Impact of Oxygen Availability on the Organelle‐Specific Redox Potentials and Stress in Recombinant Protein Producing Komagataella phaffii

ABSTRACT

The yeast Komagataella phaffii (syn. Pichia pastoris) is a highly effective and well‐established host for the production of recombinant proteins. The redox balance of its secretory pathway, which is multi‐organelle dependent, is of high importance for producing secretory proteins. Redox imbalance and oxidative stress can significantly influence protein folding and secretion. Glutathione serves as the main redox buffer of the cell and cellular redox conditions can be assessed through the status of the glutathione redox couple (GSH‐GSSG). Previous research often focused on the redox potential of the endoplasmic reticulum (ER), where oxidative protein folding and disulphide bond formation occur. In this study, in vivo measurements of the glutathione redox potential were extended to different subcellular compartments by targeting genetically encoded redox sensitive fluorescent proteins (roGFPs) to the cytosol, ER, mitochondria and peroxisomes. Using these biosensors, the impact of oxygen availability on the redox potentials of the different organelles was investigated in non‐producing and producing K. phaffii strains in glucose‐limited chemostat cultures. It was found that the transition from normoxic to hypoxic conditions affected the redox potential of all investigated organelles, while the exposure to hyperoxic conditions did not impact them. Also, as reported previously, hypoxic conditions led to increased recombinant protein secretion. Finally, transcriptome and proteome analyses provided novel insights into the short‐term response of the cells from normoxic to hypoxic conditions.

Growth of yeast cells and the synthesis and folding of recombinant proteins are impacted by oxygen supply. The cellular organelles respond differently to a shift to limited oxygen, which is explained by marked changes of transcript and/or protein levels of many gene products during the oxygenation shift.

1. Introduction

The respiratory yeast Komagataella phaffii (syn Pichia pastoris) is a well‐established and versatile expression host for heterologous protein production with more than 300 licensed industrial processes and 70 commercial products already on the market (Barone et al. 2023; García‐Ortega et al. 2019). However, the production of more complex proteins as well as the overload of secretory proteins can trigger cellular stress responses (Mattanovich et al. 2004; Raschmanová et al. 2021).

One of the main bottlenecks associated with recombinant protein production is oxidative stress. Imbalance between the generation and elimination of reactive oxygen/nitrogen species (ROS/RNS) can significantly interfere with cellular redox homeostasis (He et al. 2017; Feige 2018; Lushchak and Lushchak 2021). One major source of ROS is the oxidative protein folding machinery in the endoplasmic reticulum (ER), where disulphide bond formation takes place. Overproduction of recombinant proteins can trigger the unfolded protein response (UPR), leading to elevated ROS generation (Tyo et al. 2012; Delic et al. 2013; Zahrl et al. 2023). Other main sources of ROS are the electron transport chain in the mitochondria and β‐oxidation of fatty acids in the peroxisomes. Prolonged exposure to high levels of ROS can lead to severe oxidative stress, DNA and protein damage, and even cell death (Turrens 2003; Farrugia and Balzan 2012; Perrone et al. 2008; Aksam et al. 2009; He et al. 2021). Thus, redox balance is a crucial aspect of cell survival (Trachootham et al. 2008; González‐Siso et al. 2009).

Oxygen availability has a strong impact on cell growth and recombinant protein production. The exact effect of oxygen on productivity can vary based on the respective protein of interest as it has been demonstrated in K. phaffii in both glucose‐limited (Baumann et al. 2007) and methanol feed cultivation (Li et al. 2007; Trentmann et al. 2004; Hellwig et al. 2001; Lee et al. 2003; Baghban et al. 2019). Interestingly, in many organisms, both high levels of oxygen (hyperoxia) and low levels of oxygen (hypoxia), are related to higher leakage and accumulation of ROS (Baez and Shiloach 2014; Zhao et al. 2012; Hermes‐Lima et al. 2015; Rintala et al. 2009; Lee et al. 2020; D'Aiuto et al. 2022). Hypoxic signalling has been linked to oxidative stress due to elevated mitochondrial ROS in yeast (Poyton 1999; Dirmeier et al. 2002; Guzy et al. 2007; Liu and Barrientos 2013) and mammalian cells (D'Aiuto et al. 2022; Fuhrmann and Brüne 2017; Zheng et al. 2021; Coimbra‐Costa et al. 2017; Maiti et al. 2006; Yu et al. 2020). Mitochondria are a major source of ROS such as H₂O₂ and superoxide radicals generated by electron leakage from the electron transport chain during aerobic respiration (Farrugia and Balzan 2012). When oxygen availability declines, the rate of respiration and oxygen consumption decreases (Jouhten et al. 2008). Yeast cells adapt to hypoxic conditions by building a more sufficient mitochondrial respiratory chain. A widely indirect sensing mechanism of mitochondrial ROS and nitric oxide (NO) production and sterol homeostasis has been described, suggesting that transition from normoxia to hypoxia increases ROS levels (Guzy et al. 2007; Liu and Barrientos 2013; Guzy and Schumacker 2006; Castello et al. 2006; Hamanaka and Chandel 2010; Chung and Lee 2020).

The main antioxidant systems dealing with ROS to maintain the redox balance in the cell are the thioredoxin (Trx) and glutaredoxin (Grx) systems. Both systems can catalyse thiol‐disulphide exchange reactions and are also considered as important mediators in redox signalling (He et al. 2017; Feige 2018; Grant 2001). The main redox buffer of the cell is glutathione (GSH), which together with its dimeric form (GSSG), builds a cysteine‐cystine couple. The redox status of the cell can be determined from the respective GSH and GSSG ratio, which has been used as a biomarker for oxidative stress. Even though glutathione is only synthesised in the cytosol, different organelles accommodate different glutathione pools (Schafer and Buettner 2001; Morgan et al. 2012; Aoyama and Nakaki 2015; Oestreicher and Morgan 2019). Each compartment has different mechanisms to regulate the redox state of its glutathione pool (Elbaz‐Alon et al. 2014). Consequently, investigating redox communication between different cellular components is regarded as both important and essential to understanding cellular metabolic changes.

One of the most common techniques for monitoring intracellular redox couples is based on genetically encoded fluorescent protein redox sensors. Redox‐sensitive green fluorescence proteins (roGFPs) were developed as such in vivo redox probes by the rational engineering of wild‐type GFP (roGFP1) and enhanced GFP (roGFP2). The engineered GFPs contain two cysteine residues closely positioned to the chromophore, ready to form a disulphide bond that causes small structural changes and, as a consequence, alters the protonation state of the chromophore and introduces a relative shift in intensity between the two protonation‐dependent excitation maxima of GFP. In this context, it is important to note that different variants of roGFP1 and roGFP2 have been developed (e.g., roGFP1‐iL, roGFP1‐iE and roGFP2‐iL), which have a less reducing midpoint potential than roGFP1 or roGFP2 and can therefore be used in organelles with a more oxidising environment such as the ER (Lohman and Remington 2008; Bilan and Belousov 2017). The biosensors equilibrate with the glutathione redox couple (GSH/GSSG) in the presence of glutaredoxins (Grxs) that act as catalysts between the roGFPs and the glutathione pool (Björnberg et al. 2006; Morgan et al. 2011). The roGFPs allow ratiometric and highly dynamic measurements of the glutathione redox state with high specificity and, importantly, are targetable to different subcellular compartments in living cells (Schwarzländer et al. 2016).

The aim of this study was to investigate the impact of diverse oxygen conditions on the glutathione redox potentials of different organelles in non‐producing and producing K. phaffii under industrially relevant conditions. For this purpose, different roGFP‐based redox biosensors were targeted to the cytosol, ER, mitochondria and peroxisome of K. phaffii and tested in small‐scale cultivations. After suitable sensors for the respective organelles were identified and characterised, the impact of oxygen availability on the organelle‐specific redox potentials was tested in glucose‐limited chemostat cultures. Finally, transcriptomics and proteomics analyses were performed to gain further insights into the (transient) physiological adaptations accompanying the observed changes in redox potentials of the different organelles and recombinant protein production levels in response to varying oxygenation levels.

2. Materials and Methods

2.1. Strains and Cloning

K. phaffii CBS7435 (Centraal Bureau voor Schimmelcultures, the Netherlands) was selected as a base strain for this study. Each of the redox‐sensitive GFP variants (roGFP1, roGFP1_iE and roGFP2) was N‐terminally fused to K. phaffiii glutaredoxin (encoded by PP7435_Chr4‐0111/GRX2) by a linker consisting of four glycine amino acids (Gly)₄. For targeting the roGFPs into the respective organelles, established organelle‐specific targeting sequences for K. phaffii were used (Table 1; (Waterham et al. 1997; Gutscher et al. 2008; Rodríguez‐Manzaneque et al. 2002; Pon and Schatz 1991)). Briefly, ER targeting was achieved by the N‐terminal S. cerevisiae Kar2 signal peptide and the C‐terminal HDEL ER retention signal, while the Grx5 signal peptide was used for targeting to the mitochondrial matrix and the C‐terminal extension SKL for peroxisomal localization. The Grx2‐roGFPs fusion‐constructs were ordered as gene blocks (Integrated DNA Technologies). For transcriptional control of roGFP1 and roGFP2, the strong constitutive GAP (glyceraldehyde‐3‐phosphate dehydrogenase) promoter was used. Due to the observation that expression of roGFP1_iE in the ER using P_GAP led to a growth defect, the weaker ADH2 (alcohol dehydrogenase 2) promoter (Prielhofer et al. 2017) was employed. The roGFP sensors use a ratiometric approach (see below) which is independent of the sensor expression levels, ensuring measurement consistency regardless of sensor concentrations (Lohman and Remington 2008; Schwarzländer et al. 2016). The CYC1 terminator region was used for transcriptional termination and the nourseothricin (NTC) resistance cassette as a selection marker. All roGFP expression plasmids were targeted to the 5′‐RGI2 locus for genomic integration. Plasmid construction was done via Golden Gate Assembly (GGA) using the GoldenPiCS platform described by Prielhofer et al. (2017). Transformation into K. phaffii was done as described by Gasser et al. (2013). Integration of the expression vector into the correct locus was confirmed by colony PCR and correct targeting of the various roGFPs into the respective organelle was confirmed by fluorescence microscopy (see below). Porcine trypsinogen (trp) under control of P_GAP was used as a reporter protein. The trypsinogen expression vector described by Delic et al. (2014) served as a template for the construction of the new vectors via GGA cloning (Prielhofer et al. 2017), using the S. cerevisiae alpha‐mating factor prepro‐leader for secretion, the CYC1tt for transcriptional termination and AOXtt as genomic integration locus. For the P_GAP_trp strains the selection marker was flanked by loxP sites and removed through transient expression of Cre recombinase (Sauer 1987). The resulting trypsinogen expression vectors were transformed into CBS7435 as described before. After initial productivity screenings, one average producer was selected with a single copy of the trypsinogen expression cassette (see below). A comprehensive list of all constructed strains in this study is provided in Table 1.

TABLE 1.

List of strains used in this study. For all strains K. phaffii CBS7435 served as a background strain.

Short name	Targeted organelle	Type of roGFP	Promoter roGFP	Targeting signal	Promoter and model protein	Category
Wild‐type	—	—	—	—	—	Wild‐type
Cyt	Cytosol	Grx2‐ roGFP1	PGAP	—	—	Non‐producing + roGFPs
ER	ER	Grx2‐ roGFP1_iE	P ADH2	Kar2‐SP/HDEL	—	non‐producing + roGFPs
M	Mitochondria	Grx2‐ roGFP1	PGAP	Grx5‐SP	—	Non‐producing + roGFPs
P	Peroxisome	Grx2‐ roGFP2	PGAP	SKL	—	Non‐producing + roGFPs
C + trp	Cytosol	Grx2‐ roGFP1	PGAP	—	PGAP _trypsinogen	Producing + roGFPs
ER + trp	ER	Grx2‐ roGFP1_iE	P ADH2	Kar2‐SP/HDEL	PGAP _trypsinogen	Producing + roGFPs
M + trp	Mitochondria	Grx2‐ roGFP1	PGAP	Grx5‐SP	PGAP _trypsinogen	Producing + roGFPs
P + trp	Peroxisome	Grx2‐ roGFP2	PGAP	SKL	PGAP _trypsinogen	Producing + roGFPs

2.2. Small‐Scale Screening Cultivations

Small‐scale screenings for the determination of organelle‐specific redox potentials were performed in the 24‐deep‐well plate (DWP) format. Per construct, at least 12 (up to 24) biological clones (distinct strains) were tested per construct and strain background (non‐producing and producing K. phaffii) and at least three independent cultivations per clone were performed. For pre‐cultures, cells were grown in 2 mL of YPD (1% yeast extract, 2% peptone, 2% glucose) containing 100 mg L⁻¹ NTC for 24 h on a rotary shaker at 280 rpm and 25°C. For main cultures, 2 mL of ASM minimal medium (Baumschabl et al. 2020) containing 50 g L⁻¹ of polysaccharide (EnPump200 polysaccharide, Enpresso) and 0.7% glucose‐releasing enzyme (Reagent A, Enpresso) were inoculated to a final OD₆₀₀ of 8.0 using washed cells from pre‐cultures and incubated for approx. 48 h under the same conditions as described before. Subsequently, 1 mL of cell suspension was harvested for measuring the redox potentials.

2.3. Determination of Organelle‐Specific Redox Potentials

The protocol described by Delic et al. (2014) was down‐scaled to the 96‐well format, allowing for a more efficient and rapid sample processing procedure. Briefly, cells were diluted in 24‐DWPs to an OD₆₀₀ between 2.0 and 5.0 with PBS. For every fluorescence measurement, 3 aliquots of 160 μL diluted cell suspension were transferred to a dark fluorescence plate (FluoroNunc, Nunc) and mixed with 40 μL of the respective reagents: (a) PBS (untreated cells), (b) 1 M dithiothreitol (DTT) (full reduction of the roGFPs), or (c) 6.3 mM of 4,4′‐dipyridyl disulphide (4‐DPS) (full oxidation of the roGFPs). Fluorescence levels were measured immediately afterwards on a fluorescence photometer (Infinite M200 Tecan plate reader, Molecular Devices, Sunnyvale, CA). For the oxidised and reduced forms of roGFP1 and roGFP1_iE, the excitation wavelengths were set to 395 nm and 465 nm, respectively (emission = 515 nm), while for roGFP2, they were set to 406 nm and 488 nm (emission = 540 nm). Using the respective midpoint potential E^o' _roGFP of each roGFP variant (−287 mV for roGFP1; −236 mV for roGFP1_iE and − 280 mV for roGFP2), the organelle redox potential was determined by Equation (1) and Equation (2), while the degree of oxidation was calculated by Equation (3), where, R/(1−R) is the ratio of the reduced over oxidised roGFP indicator, F is the ratio of fluorescence intensity values of a sample excited at two wavelengths, F_ox is the ratios of fluorescence intensities of fully oxidised roGFP, F_red is the ratios of fluorescence intensities of fully reduced roGFP, I_oxλ1/I_{red λ1} is the fluorescence intensities at excitation wavelength, λ1 is the quotient of fluorescence intensities called ‘instrument factor’ and corrects for variations in the signal strength from the instrument, E^o' _roGFP is the midpoint potential of the applied roGFPs and E^o' is the determined glutathione redox potential.

$$\frac{R}{1-R}=\frac{F-F_{\mathit{ox}}}{F_{\mathit{red}}-F}\left(\frac{I_{\mathit{ox\lambda }1}}{{{\rm I}}_{\mathit{\text{redλ}}1}}\right)$$ (1)

$$E^{o^{\prime }}={\mathrm{E}}_{\text{roGFP}}^{{\mathrm{o}}^{\prime }}-29.6\mathit{mV}\times \mathit{log}\frac{\mathrm{R}}{1-\mathrm{R}}$$ (2)

$$\mathrm{OxD}=\frac{\mathrm{F}-F_{\mathit{red}}}{\frac{I_{\mathit{ox\lambda }1}}{I_{\mathit{\text{redλ}}1}}\left(F_{\mathit{ox}}-F\right)+\left(F-F_{\mathit{red}}\right)}.$$ (3)

2.4. Fluorescence Microscopy

Fluorescence microscopy was carried out to confirm the correct localization of the roGFPs in the different organelles. For this purpose, cells were harvested at the end of the 24‐DWP screening procedure as described above and resuspended in 100 μL of PBS. Three microliters of the cell suspension were placed on a glass slide, covered with a cover slip, and viewed under a Leica DMI 6000 fluorescence microscope (Zeiss), employing the appropriate bandpass filter for GFP detection (508–601 nm). The ZEN lite software (Zeiss) was used for image analysis.

2.5. Chemostat Cultivations

Chemostat cultivations were performed in 1 L DASGIP benchtop bioreactors (SR0700ODLS; Eppendorf AG, Germany) or 1.8 L DASGIP benchtop bioreactors (SR1500ODLS; Eppendorf AG, Germany). The media compositions used for batch and chemostat cultivation are described below. For the entire cultivation, the temperature was set at 25°C and the pH was controlled at 5.5 by automated addition of 12.5% NH₄OH. The bioreactors were inoculated to a starting OD₆₀₀ of 3.5. During the batch phase, the dissolved oxygen (DO) was kept above 30% by automated adjustment of the stirrer speed between 400 and 1200 rpm and the air flow between 9.5 and 30 sL h⁻¹. Glucose‐limited chemostat cultivations were initiated after the batch‐end was reached (indicated by sudden spike in DO) at a dilution rate of 0.075 h⁻¹. The chemostat working volume of 0.6 L was maintained by means of a level sensor. The stirrer speed was set between 690 and 710 rpm and the air flow was set at 0.55 vvm. The different oxygenation conditions during chemostat cultivation were established by varying the oxygen concentration in the inlet air: (a) hyperoxia (31% O₂), (b) normoxia (21% O₂), (c) oxygen‐limited (14% O₂) and (d) hypoxia (8% O₂). For the first 66 h of the chemostat phase (corresponding to five resident times), chemostats were operated under normoxia to establish base‐line conditions. When two different oxygenation conditions were tested during the same chemostat cultivation, normoxic conditions were applied in‐between the respective oxygenation conditions for 66 h in order to return to base‐line conditions.

Two biological replicates per construct and K. phaffii strain background were tested in different rounds of chemostat cultivations. The non‐producing strains expressing roGFPs were tested under normoxic, hyperoxic, oxygen‐limited and hypoxic conditions. The non‐producing and producing strains expressing roGFPs were analysed under normoxic and hypoxic conditions. Finally, for the transcriptome and proteome analysis (see below), the wild‐type strains were used.

Batch medium contained per litre: 46.5 g glycerol (86%), 2.0 g citric acid, 12.6 g (NH₄)₂HPO₄, 0.022 g CaCl₂*2H₂O, 0.9 g KCl, 0.5 g MgSO₄*7H₂O, 4 mL Biotin (0.1 g L⁻¹), 4.6 mL trace salts stock solution. The pH was set to 5.5 with 25% (w/w) HCl.

Chemostat medium contained per litre: 44.0 g glucose, 2.0 g citric acid, 17.4 g (NH₄)₂HPO₄, 0.03 g CaCl₂*2H₂O, 2.0 g KCl, 0.8 g MgSO₄*7H₂O, 3.2 mL Biotin (0.1 g L⁻¹), 1.94 mL trace salts stock solution. The pH was set to 5.5 with 25% (w/w) HCl.

Trace element solution contained per litre: 6.0 g CuSO₄*5H₂O, 0.08 g NaI, 3.36 g MnSO₄*H₂O, 0.2 g Na₂MoO₄*2H₂O, 0.02 g H₃BO₃, 0.82 g CoCl₂, 20.0 g ZnCl₂, 65.0 g FeSO₄*7H₂O, and 5.0 mL H₂SO₄ (95%–98% w/w).

2.6. Bioreactor Sample Analysis

Yeast dry mass (YDM) was determined in technical triplicates. Briefly, 2 mL of culture broth were transferred to a pre‐dried and pre‐weighted microcentrifuge tube and centrifuged for 5 min at 1200 g and 4°C. After the supernatant was removed, the pellets were washed twice with deionised water employing the same settings and then dried at 105°C for at least 25 h. Finally, the tubes were weighed again.

Extracellular glucose, glycerol, ethanol and arabitol concentrations in supernatants were quantified by HPLC as described by Pflügl et al. (2012).

For the wild‐type strains, storage carbohydrate content in the form of glycogen and was determined as described previously (Rebnegger et al. 2016, 2024). For the analysis, the pellets were resuspended and diluted in Na₂CO₃ (0.25 M) and processed as described previously (Parrou and François 1997). Glucose samples were incubated overnight with α‐amyloglucosidase (from Aspergillus niger; Sigma‐Aldrich, The Netherlands) at 57°C and measured by HPLC after overnight incubation.

For the trypsinogen producing strains, the secreted model protein concentration was quantified by microfluidic capillary electrophoresis using the LabChip GX /GXII System (PerkinElmer) system as described in Staudacher et al. (2022).

An off‐gas analyser was used to measure O₂ and CO₂ values in for the calculation of CO₂ evolution rates (CER) and O₂ utilisation rates (OUR).

Flow cytometry was performed for verification of cell viability and for in vivo detection of H₂O₂. Cell viability was measured based on propidium iodide staining by flow cytometry as described by Hohenblum et al. (2003). Assessment of H₂O₂ levels at different oxygenation conditions was done by dihydrorhodamine 123 (DHR) fluorescence dye as described in (Delic et al. 2013, 2012). Briefly, cells were diluted in PBS to OD₆₀₀ = 0.4, centrifuged and the pellet was resuspended in 2 mL PBS. DHR was added at a final concentration of 5 μg/mL and the cells were incubated for 45 min at 30°C in the dark with gentle shaking. As a negative control, unstained cells were used. As a positive control, cells were incubated with 10 mM H₂O₂ for 40 to 60 min. Flow cytometry analysis was performed using a FACS Canto with the green filter (525–550 nm).

2.7. RNA‐Seq Analysis

Total RNA was isolated from cell pellets sampled during chemostat cultivations under normoxic and hypoxic conditions as described by Ata et al. (2018). Four sampling points were selected for RNA‐Seq analysis. The first sample was taken after 66 h (T0; 5 resident times) of cultivation at normoxic conditions and the remaining three samples were taken after the switch to hypoxic conditions, specifically after 5 h (T1; 0.5 resident times), 50 h (T2; 4.0 resident times) and 70 h (T3; 5.2 resident times) had passed, respectively. Samples were further processed and analysed by the Vienna Biocenter Core Facilities GmbH (AT) for sequencing and statistical data analysis. Samtools 1.9 was used for merging and sorting the raw data and cutadapt 1.18 for trimming (Li et al. 2009; Martin 2011). Alignment was carried out with BowTie2 2.3.4.1 against the reference sequence of K. phaffii CBS7435 (Langmead and Salzberg 2012). The count data of all sample reads were calculated with kallisto quant v0.44.0 (Bray et al. 2016). Differentially expressed genes were categorised by the Genesis software tool (Sturn et al. 2002). The statistical analysis of the differentially expressed genes included a log₂ fold change threshold of 0.59 that corresponds to a 1.5‐fold change along with adjusted p < 0.05. Significantly enriched GO terms for the clusters were defined using online GO Term Finder (http://go.princeton.edu/cgi‐bin/GOTermFinder) and annotations from the Saccharomyces Genome Database (SGD) (Cherry et al. 2011).

2.8. Proteomics Sample Preparation Analysis

The samples for proteomics analysis were collected from three independent chemostat cultivations after 66 h under normoxic conditions (T0) as well as after 5 h (T1) and 70 h (T3) under hypoxic conditions. Per cultivation and sampling point, 2 mL of culture were harvested and diluted to an OD₆₀₀ of 8 in 2 mL of TE Buffer (10 mM Tris, 1 mM EDTA, pH 7.5). The cell pellets were washed twice with TE Buffer and mechanically disrupted using a FastPrep24 (3 cycles at maximum acceleration for 30 s; MP Biomedicals). The protein concentration of the resulting protein extracts was determined using a BCA assay according to the manufacturer's instructions (Pierce BCA Protein Assay Kit, ThermoFisher Scientific). For all samples, 20 μg protein was tryptically digested using the S‐Trap protocol according to the manufacturer's instructions (Protifi).

2.9. Proteomics Measurements

Peptide mixtures were separated on an Easy nLC 1200 coupled online to an Orbitrap Elite mass spectrometer (ThermoFisher Scientific). In‐house self‐packed columns (i.d. 100 μm, o.d. 360 μm, length 200 mm) packed with 3.0 μm Dr. Maisch Reprosil C18 reversed‐phase material (ReproSil‐Pur 120 C18‐AQ) heated to 45°C were loaded with 18 μL of 0.1% (v/v) acetic acid at a maximum pressure of 500 bar. Peptide elution was performed in a 180 min non‐linear gradient from 1% to 99% solvent (0.1% (v/v) acetic acid in 95% (v/v) acetonitrile) at a constant flow rate of 300 nL/min. Eluted peptides were measured in the Orbitrap with a resolution of R = 60,000 with lock‐mass correction activated. Following each MS‐full scan, up to 20 dependent scans were performed in the linear ion trap after collision‐induced dissociation fragmentation (CID) based on the precursor intensity. Dynamic exclusion was enabled (exclusion size list 500, exclusion duration 30 s) with ±10 ppm exclusion window.

Raw files were imported into MaxQuant (2.2.0.0) for data processing and protein identification. Protein database searches were performed against a forward K. phaffii CBS7435 database (UP000006853) with common contaminants and reverse entries added by MaxQuant with the following parameters: peptide tolerance, 4.5 ppm; min fragment ions match per peptide, 2; primary digest reagent, trypsin; missed cleavages, 2; variable modifications, carbamidomethyl C (+57.0215), heavy carbamidomethyl C (+61.04072, for samples with differential cysteine labelling only), oxidation M (+15.9949), acetylation N, K (+42.0106). Results were filtered for a 1% false discovery rate (FDR) on spectrum, peptide, and protein levels. Processed data were analysed using Python 3.9. Numpy and Pandas libraries were used for data importation and cleaning coupled to an in‐house pipeline. Normalised LFQ was used for relative quantification of the identified proteins with a minimum of 2 valid values per condition. Scipy and Statsmodels packages were used for statistical analysis of the quantified proteins. Fold changes were calculated from averaged log₂‐transformed LFQ intensities and t‐tested for significance. Resulting p‐values were corrected using FDR correction (α = 0.01). Significance was considered for fold change (FC) > 1.5 and adjusted p‐value (adj. p) < 0.05.

2.10. Multi‐Omics

For the integrated multi‐omics data analysis, an sPLS‐DA model was set up using the mixOmics R package, and a MOFA model was set up using the MOFA2 R package (Rohart et al. 2017; Argelaguet et al. 2018). The results were contextualised based on their annotations in SGD and the pichiagenome database (Cherry et al. 2011; Valli et al. 2016).

3. Results

3.1. Characterisation of the Biosensors in Small‐Scale Cultivation

Selecting suitable biosensors for each cellular compartment is a critical initial step in determining the glutathione redox potentials of different organelles. The selection was based on the midpoint potential of the available redox‐sensitive GFPs (see Materials and Methods), and the organelle‐specific redox potentials of yeasts (if available) and other organisms reported in literature (Meyer and Dick 2010). roGFP1 has been regularly employed for measuring the redox potential of the more reducing environment of the cytosol while roGFP1_iE has been successfully applied for measurements in the more oxidised environment of the ER (Delic et al. 2010). In Saccharomyces cerevisiae , roGFP2 has been previously used for the study of the mitochondrial (Kojer et al. 2012) and the peroxisomal (Elbaz‐Alon et al. 2014; Radzinski et al. 2018) redox state. To investigate the impact of recombinant protein production on cellular redox potentials, porcine trypsinogen was chosen as a model protein. Properly folded porcine trypsinogen contains six disulphide bonds, which requires the protein to undergo several steps of oxidative protein folding without inducing UPR (Delic et al. 2010).

To determine the organelle‐specific glutathione redox potentials of non‐producing and producing K. phaffii, small‐scale screenings in the 24‐DWP format under glucose‐limiting conditions were performed, and the redox potential of each organelle calculated as described in the Materials and Methods section (Table 2). Based on reports in literature suggesting that the fusion of a given roGFP with glutaredoxin (Grx) can provide higher specificity and speed of the thiol/disulphide exchange between the intracellular GSH/GSSG couple and the redox‐sensitive protein (Björnberg et al. 2006; Schwarzländer et al. 2016; Pastor‐Flores et al. 2017), roGFP1 and roGFP1_iE were tested with and without fusion to K. phaffii Grx2. As the Grx2‐roGFP constructs showed better reproducibility, especially in the ER (Figure S1A); it was decided to continue with the fused versions for all employed roGFPs. The Grx2‐fused version of roGFP1 in the mitochondria and of roGFP2 in the peroxisomes showed high accuracy too (Figure S1B). The correct localization of the roGFPs was confirmed by fluorescence microscopy (Figure S2).

TABLE 2.

Organelle‐specific redox potentials and degrees of oxidation of the roGFPs in small‐scale screenings in minimal media under glucose‐limited conditions.

Organelle	Redox potential ± SD (mV)		Degree of oxidation ± SD (%)
	Non‐producing	Producing	Non‐producing	Producing
Cytosol	−292 mV ± 0.6	−292 mV ± 1.2	72% ± 1.0%	67% ± 2.5%
ER	−230 mV ± 1.1	−222 mV ± 0.4	80% ± 1.5%	86% ± 0.8%
Mitochondria	−274 mV ± 0.6	−275 mV ± 2.2	87% ± 0.4%	89% ± 2.1%
Peroxisome	−275 mV ± 2.0	−277 mV ± 1.4	88% ± 2.1%	77% ± 1.2%

In the non‐producing K. phaffii strains, the average redox potential of the cytosol was −292 mV, indicating a highly reduced glutathione pool. In contrast, the glutathione pool of the ER was found to be highly oxidised, with a redox potential of −230 mV. Both results were in good agreement with results for other organisms described in literature and the previous measurements of the cytosolic and ER redox potentials in K. phaffii strain X‐33 by Delic et al. (2012), in which the redox potentials of the cytosol and the ER using non‐fused roGFP sensors were determined to be −303 and −242 mV, respectively. In the present study, the range of organelles was extended to the mitochondria and the peroxisomes, for which the redox potentials were measured for the first time in K. phaffii. Mitochondria are a primary source of ROS, formed by electron leakage from the electron transport chain (Farrugia and Balzan 2012) while peroxisomes are crucial organelles in glucose conditions based on their involvement in multiple metabolic processes (e.g., fatty acid oxidation, ether lipid synthesis and ROS metabolism) and their critical role in regulating cellular reactions to different types of stress such as oxidative stress and hypoxia (He et al. 2021). It was found that the mitochondria and peroxisomes have a less reduced glutathione pool than the cytosol, with a glutathione redox potential of −274 and −275 mV, respectively. In the producing strains, the redox potentials of the cytosol, mitochondria and peroxisomes were similar to those of the corresponding non‐producing strains. However, the ER showed a slightly more oxidised environment, with a redox potential of −222 mV, which likely stems from the impact of recombinant protein production and disulphide bond formation in the ER. The degree of oxidation of the different roGFPs was found to vary between 67% and 90% (Table 2), confirming that the employed roGFPs were well suited for measuring the redox potentials of the respective organelle (Meyer and Dick 2010). The difference between the degree of oxidation of the non‐producing (88%) with the producing strains (77%), expressing roGFP2 in the peroxisomes, is a result of the fluorescence signal deviation of the fully reduced cells between the two independent screenings.

3.2. Hyperoxic and Oxygen‐Limited Conditions Do Not Influence the Organelle‐Specific Redox Potentials of K. phaffii Grown in Glucose‐Limited Chemostat Cultures

As mentioned above, cell adaptation to different oxygen levels is highly important for survival (Chung and Lee 2020). In S. cerevisiae , hyperoxia toxicity is linked to mitochondrial redox state shift and superoxide‐related damage (Outten et al. 2005), while hyperoxia was not studied in K. phaffii before. Oxygen‐limited conditions have a strong impact on cellular metabolism, which was shown to be different between S. cerevisiae and K. phaffii (Baumann et al. 2010). Previously, redox potential measurements were limited to small‐scale cultivations that pose limitations to a fully controlled environment. Therefore, chemostat cultivations were chosen to assess the impact of different oxygen levels on the organelle‐specific redox potentials.

In order to get a better understanding of the impact of different oxygen conditions on the organelle‐specific redox potentials, representative, non‐producing K. phaffii strains expressing the different roGFPs were selected from small‐scale screenings and exposed to excess of dissolved oxygen (DO) (Hyperoxia, DO = 80%, inlet O₂ = 31%) and limited‐oxygen availability (Oxygen‐limited, DO = 4%, inlet O₂ = 14%) in glucose‐limited chemostat cultures operated at a dilution rate of 0.075 h⁻¹. Baseline conditions were established by cultivating the cells at normoxic conditions (Normoxia, DO = 30%, inlet O₂ = 21%) until steady‐state was reached (66 h/5 resident times). Unexpectedly, the redox potentials of the investigated organelles remained rather unaffected by the different oxygenation conditions (Figure 1A,B). The estimated average redox potentials in chemostats for the cytosol and the ER were very similar to those measured in small‐scale screenings (Table 2), while the average redox potential of the mitochondria (E_GSH = −265 mV) was slightly more oxidised and the average redox potential of the peroxisomes (E_GSH = 294) substantially more reduced than in the screenings (E_GSH = −274 mV and E_GSH = −275 mV, respectively). As the biomass concentration was not affected in the low‐oxygen condition (Figure S3) and no ethanol production was observed, the limited‐oxygen supply was seemingly still sufficient to support full respiratory growth. Hence, it was decided to test if the redox state of K. phaffii is affected by an even stronger oxygen‐limitation (hypoxia) which induces a switch to respiro‐fermentative metabolism.

FIGURE 1.

Measurements of glutathione redox potentials of non‐producing K. phaffii strains during hyperoxic, normoxic and oxygen‐limited conditions. Chemostat cultivations were performed in glucose‐limiting conditions using three different oxygenation conditions. (A) The redox potential of four organelles, namely the cytosol (black dots), the ER (blue triangle), the mitochondria (green square) and the peroxisome (red triangle) are shown across the different oxygen conditions (inlet O₂; secondary axis; blue area) over chemostat cultivation time. (B) Redox measurements in the table refer to the average values and standard deviation of two technical replicates for each oxygen condition.

3.3. Hypoxia has a Strong Impact on the Organelle‐Specific Glutathione Redox Potentials of Non‐Producing and Producing K. phaffii

As hyperoxia and oxygen‐limited conditions in chemostat cultures did not lead to any significant changes in the redox potentials of the different cell organelles, further investigations were focused on hypoxic conditions. As mentioned before, hypoxia was previously demonstrated to increase recombinant protein production rates of K. phaffii (Baumann et al. 2007) and is therefore not only of interest from a physiological perspective but also from a bioprocessing point of view. Thus, the trypsinogen producing strains were included in this set of experiments.

Glucose‐limited chemostat cultivations of non‐producing and producing K. phaffii expressing the different roGFPs were performed in a similar manner as described before. Initially, baseline conditions were established by cultivating the cells under normoxic conditions (DO = 30%, inlet O₂ = 21%) until steady‐state was reached. After the switch to hypoxia (DO = 0%, inlet O₂ = 8%), sampling intervals were again kept rather short in order to be able to also monitor responses of a more transient nature to hypoxic conditions of the respective glutathione redox potentials of the four targeted organelles.

In addition to the roGFP measurements, the biomass and secreted metabolite concentrations were also analysed. As expected under hypoxic conditions, a clear gradual decrease in the biomass concentration from approx. 20 g L⁻¹ (normoxic conditions) to approx. 10 g L⁻¹ was observed (Figure S4), while the ethanol and arabitol concentrations gradually increased to roughly 6 g L⁻¹ ethanol and 3 g L⁻¹ arabitol for all cultivated strains at the end of the hypoxic phase (Figure S5A), clearly demonstrating that the cells had switched their metabolism from fully‐respiratory to respiro‐fermentative growth. Considering the decrease in biomass, the specific productivity (q _P) of ethanol and arabitol also exhibited a steady increase by the end of hypoxia (Figure S5B). Additionally, the specific productivity (q _P) of the porcine trypsinogen was calculated. In agreement with previous reports for P_GAP‐based recombinant protein production under hypoxic conditions (Baumann et al. 2007), an approx. 1.7‐fold higher q _P compared to normoxic conditions was estimated for all analysed producing strains (Figure S6). It should be noted that the last time point considered in q _p calculations was at 50 h of hypoxic conditions. Up to that point, the cells exhibited similar physiological behaviour. However, at 70 h of hypoxic conditions, the cultures deviated in terms of their biomass concentration.

Strikingly, the hypoxic condition had indeed a drastic impact on the redox potentials of all four investigated organelles and, interestingly, the measured redox potentials of the ER and even more so of the mitochondria of producing K. phaffii diverged clearly from those of non‐producing K. phaffii. Slight differences were also observed in how the redox potentials of the cytosol adapted, while the redox potentials of the peroxisome remained seemingly unaffected by the production of the recombinant protein (Figure 2).

FIGURE 2.

Organelle‐specific glutathione redox potentials of non‐producing and producing K. phaffii strains under hypoxic conditions. Chemostat cultivations were performed to determine the transition of the redox potentials from normoxic to hypoxic conditions. Different patterns were observed between the different organelles as well as between the non‐producing and the producing strains. The measurements represent the average value of two biological replicates. Error bars represent the standard deviation calculated between the two biological replicates. The dotted line represents two period average moving trendline. Time intervals between biomass samplings are indicated by shaded backgrounds, with the respective generation numbers (n _gen) indicated.

In the cytosol, the glutathione redox potential of the non‐producing strains became more oxidised as it increased from −293 mV to −285 mV within the first 5 h of hypoxia and remained at this state for at least 25 h, after which it returned to the levels observed in normoxic conditions. The producing strains showed a similar pattern throughout the first 30 h in hypoxic conditions, but instead of returning to the more reduced state, the cytosolic redox potential remained at the more oxidised level (E = −285 mV) until the end of the cultivation.

Long‐term adaptation (50 h) to hypoxic conditions led to a more oxidised redox potential of the ER, in both, the non‐producing and in the producing strains. However, while in the non‐producing strains the redox potential gradually became more oxidised, it remained mostly stable until at least 30 h of hypoxic cultivation in the producing strains, after which it became oxidised to −195 mV. In the last 20 h of the cultivation, the redox potential in the producing strains clearly returned to a less oxidised state, however, similar to the non‐producing strains, a large variation between the two biological replicates was observed throughout the entire cultivation, which was even more pronounced at the end of the cultivation.

The biggest differences in response to hypoxia of the glutathione redox potential between the non‐producing and producing strains were seen for the mitochondria. In case of the non‐producing strains the glutathione redox potential increased from −256 mV to −245 mV within the first 5 h, while after 20 h in hypoxic conditions, the redox state was already more reduced and continued to become even more reduced throughout the remaining cultivation time (−275 mV). In contrast, the mitochondrial redox potential of the producing strains remained stable at −253 mV for at least 30 h in hypoxic conditions. Nevertheless, after 50 h, the redox potential switched from −256 mV to −244 mV for biological replicate 1 and to −251 mV for the other. Regardless of the last change, at the end of hypoxic cultivation (70 h), the redox potential of both strains went back to its reduced state with E = −265 mV.

It is important to note that both, roGFP1_iE in the ER and roGFP1 in the mitochondria, reached a maximum OxD of approx. 95% at various time‐points. Thus, specifically the highest redox potentials for the ER and mitochondria should be interpreted with some caution. Nevertheless, both organelles became (partly) very oxidised in response to hypoxia which might indicate that the cells were facing oxidative stress.

The redox potential of the peroxisomes remained relatively stable throughout the first 5 h in hypoxia but gradually decreased to −301 mV thereafter. However, throughout the last 15 h of cultivation the redox potentials returned to normoxic levels.

3.4. At the End of the Hypoxic Conditions the Specific Oxygen Uptake Rates Reach Levels Similar to Those Observed in Normoxic Conditions

One important step in determining the effects of oxidative stress and the production of reactive oxygen species on cell redox biology is to understand the rate at which cells consume oxygen (Wagner et al. 2011). To assist this, the specific oxygen uptake rates (q _O2) of the non‐producing and producing strains expressing roGFPs as well as the wild‐type strains were calculated during the transition phase from normoxia to hypoxia. Additionally, the respiratory quotient (RQ) values confirmed the establishment of hypoxic conditions (Table S1).

When hypoxic conditions were applied to the chemostat cultures, the q _O2 of all strains was significantly reduced from approx. 1.5 mmol g⁻¹ h⁻¹ to approx. 0.85 mmol g⁻¹ h⁻¹. As the cells were adapting to hypoxia, q _O2 was steadily increasing again, reaching similar or even higher levels than those observed in normoxic steady‐state. The non‐producing strains reached approx. 1.6 mmol g⁻¹ h⁻¹, the producing strain approx. 2.0 mmol g⁻¹ h⁻¹and the wild‐type 1.5 mmol g⁻¹ h⁻¹ (Figure 3). This was an unexpected result. Both the wild type and non‐producing roGFP strains that have seemingly fully adapted to hypoxia managed to consume the same amount of oxygen as in normoxic steady‐state while the producing strains that co‐expressed roGFP showed an even higher q _O2 at the end of the hypoxic phase. If biomass reduction is taken into consideration, one can speculate that cells can still uptake oxygen in a sufficient manner. Also, the higher q _O2 values of the producing strains (expressing trypsinogen and roGFPs) could be a result of an increased energy demand imposed by recombinant protein production. The only strain that showed reduced specific oxygen uptake rates at the end of hypoxia was the trypsinogen producing strain expressing roGFP in the ER (normoxia = 1.9 mmol g⁻¹ h⁻¹, end of hypoxia = 1.6 mmol g⁻¹ h⁻¹). That strain, though, had shown growth deficiencies and very low trypsinogen productivity. The latter is probably due to the fact that there is competition between the trypsinogen and the roGFP‐sensor for entry into the ER, as described by Barrero et al. (2018). Consequently, the q _O2 values of this strain were not included in the calculations.

FIGURE 3.

Specific oxygen uptake rates of non‐producing and producing K. phaffii expressing roGFPs. Average q _O2 values were calculated from independent glucose‐limited chemostat cultivations under normoxic (0 h) and hypoxic conditions. Average values were derived from two biological replicates of different roGFP‐expressing strains in the non‐producing (red square) and producing strains (blue circle), respectively, while values for the wild‐type strains (black rhombus) were calculated from four technical replicates. The producing strains expressing roGFP in the ER was not included in the calculations. Error bars represent the standard deviation. Time intervals between biomass samplings are indicated by shaded backgrounds, with the respective generation numbers (n _gen) indicated.

3.5. Global Transcriptome and Proteome Analysis Revealed Initial Metabolic Response to Hypoxia and Gene Regulation for Cell Adaptation Under Hypoxic Conditions

The profile of the organelle‐specific redox potentials and the q _O2 levels in response to hypoxic conditions indicate that short‐term response and long‐term adaptation to hypoxia in K. phaffii may vary. Long‐term adaptation of K. phaffii to hypoxia in chemostat cultures (O₂ = 8%; D = 0.1 h⁻¹; 5 resident times/66 h) on a transcriptome, proteome and metabolic flux level has been studied previously by Baumann et al. (2010) and included the transcriptional upregulation of glycolytic and non‐oxidative pentose phosphate pathway genes, ergosterol and sphingolipid biosynthesis genes as well as genes related to the general stress response and, more specifically, protein folding stress. Furthermore, marked downregulation in response to hypoxia was observed for genes with a role in the TCA cycle. The respective transcriptional changes were positively correlated to the gathered proteome and metabolic flux data sets, while in a follow up study it was demonstrated that in contrast to the expression trends of the respective biosynthetic genes, the cellular ergosterol and sphingolipid content actually decreased significantly (Adelantado et al. 2017).

To investigate if there are indeed differences between the short‐ and long‐term response of K. phaffii to hypoxic conditions, glucose‐limited chemostat cultivations (D = 0.075 h⁻¹) were performed and samples for transcriptome and proteome analysis were taken from steady‐state cultures in normoxic conditions (T0), as well as 5 h (T1), 50 h (T2) and 70 h (T3) after the switch to hypoxia. Both analyses were limited to the K. phaffii wild‐type as transcriptome comparison of the non‐expressing and expressing K. phaffii strains by Baumann et al. (2010) only yielded 6 differentially expressed genes.

Pairwise comparisons between sampling points T1 and T0, T2 and T0 and T3 and T0, yielded 1238, 1207 and 1067 differentially expressed genes (DEGs) (cut‐off criteria: adj. p < 0.05; FC = 1.5), respectively. Furthermore, pairwise comparisons between sampling points T1 and T2 as well as T1 and T3 resulted in 693 and 551 DEGs, respectively, suggesting that there is indeed a substantial change in the transcriptional make‐up of cells that have been exposed for shorter or longer periods to hypoxic conditions. In contrast, only six DEGs were identified between sampling points T2 and T3. Hence, this comparison was not further investigated. Venn diagram analysis was applied to identify shared and unique DEGs between the respective pairwise comparisons to sampling point T0 (Figure 4A). According to this analysis, 664 DEGs were shared across all three comparisons, whereas the second largest gene set of 430 DEGs was unique to the T1 vs. T0 comparison, further supporting the notion that a transcriptional short‐term response of K. phaffii to hypoxic conditions indeed exists and varies from the long‐term adaptation. GO term enrichment analysis (biological process; BP) was done separately for up‐ and down‐regulated DEGs of the respective shared and unique gene sets (Figure 4B).

FIGURE 4.

(A) Venn diagram analysis of the DEGs of comparisons T1 versus T0 (5 h of hypoxia vs. normoxia), T2 versus T0 (50 h of hypoxia vs. normoxia) and T3 versus T0 (70 h of hypoxia vs. normoxia). (B) Number of up and down regulated DEGs among the shared gene set between all three comparisons as well as the DEGs unique to the three comparisons.

Main enriched GO terms for the set of genes that were upregulated in all three comparisons in response to hypoxia (245 genes) were amino acid metabolic process, carbohydrate metabolic process, generation of precursor metabolites and energy, protein folding and vitamin metabolic process. Enriched GO terms for the downregulated genes (419 genes) were the terms transmembrane transport, response to chemicals, transcription by RNA polymerase II, peroxisome organisation, cell wall biogenesis and regulation of the cell cycle. The GO terms lipid metabolism and monocarboxylic acid metabolic process were enriched in both, up‐ and downregulated gene sets. Compared with the previous transcriptomics study of K. phaffii in hypoxia using microarrays (Baumann et al. 2010), a higher number of DEGs as well as some differences in the GO enrichment analysis were found, which might be based on a better refined genome sequence and ORF annotations (Valli et al. 2016, 2020), as well as differences in the experimental, bioinformatics and statistical tools employed in this study. As observed before, a strong upregulation of glycolytic genes (PFK 1/2, PFK 300, PGK 1, GPM 1, GPM 3, ENO 1, CDC 19, PDC 1), and the oxygen‐requiring genes of the ergosterol biosynthesis pathway (ERG1/3/11/25) was occurring during hypoxia. Genes encoding the low‐affinity hexose transporters (HXT 1 and HXT 400) were induced throughout hypoxia, while the high‐affinity transporters GTH1 and HGT2 were strongly downregulated. In addition, higher transcript levels of the biosynthetic pathways leading to heme as well as thiamine were found in hypoxia, which have not been reported before in K. phaffii. Upregulation of heme biosynthesis can be easily explained as cells sense hypoxia through the inability to maintain oxygen‐dependent heme biosynthesis (Shimizu et al. 2016). Increasing the levels of enzymes that require oxygen as a substrate is a commonly applied mechanism how cells respond to hypoxia, aiming to make the best use of the limiting oxygen supply. Similar to the ERG pathway, the genes encoding the oxygen requiring steps (HEM13 and HEM14) are strongly upregulated during hypoxia, while the heme degrading HMX1 is downregulated. The strong and consistently higher expression levels of all genes involved in thiamine biosynthesis was unexpected but is most probably due to requirement for thiamine pyrophosphate (TPP) as cofactor for pyruvate decarboxylase, which was previously reported in filamentous fungi (Shimizu et al. 2016). Indeed, hypoxic upregulation of PDC1 as well as of the alcohol dehydrogenase isoform ADH900 was observed. In contrast, ADH2 was not differentially expressed, which confirms that Adh900 is responsible for ethanol formation in K. phaffii (Karaoğlan et al. 2020).

As the redox status of the cell can be determined by the GSH/GSSG ratio, the expression of the genes related to glutathione biosynthesis and glutathione‐requiring enzymes was investigated. The transcript levels of glutathione synthase enzymes, GSH1 and GSH2, were not affected by hypoxic conditions while cystathionine γ‐lyase, CYS3, and cystathionine β‐synthase, CYS4, were differentially expressed and strongly upregulated in hypoxic compared to normoxic conditions. Expression levels of genes related to glutathione‐requiring enzymes (GRX 3, GRX 5, GRX 6, GLR 1, GTT 1, DUG 1, DUG2) remained unchanged. Despite being an oxygen requiring step, the whole methanol utilisation (MUT) pathway genes (AOX1/2, DAS1/2, FLD, FBA1‐2, TAL1‐2, RKI1‐2, RPE1‐2 and the related transcriptions factors (TFs) MIT1, MXR1, TRM1 and also MIG1‐1) exhibits lower expression levels in hypoxic conditions. Alongside the MUT pathway and fatty acid β‐oxidation, most peroxisomal biogenesis genes showed lower abundance. Additionally, a clear down‐regulation of the genes involved in mating and pheromone‐response (MATa1/2, MATalpha1/2, the STE2/3 receptors, transcriptional regulators STE12 and several other genes involved in mating cascade) was observed. Notably, the general stress response does not seem to be induced, but it seems really a hypoxia‐specific stress pattern. In this respect, ROX1 encoding the TF responsible for repressing hypoxic genes in aerobic conditions is strongly down‐regulated upon hypoxia, especially in the later timepoints T2 and T3.

The Venn analysis of the T1 vs. T0 comparison showed a unique cluster of 229 genes which were upregulated after 5 h of hypoxia but not differentially expressed in the other two comparisons. This set of DEGs was enriched in genes related to amino acid metabolism, especially genes related to biosynthesis of lysine, arginine, histidine, serine threonine, methionine, proline, branched‐chain amino acids and aromatic amino acids. Furthermore, the DEGs were also enriched in genes with a role in mitochondrial organisation, in particular genes encoding mitochondrial complex IV as well as genes connected to mitochondrial translation, tRNA aminoacylation for protein translation and sterol biosynthesis. Transcription factors GCN4, MET4, CRA1 and CIN5‐1 were found upregulated in this cluster as well. The respective downregulated set of DEGs (201 genes) was enriched in genes with a role in lipid metabolic process, response to chemicals, transmembrane transport and regulation of cell cycle.

In the genome of K. phaffii, approximately 115 genes possess one or more isoforms. One interesting finding is that one third of them exhibit an opposite expression pattern upon hypoxia (e.g., one UP, the other DOWN or unregulated, Data S1). GO term analysis showed that their molecular functions were primarily related to transmembrane transport activity (33%), oxidoreductase activity (19%) and ion binding (14%). This indicates that many of the so far uncharacterized isoforms are required to accommodate for stress‐specific finetuning, as demonstrated also for other yeasts (Creamer et al. 2022; Causton et al. 2001), hence their differential expression patterns might allow to obtain some more insights into their functionality in future.

Cellular component GO enrichment analysis was conducted for the DEGs in the T1 versus T0, T3 versus T0 and T3 versus T1 comparisons, with a primary focus on the organelles where the redox potential was measured in this study (Figure 5A). This analysis demonstrated that while for the cytosol and the mitochondria the number of allocated DEGs was substantially lower in the T3/T0 versus the T1/T0 comparison, for the ER and peroxisome this number was only slightly lower and more than 25% of genes connected to the peroxisome were affected. Furthermore, while for the other organelles the split between up‐and downregulated genes was more evenly distributed, a large majority (approx. 81%) of DEGs connected to the peroxisome was actually downregulated in hypoxic conditions.

FIGURE 5.

(A) GO cellular component analysis for DEGs located in the cytoplasm, ER, mitochondria and peroxisome. The graph shows the relative percentages of up (red) and down (blue) regulated genes calculated from the total number of genes included in each component category. (B) Log₂‐fold change of genes related to the antioxidant system and ER functions in K. phaffii.

The redox potential of the cytosol and mitochondria of the non‐producing strains became more oxidised between 5 and 30 h of hypoxic cultivation, while they became reduced again at around 50 h of hypoxia. It is known that in S. cerevisiae the regulation of the redox exchange between the cytoplasm and the mitochondria but also ROS generation is strongly regulated by hypoxic response (González Siso and Cerdán 2012). GO analysis of the T1 versus T0 comparison showed enrichment of amino acid and mitochondrial‐related genes and almost twice as many upregulated genes in the cytoplasm and the mitochondria in the T3 versus T0 comparison. Hence, amino acid metabolism and mitochondria organisation could potentially explain the oxidation of these organelles under hypoxic conditions.

Interestingly, the redox potential of the ER was steadily increasing during hypoxic cultivation and remained highly oxidised until the end of the cultivation. Concordantly, the number of DEGs allocated to the ER remained high throughout hypoxia and the GO term analysis of the ER returned similar terms for the T1 versus T0 and T3 versus T0 comparisons. ERO1, encoding oxygen requiring ER oxidase, was upregulated throughout the entire hypoxic phase. In contrast, its substrate PDI1 as well as several other UPR/folding‐related genes, including the transcriptional UPR activator HAC1 and genes regulated by it (e.g., KAR2, SCJ1, ERP41, LHS1, SEC61 translocon subunits, OST/PMT complex members) were only upregulated at the T2 and T3 time points (and showed only a positive regulation trend at the T1 timepoint), indicating that they are rather part of the adaptive response to hypoxia (Figure 5B). Moreover, ROS levels in the non‐producing strains were found significantly increased after 50 h of hypoxia (Figure S7). Since the ER was the only organelle that was more oxidised after 50 h of hypoxia when compared to normoxic redox potential measurements, one could speculate that the increased levels of H₂O₂ were mostly seen due to the strong upregulation of ERO1. As a result of the adaptation response, UPR related genes were upregulated in the later stages of hypoxia, eventually affecting the redox balance of the organelle.

The redox potential of the peroxisomes became slightly more reduced throughout the first 5 h in hypoxia. The transcriptome analysis showed that most of the genes related to the peroxisome were downregulated compared to normoxic conditions. However, in the T3 versus T1 comparison, the majority of peroxisome‐related genes were actually upregulated (Figure 4A), indicating reduced down‐regulation in response to prolonged hypoxic conditions. The expression levels of CTA1 (catalase 1), the gene responsible for detoxification of H₂O₂ formed by acyl‐CoA oxidase in the peroxisomal matrix during fatty acid β‐oxidation, showed a similar regulation profile as other peroxisomal genes. Strong initial downregulation of peroxisome‐related genes could be explained by the fact that β‐oxidation is downregulated in hypoxia and, consequently, H₂O₂ generation is also reduced. However, while the cells are adapting to hypoxia, they start to express lipid metabolism and peroxisomal genes again, which might also explain the increase of the peroxisomal redox potential back to normoxic levels at the end of hypoxic cultivation phase.

In parallel to the RNA‐Seq data, proteome analysis was performed for the non‐producing strains. The sampling points from the chemostat cultivation in normoxic and hypoxic conditions were the same as transcriptome sampling points: normoxic conditions (T0), 5 h of hypoxia (T1) and 70 h of hypoxia (T3). For each timepoint, three biological replicates were collected, lysed, trypsin‐digested and analysed with Orbitrap Elite mass spectrometer. Raw data processing and protein identification took place with MaxQuant while processed data were analysed with Python 3.9. Differentially abundant proteins were identified by FC > 1.5 and adj. p < 0.05 as cut‐off and were further analysed by using GO databases. The comparisons of T1 versus T0 and T3 versus T0 were selected for data interpretation.

In total, 65 and 127 proteins were found to have a statistically significant different abundance in the T1 versus T0 and T3 versus T0 comparisons, respectively. Enriched GO terms of the category biological process for the proteins with increased abundance (30 proteins in T1 vs. T0 and 61 proteins in T3 vs. T0) showed an even distribution between the two comparisons, while for the proteins that showed decreased levels (35 proteins in T1 vs. T0 and 66 proteins in T3 vs. T0), GO terms related to nucleobase‐related small molecule metabolic process, generation of precursor metabolites and energy, monocarboxylic acid metabolic process and transcription by RNA polymerase III were higher represented in the T3 versus T0 comparison (Data S2). Glycolytic proteins such as (Hxt1, Pfk2, Fba1‐1) as well as several amino acid, heme or thiamine biosynthesis proteins showed higher levels, while PPP (Rki1‐1, Tkl1, Zwf1) and TCA cycle enzymes (Mdh3, Idp2) as well as the peroxisomal enzymes Aox1, Fox1 and Pox1 were less abundant in T1. The response was more pronounced in T3, where more proteins of the mentioned pathways showed increased or decreased abundance. For the category cellular component, the GO enrichment analysis revealed increased protein levels in the mitochondria, decreased protein levels in the peroxisomes and an even distribution between increased and decreased protein levels in the cytosol and the ER, during the T1 versus T0 comparison. However, for the T3 versus T0 comparison those proteins with increased levels were found to a higher degree to be localised in the cytoplasm, while those proteins with decreased levels were connected to a higher degree to the ER. In the mitochondria, the increased and decreased protein levels were even in T3 versus T0 comparison (Figure S8).

It is fairly noticeable that the proteome analysis resulted in a small number of differentially expressed proteins even though approximately 900–1000 proteins were quantified at each time point. A possible reason could be that the statistics used in the analysis were very strict (p‐value: 0.05 and FC > 1.5). To test this hypothesis, the transcriptome and the proteome data sets of the non‐producing K. phaffii strains were integrated by multi‐omics data analysis, namely mixOmics and multi‐omic factor analysis (MOFA). MixOmics is an R package that provides a wide range of linear multivariate methods for exploration, integration, dimension reduction and visualisation of biological data (Argelaguet et al. 2018) while MOFA is a framework for the integration of multi‐omic data sets in a completely unsupervised fashion by indicating an interpretable low‐dimensional data representation in terms of (hidden) factors. These factors are regarded as the driving sources of variation across data sets, thus identifying different cellular states (Argelaguet et al. 2018, 2020). MixOmics analysis showed a high correlation between the transcriptome and the proteome (0.99) and conditions were perfectly separated in each dataset (Figure S9A). The main separation observed along dimension 1 of the projection divided the samples between normoxic conditions (T0) and 70 h of hypoxic conditions (T3). The samples from 5 h of hypoxic conditions (T1) were located between the two and separated from them in dimension 2 (Figure S9B). From the MOFA analysis, Factor 2 separated normoxic (T0) from hypoxic conditions (T1 and T3) while Factor 4 separated the data of the 70 h of hypoxia (T3) from the other two conditions although no clear pattern could be identified (Figure S10A). Factors 1 and 3 described the omics‐layer internal spread of the data for transcriptomics and proteomics respectively, so they do not really explain the differences between the three conditions.

In general, both multi‐omics analyses showed similar gene and protein regulation patterns as the ones identified in the RNA‐Seq and proteome analysis. The mixOmics analysis showed that many of the proteins contributing the most to the separation along dimension 1 are associated to mitochondrial activity while proteins that separate T1 in dimension 2 are associated with the cytoplasmic thioredoxin reductase, the translation elongation protein and the adenosine kinase. At the transcript level, the most important gene contributing to dimension 1 was GSY2, a glycogen synthase, which is associated with T0 (normoxia). Interestingly, measurement of the intracellular storage carbohydrates in the wild‐type strains showed that the levels of glycogen in the biomass considerably decreased from 13% in normoxic conditions (T0) to 1% by the end of hypoxic conditions (T3). At the same time, GSY2 transcript levels were downregulated during hypoxic compared to normoxic conditions. Finally, both, dimension 1 and 2, showed negative correlations of the pyridoxine and thiamine metabolisms and T0 (normoxia). The MOFA analysis showed a positive correlation of T1 and proteins involved in phosphorylation and in transcriptional regulation while the proteins associated with COP1 vesicle complex and deadenylation‐dependent mRNA‐decapping factor were negatively correlated to T3. In the transcriptome analysis of factor 4, most of the transcripts were involved in the NADPH metabolism and related processes. However, on the transcript level, numerous genes significantly contributed to both factors, suggesting a polygenic effect with substantial changes in metabolic processes. More detailed results of the multi‐omics analysis with all identified genes and proteins are listed in the supplement (Figures S10B and S11A,B).

4. Discussion

In this study, the use of roGFPs for measuring organelle‐specific glutathione redox potentials was demonstrated as a valuable tool for investigating how oxygen availability affects redox balance in cells. This was accomplished by targeting roGFPs to the cytosol, ER, and, for the first time in K. phaffii, also to the mitochondria and peroxisomes. Both non‐producing and producing K. phaffii strains expressing roGFPs were cultivated in small‐scale screening cultures (24‐DWPs) and glucose‐limited chemostats. Bioreactor cultivations aimed to monitor the organelle redox potentials in vivo under industrially relevant carbon‐limited conditions, as previous studies were limited to glucose batch shake flasks. Well‐controlled and reproducible culture conditions were achieved through bioreactor cultivation, which is important for studying parameters such as oxygen availability. Under the premise that high but also low oxygenation levels might affect the redox homeostasis of the cells, chemostat cultivations were carried out under normoxic (21% O₂ inlet air), hyperoxic (30% O₂ inlet air), oxygen‐limited (14% O₂ inlet air) and hypoxic conditions (8% inlet O₂ air). Finally, insight into the transition phase from normoxic to hypoxic conditions and cell adaptation to hypoxia was gained through transcriptome and proteome analysis. The gathered data sets contribute to a better comprehension of the cellular processes underlying the observed changes in organelle‐specific redox potentials in this study.

Measurements of the redox potential of the cytosol and the ER have been done before in different redox‐engineered K. phaffii strains cultivated in shake‐flasks (Delic et al. 2010). Even though the cultivation conditions were different in this study (glucose‐limitation vs. glucose surplus), the redox potentials for the non‐producing strains only differed to a small degree from the previous data, and again the cytosol was found to be highly reduced and the ER was found to be highly oxidised. In contrast, differences were clearly seen in respect to the effect of recombinant trypsinogen production. Delic et al. (2012) found that production of trypsinogen did not have an impact on the redox potential of the ER but caused a significant reduction in the redox potential of the cytosol. In the fed‐batch mimicking small‐scale cultures in the present study, the redox potential of the cytosol remained the same as in the non‐producing strain while the ER became more oxidised compared to the non‐producing strain.

In S. cerevisiae, the redox potential of the mitochondrial matrix was shown to have a reduced environment with a redox potential around −296 mV (Hu et al. 2008) while the redox potential of the peroxisomes was found even more reduced (−319 mV) (Elbaz‐Alon et al. 2014). In K. phaffii, both organelles were found to have a reduced environment with very similar redox potential values (mitochondrial E_GSH = −274 mV; peroxisomal E_GSH = −275 mV) in screening cultures in the non‐producing and producing strains. However, in chemostats, a much more reduced redox potential was measured for the peroxisomes, while for the mitochondria the redox potential was more oxidised compared to the screening cultures. Recombinant trypsinogen production in screening cultures and chemostats under normoxic conditions did not affect the redox potentials of these organelles.

Oxygen availability has a strong impact on the energy and redox metabolism (Bettenbrock et al. 2014). Even though many studies have connected hyperoxia with elevated H₂O₂ generation and oxidative stress (Baez and Shiloach 2014), no differences in the organelle‐specific redox potentials were observed in this study during the transition from normoxic to hyperoxic conditions. This phenomenon might be attributed to ecological factors and the lifestyle of yeasts. For example, S. cerevisiae is a Crabtree‐positive yeast that, even under aerobic conditions, is fermenting. Fast glucose consumption, alcoholic fermentation and the ability to propagate at least nearly anaerobically indicate an environment characterised by the presence of ethanol and CO₂ but with minimal reliance on oxygen (Dashko et al. 2014). In contrast, K. phaffii is a Crabtree‐negative yeast since it exclusively performs respiration under fully aerobic conditions. Thus, it may have developed the ability to withstand elevated oxygen levels and exhibit greater resistance to oxidative stress.

A main finding of this study was that hypoxia had a profound effect on the glutathione redox potential of all studied cellular compartments. Differences in the response to hypoxia were also observed between the organelle‐specific redox potential of the non‐producing and producing K. phaffii strains. For the non‐producing strains, the cytosol became more oxidised after the cells were exposed to hypoxia for 5 h, while during prolonged hypoxia it returned to its more reduced redox state as in normoxia. The redox potentials of the ER gradually increased throughout the hypoxic phase. The transcriptome analysis confirmed upregulation of ERO1 early after the hypoxic shift before induction of known UPR genes as an indication of ER stress in T2 and T3. As de novo disulphide bond formation requires O₂, which is scarce during hypoxia, induction of ERO1 can be seen as a primary response. To keep Ero1 active, it needs more oxidising glutathione redox conditions. To balance Ero1 levels, other chaperones such as Pdi1 and Kar2 are induced as well, and an accumulation of ROS is observed in the producing strain. Since the ER redox state is tightly related to oxidative protein folding homeostasis, ER stress conditions can lead to ROS accumulation, resulting in oxidative stress (Cao and Kaufman 2014).

The redox potentials of the mitochondria became more oxidised upon short‐term exposure to hypoxia, but later on (after 20 h) decreased to even more reduced levels than those measured in normoxic conditions. Data analysis of the transcriptome and the proteome showed upregulation of mitochondrial complex IV and ribosomal subunit related genes after 5 h in hypoxic conditions while the expression of the cytochrome c peroxidase gene was downregulated during hypoxic conditions. This is an interesting result since the electron transport chain (ETC) complex generates ROS while cytochrome c peroxidase eliminates ROS. Changes in the composition of ETC complexes are possibly necessary for cell adaptation to hypoxic conditions while ROS generation could have a role in hypoxic signalling (Liu and Barrientos 2013; Fuhrmann and Brüne 2017). These changes could potentially explain the redox potential changes in the mitochondria although that pattern was not observed in the producing strains.

For the trypsinogen producing strains, the cytosol became more oxidised after 5 h of hypoxia and remained at this level until the end of the cultivation. The redox potential of the ER and mitochondria showed initially nearly no response to hypoxia, however, after 50 h had passed, the redox potential of both organelles suddenly increased substantially. At the same time, the specific trypsinogen production rate reached its maximum, which theoretically should mean that the oxidative protein folding machinery and therefore ER‐based ROS generation peaked as well. This phenomenon has been observed before in mammalian cells (Siegenthaler and Sevier 2019). Simultaneously, the cytosol of the producing strain remained in a more oxidised state as well, potentially hinting at a cross‐talk between the glutathione pools of the cytosol, ER and mitochondria.

Surprisingly, peroxisomes showed the opposite response from the other organelles by becoming more reduced during hypoxic cultivation and then returning to normoxic values at the end of cultivation. Interestingly, in both non‐producing and producing strains, the redox potential changes of the peroxisomes followed the exact same pattern. It has been suggested that peroxisomal redox balance regulates peroxisomal matrix protein import (He et al. 2021). Ma et al. (2013) showed that by exposing K. phaffii to H₂O₂, a clear delay in the import of PTS1‐carrying proteins occurs and identified Pex5, Pex13 and Pex14 as modifying factors for cellular oxidative stress sensitivity. In this study, all the PEX‐related genes as well as peroxisomal catalase A (CTA1) were downregulated in hypoxia compared to normoxia. These results could potentially explain why peroxisomes were more reduced during hypoxic conditions. By the end of the hypoxic phase, the peroxisomal redox potential returned to normoxic values. Transcriptome analysis comparing late (70 h) to early (5 h) hypoxia indicated upregulation of lipid metabolism and peroxisomal‐related genes. This upregulation could explain the adaptation to normoxic redox potentials values measured at the end of hypoxic cultivation. Finally, even though many studies imply a redox cross talk between mitochondria, ER and peroxisomes (Yoboue et al. 2018), in this study, the peroxisome was the only organelle that did not become more oxidised during hypoxia, indicating a tight redox regulation potentially to avoid ROS generation in cells that are under oxidative stress.

The short term decrease, followed by an increase of specific oxygen uptake rates during hypoxic cultivation deviates partly from previously published data. Adelantado et al. (2017) found no difference in q _O2 between normoxic and hypoxic conditions, whereas Carnicer et al. (2012) observed lower q _O2 in hypoxia for both wild‐type and producing strains. In contrast, Baumann et al. (2010) reported an almost two‐fold increase in q _O2 for the producing strain in fed batch. This difference is partly attributed to different severity of hypoxia in these studies (Carnicer et al. 2012) and it indicates the importance to follow the trajectory of the responses over an extended time period. Increased oxygen consumption in yeast has been linked to increased mitochondrial respiration as an adaptive response to ER stress (Knupp et al. 2018). Secretory stress is also connected with increased oxygen consumption and ATP requirements. Under hypoxic conditions, cells may experience increased metabolic demand to maintain cellular functions, increasing metabolic activity that could lead to higher oxygen uptake rates to meet the cellular energy requirements.

The transcriptome data presented in this study revealed the gene regulation of the initial transition phase from normoxic to hypoxic conditions. The transcripts of most amino acid biosynthetic genes were significantly upregulated after 5 h of hypoxia, while their expression was mostly decreased at the later sampling points. A similar response was observed in the yeast Candida albicans (Burgain et al. 2020). An exception to this trend was observed for 17 amino acid biosynthetic gene transcripts, mostly related to cysteine, glycine, methionine and lysine biosynthesis, that remained upregulated throughout the hypoxic conditions (70 h) (Data S1). In K. phaffii, alterations in amino acid pools correlated with increased glycolytic activity and reduced TCA cycle fluxes under conditions of reduced oxygen availability. Notably, these changes in a producing strain were linked to the energetic cost of amino acids, rather than the amino acid composition of the recombinant product or the cell's proteome (Carnicer et al. 2012). Furthermore, proteome analysis demonstrated comparable results to transcriptome analysis, and the high correlation between the two datasets was confirmed to agree with the literature (Rußmayer et al. 2015; Lu et al. 2007).

Gene transcripts related to mitochondrial organisation and translation were found upregulated only in early hypoxia (T1 vs. T0 comparison). Mitochondrial organisation may result from metabolic changes induced by hypoxia, facilitating the switch from respiratory to respiro‐fermentative metabolism. Mitochondrial translation could potentially be connected with amino acid starvation stress. Johnson et al. (2014) proved that amino acid starvation increases mitochondrial translation and that only the mitochondrial translation elongation factor is strictly linked with amino acid deprivation in mammalian cells. In this study, the key transcription factor GCN4 which activates amino acid biosynthesis in response to amino acid starvation in K. phaffii as well as the mitochondrial translation elongation factor TUF1 were strongly upregulated in early hypoxia. Furthermore, even though genes related to TCA cycle were downregulated during hypoxia (Baumann et al. 2010), gene transcripts encoding the enzymes mitochondrial aconitase (activated during amino acid starvation), mitochondrial NADP‐specific isocitrate and mitochondrial aspartate aminotransferase (both participate in amino acid biosynthetic process) were upregulated in early hypoxia as well. Hence, initial transition to hypoxia may have led to amino acid starvation due to oxygen depletion, resulting in upregulation of genes related to amino acid metabolism and mitochondrial translation in the early stage of hypoxic cultivation.

In conclusion, we have studied organelle specific redox states in non‐producing and producing K. phaffii in different oxygenation levels. Distinct changes in the redox potentials were observed directly after the switch from normoxic to hypoxic conditions, while they adapted either to the initial normoxic level or to a new equilibrium after prolonged cultivation in hypoxia. Remarkably, specific oxygen consumption rates adapted to the same or even higher levels in hypoxia after having dropped markedly directly after the switch from normoxia. The genome‐wide changes of the initial response and the long‐term adaptation to hypoxia have been studied at the transcriptome and proteome level, pointing at multiple layers of immediate vs. long term reaction to oxygen limitation.

Author Contributions

Aliki Kostopoulou: investigation, formal analysis, methodology, writing – original draft. Corinna Rebnegger: writing – review and editing, supervision, methodology, formal analysis, conceptualization. Borja Ferrero‐Bordera: methodology, investigation, data curation. Matthias Mattanovich: data curation, formal analysis, methodology. Sandra Maaß: supervision, methodology, data curation. Dörte Becher: data curation, methodology, supervision, funding acquisition. Brigitte Gasser: conceptualization, writing – review and editing, supervision, methodology, funding acquisition, formal analysis. Diethard Mattanovich: conceptualization, formal analysis, writing – review and editing, methodology, supervision, funding acquisition.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Data S2.

Figure S1.

Data Availability Statement

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (Perez‐Riverol et al. 2022) with the dataset identifier PXD055501. Transcriptomics data have been deposited to NCBI GEO with the accession number GSE277464.