Application of Stable Isotope Tracing to Elucidate Metabolic Dynamics During Yarrowia lipolytica α-Ionone Fermentation.

Targeted metabolite analysis in combination with 13C-tracing is a convenient strategy to determine pathway activity in biological systems; however, metabolite analysis is limited by challenges in separating and detecting pathway intermediates with current chromatographic methods. Here, a hydrophilic interaction chromatography tandem mass spectrometry approach was developed for improved metabolite separation, isotopologue analysis, and quantification. The physiological responses of a Yarrowia lipolytica strain engineered to produce ∼400 mg/L α-ionone and temporal changes in metabolism were quantified (e.g., mevalonate secretion, then uptake) indicating bottleneck shifts in the engineered pathway over the course of fermentation. Dynamic labeling results indicated limited tricarboxylic acid cycle label incorporation and, combined with a measurable ATP shortage during the high ionone production phase, suggested that electron transport and oxidative phosphorylation may limit energy supply and strain performance. The results provide insights into terpenoid pathway metabolic dynamics of non-model yeasts and offer guidelines for sensor development and modular engineering. Published by Elsevier Inc.

Introduction

Liquid chromatography-tandem mass spectrometry (LC-MS/MS)-based metabolite quantification is increasingly useful to guide metabolic engineering, and when paired with 13C-isotopic tracing methods it can be used to track fluxes through metabolism and carbon movement (Jang et al., 2018, Abernathy et al., 2019, Allen, 2016). The molecular measurements give insight into the status of engineered cellular systems, facilitating further strain improvement and greatly reducing the time spent developing industrial production strains (Campbell et al., 2017). Metabolic measurements from bioreactors at production scale can identify nonuniformities such as imperfect mixing, fluctuations in pH, and temperature and substrate gradients, phenomena that can lead to suboptimal strain performance and enhanced rates of genetic mutation (Käß et al., 2014, Elena and de Visser, 2003). Thus, detection of metabolic changes that are a result of large-scale cultivations (such as alterations in metabolite bottlenecks or low oxidative phosphorylation efficiency) enables more robust industrial strain designs or fermentation optimizations (Wehrs et al., 2019). However, production-scale reactors are not readily available and are cost-prohibitive. Therefore, scaled-down bench-top fermenters provide a compromise to rigorously assess initial strain performance.

Analytically, MS-based 13C-isotopic tracing methods and, more generally, metabolite profiling analyses cannot accurately identify and quantify all metabolites within a single run due to challenges in separation and detection. Best practices use gas chromatography (GC) or LC in tandem with MS to separate and quantify a subset of metabolites including isotopologues. GC is useful for metabolite quantification but requires a derivatization step (via silylation or alkylation) to volatilize polar sugars, amino acids, and organic acids and results in reduced sample stability (Halket et al., 2005, Allen and Ratcliffe, 2009). LC does not require chemical modification of analytes, but the presence of sample matrix and solvent effects can suppress ion signals. In some instances, the signals can be enhanced with the addition of an ion pairing (IP) reagent; however, the IP reagent is present at millimolar quantities in the buffer and contaminates instrumentation, requiring considerable effort in post-operation cleanup. Hydrophilic interaction chromatography (HILIC) columns and methods separate amino acids (Kambhampati et al., 2019), sugars (Xia et al., 2019), and central carbon intermediates (Rodriguez et al., 2016, Teleki et al., 2015) and are now used for targeted metabolite quantification in conjunction with dynamic 13C-tracing to examine Escherichia coli and cyanobacteria metabolic dynamics (Li et al., 2017, Gao et al., 2016b). In the current study, a HILIC method using the modifier medronic acid was developed and employed to retain and separate isobaric metabolites within a single 20-min run. Limits of detection were determined for 15 amino acids and 42 metabolite intermediates of the central carbon metabolism and mevalonate (MVA) pathway. The developed HILIC method and MS/MS were used to analyze an oleaginous yeast Yarrowia lipolytica strain engineered to produce α-ionone.

Ionones are a subclass of terpenoid aroma compounds in plants that are valuable to the fragrance and flavor industries. Typically, plants only produce low amounts of aroma compounds. These low natural levels combined with increasing consumer demand for all-natural products has strained the current ionone supply chain and provided an opportunity for the development of new production methodologies. Recent efforts have focused on producing ionones via biotechnological processes (Jiang et al., 2019, Zhang et al., 2018, Chen et al., 2019), including a study reporting titers of approximately 400 mg/L of α-ionone from E. coli in 10,000-L fermenters (Lukin et al., 2019). However, organisms that are Generally Recognized As Safe (GRAS) are desirable. Y. lipolytica is an oleaginous yeast that is capable of accumulating large amounts of lipids and fatty acids (>70% dry cell weight (DCW), after genetic engineering, Friedlander et al., 2016, Xu et al., 2016). The large flux through acetyl-CoA and the ability to consume a wide range of substrates suggest that Y. lipolytica may be well suited for terpene production. Previously, a strain was engineered for the production of β-ionone (Czajka et al., 2018). Here, a Y. lipolytica strain engineered to produce α-ionone is described in high-cell-density fermentations. Using HILIC-MS/MS, temporal shifts in metabolism, which resulted in MVA excretion during the growth phase followed by uptake during the ionone production phase, were captured over the course of fermentation. Labeling measurements from pulse-trace experiments indicated that tricarboxylic acid (TCA) cycle activity was limited under high-cell-density fermentations, re-capitulating a trend previously described in E. coli (Chen et al., 2011). The findings here have broad implications for the metabolic engineering community, as the MVA pathway is the source of a number of valuable natural compounds.

Results

Metabolite Detection via HILIC Method

A comprehensive HILIC method was developed to detect and achieve efficient separation of sugar phosphates, sugars, and organic acids in central carbon metabolism; intermediates in the MVA pathway; and amino acids within a single, 20-min run. The aqueous running buffer contained 10 mM ammonium acetate (pH 9.0) in water (buffer A), and the organic buffer contained 10 mM ammonium acetate (pH 9.0) in acetonitrile:water (9:1) (buffer B); 5 μM medronic acid (CH6O6P2) was added to both buffers. The binary gradient started at 95% B and was gradually reduced over a 20-min run time that included 6 min of equilibration back to the starting conditions (described in the Methods). Limits of detection of 17 sugar phosphates, 7 sugars, 8 organic acids, 7 MVA pathway intermediates, 15 amino acids, and the energy molecules (adenosine triphosphate [ATP], adenosine diphosphate [ADP], and adenosine monophosphate [AMP]) were quantified. The chromatographic separation for a select set of metabolites relevant to this study is presented in Figure 1. A complete list with retention times, limits of detection, and MS parameters is provided in Table S1.

Figure 1

HILIC Metabolite Peaks

(A–C) (A) Isobaric sugars and sugar phosphates, (B) central carbon metabolite intermediates, and (C) MVA pathway intermediates. *FPP and GGPP eluted at approximately the same time.

One challenge with the detection and quantification of central carbon intermediates is that a number of important metabolites are isomers. Triose phosphates (glyceraldehyde-3-phosphate and dihydroxyacetone phosphate), pentose phosphates (ribose-5-phosphate and ribulose-5-phosphate), hexose phosphates (glucose-1-phosphate, fructose-6-phosphate, and glucose-6-phosphate), and some sugars/alcohols (glucose and fructose, sucrose, maltose, galactinol, etc.) are not distinguishable from each other by MS because of identical chemical formulas. Chromatographic separation of these isobaric compounds was achieved using HILIC and alkaline pH conditions (pH 9.0) (Figure 1A), with the exception of the pentose phosphates (quantified as pentose-5-phosphates). Medronic acid and a slow initial solvent gradient (95%–70% buffer B over 8 min) played a crucial role in the sensitive detection and resolution of sugar phosphates and MVA pathway intermediates (Figures 1B and 1C). The optimal peak shapes resulted in lower limits of detection (<5 pmol) sufficient for quantification. The HILIC method was then verified with a yeast production system, Y. lipolytica, engineered to produce α-ionone. As the yeast was cultivated in rich medium (YPD) that contains peptone, the amino acids were not monitored. 13C-isotopic tracing methods were employed with the resulting isotope measurements collected via a scheduled multiple reaction monitoring (MRM) method.

Yarrowia lipolytica Engineering for α-Ionone Production

Arch Innotek's proprietary Y. lipolytica-producing α-ionone strain was used throughout the study. The strain was established based on a β-ionone-producing Y. lipolytica strain previously described (Czajka et al., 2018). The α- and β-ionone production pathways share 12 common biosynthetic steps through the MVA pathway (Zhang et al., 2018). Three molecules of acetyl-CoA are condensed to form 3-hydroxy-3-methylglutaryl-CoA, which is then converted to the five-carbon terpene building blocks isopentyl-pyrophosphate (IPP) and dimethylallyl pyrophosphate (DMAPP). IPP and DMAPP are sequentially condensed to form the C40 compound, lycopene, the last shared intermediate between the ionone isomer pathways. For α-ionone, the ends of the same lycopene molecule are chemically linked to form δ- (one ring) or ε- (two rings) carotene. The introduction of lycopene cyclase from Lactuca sativa (LsLCYe) and the carotene cleavage dioxygenase from Osmanthus fragrans (OfCCD1) (Wang, 2019) led to the detectable production of α-ionone. When the two genes were added to a previously constructed and described strain engineered to overexpress the MVA pathway (Table S2; Czajka et al., 2018), a titer of 43 ± 3 mg/L of α-ionone was achieved in shake-flask growths (Figures S1 and S2). Strains used in this study and their genetic information are presented in Table S2.

Bench-Top Fermentation Achieved ∼400 mg/L of α-Ionone

A bench-top bioreactor (Sartorius BIOSTAT A, 5 L) was operated in fed-batch mode with a 10% (v/v) dodecane layer to extract, retain, and quantify ionone titers. A representative fermentation curve is shown in Figure 2A. The maximal biomass, measured by OD600, was achieved within 75 h. The dissolved oxygen saturation percentage dropped to ∼2%–5% during the early growth phase and maintained this low level until the late production phase (∼150 h). The residual glucose concentration is shown in Figure S3. α-Ionone was produced throughout the entire fermentation period; however, there was an increase in the average production rate from 0.9 ± 0.5 mg/L/h to 3.7 ± 0.4 mg/L/h (Figure 2B) at 104 ± 10 h after inoculation. 408 ± 50 mg/L of α-ionone was obtained at the end of the fermentation, among the highest reported titers to date (Figure 2C). The ratio of α- to β-ionone was 90% ± 2%, with 80% ± 5% of the quantified ionone in the dodecane overlay (i.e., ∼20% remained intracellular). Previously, it was shown that the ionone capture efficiency of a dodecane overlay is <100% (∼76%), thus the reported α-ionone production estimate is conservative (Czajka et al., 2018). Based on the strain fermentation characteristics, metabolic changes due to pathway engineering and product synthesis were further investigated.

Figure 2

Ionone Production Parameters in 5-L Bioreactor

(A–C) (A) Representative OD600 and α-ionone production over time, (B) α-ionone production rate before and after 104 h, and (C) final titers of α- and β-ionones at the end of 5-L bioreactor cultivation. Data are represented as mean ± SD from biological triplicates.

Metabolite Comparisons and Isotopic Pulse-Chase Indicated Minimal Changes to the Flux Profile

Engineering the overexpression of multi-step enzymatic pathways can lead to metabolic burdens and unforeseen metabolic changes. To determine the cellular response to the engineered pathway, intracellular metabolite concentrations (pool sizes) were measured in exponentially growing cells (in shake-flasks) via targeted metabolite analysis. A comparison with the wild-type (WT) indicated that the engineered strain had reduced TCA cycle pool sizes (Figures 3A and 3B). Previous studies have also reported reduced levels of citrate (CIT) and other TCA intermediates during shake-flask cultivations of strains producing carotenoids (Zhao et al., 2017, Nagai et al., 2018), as a consequence of re-routing of TCA cycle carbon to product synthesis (through the ATP-citrate lyase reaction or acetyl-CoA competition). To minimize the competition for TCA intermediates, glutamate, acetate, and ethanol were supplemented to shake-flask cultures. Interestingly, only ethanol resulted in an increase of α-ionone titer (Figure S2), consistent with previous results for a beta carotene-producing strain (Czajka et al., 2018). Isotopic tracing was then used to compare the pathway activity of the WT and engineered strains during fermentation processes.

Figure 3

WT and Engineered Strain Pool Size Comparisons and Pulse-Trace Results

(A) Intracellular metabolite pool size ratios of the upper glycolysis pathway metabolites of the engineered strain compared with the WT.

(B) Pool size ratios of the TCA cycle and MVA pathway metabolites. The dotted line shows a ratio of 1 and represents the metabolites that have equal pool sizes in each strain.

(C) Pulse-trace labeling incorporation. 13C content of central carbon metabolites after 20 min of metabolite labeling using a U-13C6 glucose tracer for wild-type (WT) and engineered (Eng) strains. Labeling percentage is normalized to the average G6P labeling.

(D) MVA labeling.

See also Figures S5 and S6. (A and B) Data are represented as mean ± SD from biological triplicates; (C and D) data are represented as mean ± SD from biological duplicates.

Cells from fed-batch bioreactors were collected over the course of fermentation and immediately pulsed into a medium containing fully labeled (U-13C6) glucose in shake-flasks. After a specified time in the labeled medium, cells were mixed with a cold ethanol:sodium chloride solution (final solution temperature ∼ −20°C) to rapidly quench cell metabolism (Figure S4). The dynamic labeling results revealed four main observations. First, the central carbon-labeling profiles were reproducible across bioreactor runs (Figures 3 and S5). Second, the incorporation of 13C into TCA cycle metabolites (CIT, α-ketoglutarate [AKG], succinate [SUC], malate [MAL]) in both WT and engineered strains was much lower relative to the sugar phosphates in both growth and production phases (Figures 3, S5, and S6). Third, the MVA labeling enrichment was higher in cells from the growth phase than in those from the production phase (Figure S5). Fourth, MVA was labeled to lower levels in the engineered strain relative to the WT (Figure 3). Extending the labeling period from 20 min to 2 h did not result in significant MVA labeling (Figure S5). As the engineered strain had an overexpressed MVA pathway, MVA labeling enrichment was expected to be higher than in the WT, yet the opposite was found to be the case. Therefore, additional metabolite quantification was performed.

Extracellular MVA Quantification Indicates MVA Uptake from the Media and a Shifting Metabolic Bottleneck

Metabolite concentrations in the fermentation medium of the engineered strain were quantified to gain insights into the low MVA 13C incorporation and to understand the metabolic shifts over the course of the fermentation. A significant amount of MVA secretion was identified, with concentrations peaking at 405 ± 33 mg/L (Figure 4). In a separate test, MVA was also secreted during fermentation without a dodecane overlay and reached ∼270 mg/L, indicating that the secretion is inherent to the biological system and not an artifact of the dodecane overlay. The MVA concentration decreased following the peak, with a corresponding increase in the production of ionone. A similar dynamic was observed for CIT, MAL, SUC, fumarate (FUM), and pyruvate (PYR), which peaked at a total concentration of 4.2 ± 0.9 g/L (Figure 4B). The extracellular metabolite dynamics suggested that secreted MVA, which was unlabeled, was taken up by the yeast cells and used to produce ionone, consistent with the low level of MVA 13C incorporation observed in the engineered strain (Figures 3 and S5). Organic acids have been previously reported to be consumed by Y. lipolytica (Papanikolaou et al., 2002, Sabra et al., 2017), whereas MVA consumption has been widely reported in E. coli (Yoon et al., 2006, Wang et al., 2011), and to a lesser extent in some yeast (Rhodotorula glutinis, Ikeura et al., 1988; Phaffia rhodozyma, Calo et al., 1995), but not others (S. cerevisiae, Rodriguez et al., 2016). To test whether the engineered Y. lipolytica strain was capable of MVA uptake for production of ionone, cultures were grown in shake-flasks and supplemented with 2 g/L MVA. The resulting ionone production, measured via GC-MS, was increased by 35% (p < 0.05, Figure S2). The results indicate that MVA can serve as a valuable carbon source for enhanced production of ionone.

Figure 4

Representative Extracellular Metabolite Dynamics Over the Course of Fermentation

(A and B) (A) MVA and ionone production and (B) organic acids. Error bars represent standard deviation of duplicates. Data are represented as mean ± SD from technical duplicates.

Pool Size Quantification Indicated Limited TCA Cycle Efficiency and Energy Limitation

Intracellular pool sizes were measured in the engineered strain during the fermentations. Intracellular MVA concentrations decreased over the course of the fermentation (Figure 5). Of the downstream MVA pathway intermediates, IPP had the highest concentration, geranyl pyrophosphate (GPP) and farnesyl pyrophosphate (FPP) were present in low abundance and geranyl-geranyl pyrophosphate (GGPP) was not present at detectable levels. Of the quantifiable metabolites, MAL, FUM, and AKG were the most abundant (Figure 5).

Figure 5

Representative Intracellular Metabolite Concentrations during the 5-L Bioreactor Fermentation

(A–D) (A) Glycolysis and pyrophosphate pathway metabolites, (B) TCA cycle metabolites, (C) MVA pathway metabolites, and (D) MVA concentration and CIT pools reported as relative intensities *nd, not detected. Colors correspond to sampled time points with 50 h (blue), 71 h (red), 98.5 h (black), and 146 h (gray). Data are represented as mean ± SD from technical duplicates.

Most metabolite concentrations declined over the fermentation, with significant decreases occurring during the late production phase (146 h). The intermediates of the TCA cycle had a 2.9-fold (MAL), 3.1-fold (FUM), 3.0-fold (SUC), and 3.5-fold (CIT) reduction between the 98.5- and 146-h samples (Figure 5). The general decrease of metabolites may be the result of reduced metabolic and TCA cycle activity when cells entered the stationary phase. As an obligate aerobe, Y. lipolytica relies on the TCA cycle and oxidative phosphorylation to supply energy required for metabolic activity. Thus, reduced TCA cycle activity and capacity for electron chain transport due to low oxygen availability would be expected to result in energy deficiencies. Indeed, when the concentrations of ATP, ADP, and AMP were measured, the calculated energy charge (defined as ([ATP] + ½[ADP])/([ATP]+[ADP]+[AMP]) was significantly lower in the production phase (0.5) relative to the growth phase (0.9) and characteristic of an energy-limited cellular state (Wu et al., 2016); however such numbers should be viewed qualitatively because energy metabolites can be difficult to quantify accurately. A similar reduction in TCA cycle metabolites was observed over time in Y. lipolytica lycopene-producing strains (Zhao et al., 2017).

Discussion

Metabolite profiling and isotopic labeling has rapidly become an important tool for analyzing biological systems; however, it is often the limiting analytical step due to challenges in separating compound classes (hydrophobic lipids, polar sugar phosphates, etc.) and detecting pathway intermediates with one short chromatographic method. Reverse-phase chromatography methods have been able to successfully separate many central carbon metabolites using an IP agent, but the extensive cleaning needed to remove the IP modifier is time-consuming. HILIC methods have recently become more widely utilized because of their ability to separate multiple types of compound classes within central carbon metabolism (Li et al., 2016, Li et al., 2017, Luo et al., 2019, Kambhampati et al., 2019). Alkaline conditions in HILIC methods provide better detection capabilities than acidic conditions (Teleki et al., 2015). A recent study demonstrated that the addition of medronic acid to alkaline conditions in HILIC significantly improved sugar phosphate peak quality (Hsiao et al., 2018). Here, a medronic acid- and alkaline-based HILIC method was developed with a high initial buffer B percent (95%) and slow gradient changes that results in adequate and robust separation of sugar phosphates, sugars, and organic acids in the central and secondary metabolism as well as amino acids. The HILIC method's ability to separate important isobars along with robust methods for quenching cellular metabolism and metabolite extraction were examined in a Y. lipolytica strain that produced high titers of α-ionone, in part due to a change in MVA use during ionone production.

Shifts in Metabolism during High-Cell-Density Fermentation Processes Can Result in Unbalanced Production Pathways

Y. lipolytica can naturally produce large quantities of lipids and therefore has the capacity to generate significant amounts of acetyl-CoA, the precursor for the MVA pathway. Thus, Y lipolytica can be metabolically engineered for terpene production (Kildegaard et al., 2017, Schwartz et al., 2017, Larroude et al., 2018). However, if the goal is to reach industrial production, the metabolic response to both engineering and high-cell-density fermentations must be considered carefully. Pathway engineering and introduction of heterologous genes can lead to metabolic reorganization in response to cellular burdens that affect strain productivity differently in small-scale (i.e., shake-flasks) growth conditions relative to high-cell-density bioreactors (Sabra et al., 2017). Here, the MVA secretion by the engineered strain during the growth phase and uptake during the production phase suggested that a major metabolic shift occurred and that the production pathway bottleneck changed over the course of fermentation. Although it is possible that the secretion was promoted by the dodecane overlay (an effective extraction solvent for terpenes), an additional fermentation without the overlay found that MVA secretion still occurred (albeit at slightly lower levels), ensuring the phenomena was inherent to the biological system. Thus, the secretion of MVA during growth suggests that a downstream enzyme step is initially rate limiting. Some studies have shown that the expression levels of the carotenoid-producing gene remain relatively constant over the course of one cultivation in natural (astaxanthin-overproducing mutant, Phaffia rhodozyma) and heterologous (E. coli) systems (Miao et al., 2011, George et al., 2018), whereas other reports have indicated that oxygen content and mannitol accumulation can affect gene expression level and carotenoid accumulation (Wu et al., 2011, Zhang et al., 2012).

In Y. lipolytica, low nitrogen levels can lead to an upregulation of the ergosterol pathway and increased ergosterol accumulation (Kerkhoven et al., 2016, Kerkhoven et al., 2017), with the transcription factor Mhy1 indicated as having a role in regulating the steroid biosynthesis pathway (Wang et al., 2018). Other transcription factors that regulate lipid production, such as Mig1 or Mga2, may also have effects on the transcription of MVA pathway genes (Wang et al., 2013, Liu et al., 2015, Morin et al., 2011). Interestingly, the translation elongation factor-1α promoter used to constitutively express the engineered genes was shown to have varying activity during cultivation with high activity under late-stage cultivation and low nitrogen conditions (Liu et al., 2019). Increased transcription of downstream MVA enzymatic steps during late stage fermentations may help alleviate potential pathway bottlenecks. In addition, several MVA pathway enzymes are known to have metabolite feedback inhibition mechanisms, including MVA kinase, which is highly inhibited by GPP and FPP (Miziorko, 2011). Thus, the rate-limiting step may be due to interactions between multi-level cellular responses under stressed fermentation conditions. Decreased levels of GPP and FPP measured over the course of the fermentation may alleviate inhibition of MVA kinase, leading to higher MVA utilization. Overall, the resulting 4-fold increase in rate during the production phase indicates a shift in metabolism to consume MVA and produce ionone more quickly than would be possible through MVA biosynthesis alone without additional engineering efforts.

Both yeast and E. coli have been reported to secrete MVA when engineered to overexpress the MVA pathway. However, whereas some intermediates are secreted and then taken up for enhanced productivity, e.g., IPP in E. coli (George et al., 2018), the uptake of MVA has not been reported to enable ionone production. Ionone production studies in E. coli have indicated more rapid ionone accumulation during the stationary phase, but as extracellular MVA was not measured, the cause of the increase was not determined (Zhang et al., 2018, Chen et al., 2019). Perhaps it is not surprising that Y. lipolytica is capable of using exogenous MVA, as its capacity to take up secreted metabolites like pyruvate (Yovkova et al., 2014) and acetate (Gao et al., 2016a) over the course of fermentation is well known, with another study describing the secretion of the precursor dammarenediol-II during the production of protopanaxadiol that has a subsequent reduction of the extracellular precursor concentration (Wu et al., 2019). Citric acid secretion and uptake after glucose exhaustion has also been observed in fermenters under varying conditions (Sabra et al., 2017). These studies indicate that the flexible metabolism of Y. lipolytica may present an advantage for high-cell-density cultivation relative to other species. Although the exact mechanism resulting in uptake remains under study, the temporal metabolic shifts of engineered strains in fermenters are intriguing and may provide guidelines for strain engineering. For example, several other oleaginous yeast and fungi have been reported to display similar lipid turnover dynamics during fermentation as Y. lipolytica, such as Cunninghamella echinulata and Cryptococcus podzolicus (Fakas et al., 2007, Qian et al., 2019) and in Saitoella coloradoensis and several Lipomyces species (Slininger et al., 2016).

Changes in TCA Cycle Metabolite Levels and Activity Further Suggest Altered Metabolism over the Course of Fed-Batch Fermentations

The decrease in TCA metabolite pool sizes and the limited label incorporation suggest that the TCA cycle has a decreased role during fermentations. It has been previously reported that the TCA cycle can be incomplete during aerobic growth in high-cell-density E. coli fermentations (Chen et al., 2011). During cultivations, dissolved oxygen becomes scarce, resulting in reduced electron transport and reduced TCA cycle metabolism. As Y. lipolytica is an obligate aerobe, the limited oxygen availability could result in decreased TCA cycle and cellular metabolic activity. Previous studies have shown that low oxygen availability reduces isocitrate, CIT, and AKG formation in Y. lipolytica (Sabra et al., 2017, Kamzolova et al., 2012, Kamzolova et al., 2013, Bellou et al., 2014). During growth on glucose, the dissolved oxygen content has an increased effect on the amount of CIT produced relative to growth on other carbon sources (Sabra et al., 2017, Magdouli et al., 2018). Flux balance analysis modeling supported the suppression of the TCA cycle under conditions that still had oxygen uptake (Figure S7), which combined with previous experimental results indicates that TCA cycle functionality is highly sensitive to oxygen levels (Sabra et al., 2017, Magdouli et al., 2018). Three Y. lipolytica models iYLI647 (Mishra et al., 2018), iYali4 (Kerkhoven et al., 2016), and iMK735 (Kavšček et al., 2015) were tested and confirmed the in silico reduction of the TCA cycle activity with different oxygen uptake rates leading to the inactivation (Figure S7). It is interesting to note that TCA cycle metabolite decreases were not found for CIT, MAL, and FUM during shake-flask cultivation of the Crabtree-positive, natural carotenoid producer, Xanthophyllomyces dendrorhous (Alcalde and Fraser, 2018), but decreases were shown for lycopene-producing Y. lipolytica (Zhao et al., 2017), further suggesting a unique response of Y. lipolytica to oxygen scarcity.

Although evidence may support the reduced role of the TCA cycle during fermentations, other mechanisms and factors may contribute to the decreases in TCA cycle intermediates during the production phase. Some examples include (1) uptake of previously secreted TCA metabolites, which can reduce the need for Y. lipolytica to maintain large pool sizes and lead to diluted labeling; (2) dilution from relatively large pool sizes of TCA metabolites in both mitochondria and cytosol; and (3) increased diversion of carbon to products. In addition, Yarrowia may undergo a yeast-to-hyphae transition under low oxygen levels or fermentation stresses, leading to both metabolic alterations and morphological changes (Bellou et al., 2014).

Metabolic stress may also contribute to the low oxidative phosphorylation efficiency (Wu et al., 2016). Stress-induced metabolites such as carotenoids (e.g., beta carotene) can partition to the cell membrane (Wang et al., 2019) and damage cell membrane integrity (Verwaal et al., 2010, Liu et al., 2016). Prior work indicates that strains producing high levels of beta carotene are less stable than those grown for ionone production (Czajka et al., 2018), which could also be tied to differences in ATP demands fulfilled by electron transport. A recent study conducted in E. coli determined that an IPP-ATP complex was formed in IPP-overexpressing cells, and that the energy of cells dropped to a value of 0.40 from ∼0.80 during the fermentation (George et al., 2018). Although attempts to quantify the IPP-ATP complex were not made in this study, it may also contribute to the observed low energy charge (EC) state.

Limitations of the Study

Isotopic pulse-tracing methods cannot completely quantify and decipher all metabolic fluxes. However, 13C-metabolic flux analysis could not be applied as (1) the engineered strain requires complex, rich medium for growth and (2) unknown flux exchange routes exist among cellular compartments. In addition, although controls were used as described throughout the text, the quenching solution may not completely halt metabolism, which can add variability to the measurements (Figure S4). More comprehensive analyses (including transcriptomics, proteomics, metabolite overflow kinetics, and cell imaging) could reveal the cellular metabolic regulation during ionone fermentation. During 13C-pulse-tracing experiments, cells were sampled from the reactor and labeled with 13C-glucose under shake-flask conditions. The ex situ labeling may have unavoidable influence on cell metabolism.13C-pulse-tracing methods can adequately capture the activities of pathways; however, inactive pools and the high interaction of terpene synthesis with lipid metabolism may further confound the results (Liu et al., 2019).

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.