped the fermentation to achieve higher EPA titers in biomass and an overall EPA conversion yield from glucose. 3. Significantly higher cell density was achieved due to an overall less dilution effect in the continuous process, which further helped the fermentation improve EPA productivity since the EPA titer in biomass was maintained the same (Table I). The majority of water in a continuous fermentation came from the glucose feed solution while the water in fed-batch fermentation was from both initial medium and glucose feed. 4. The Y. lipolytica Z7334 strain retained its high lipid and EPA production (Figs. 3 and 5) after a long period of continuous operation due to its inherent high degree of genetic stability (Xue et al., 203). Though the success of the two-stage continuous fermentation experiment was expected based on the model simulation results, the evolution of the Y. lipolytica strain to a higher productivity variant 0 Biotechnology and Bioengineering, ol. 9999, No. xxx, 206