Scale‐up of a Type I secretion system in E. coli using a defined mineral medium

Secretion of heterologous proteins into the culture supernatant in laboratory strains of Escherichia coli is possible by utilizing a Type I secretion system (T1SS). One prominent example for a T1SS is based on the hemolysin A toxin. With this system, heterologous protein secretion has already been achieved. However, no cultivations in a defined mineral medium and in stirred tank bioreactors have been described in literature up to now, hampering the broad applicability of the system. In this study, a mineral medium was developed for cultivation under defined conditions. With this medium, the full potential and advantage of a secretion system in E. coli (low secretion of host proteins, no contamination with proteins from complex media compounds) can now be exploited. Additionally, quantification of the protein amount in the supernatant was demonstrated by application of the Bradford assay. In this work, host cell behavior was described in small scale by online monitoring of the oxygen transfer rate. Scalability was demonstrated by stirred tank fermentation yielding 540 mg/L HlyA1 in the supernatant. This work enhances the applicability of a protein secretion system in E. coli and paves the way for an industrial application.


| INTRODUCTION
The secretion of proteins into the culture medium or the periplasmic space offers several advantages compared to a classic intracellular production. The protein can be obtained in correctly folded form and the purification is simplified compared to purification from lysed cells or inclusion bodies. 1 Different strategies can be distinguished to achieve the extracellular localization of a protein using gram-negative bacteria: Active transport into the periplasm, followed by passive transport across a leaky or permeabilized outer membrane is one approach. 2 The increasing permeability of the outer membrane can be realized by different strategies, for example, the use of so-called "leaky mutants," which have mutations in the membrane or cell wall structure. 2 It might also be obtained by co-expression of the kil gene that causes the release of periplasmic proteins. 3 Besides passive transport across the outer membrane, the use of a secretion system can directly target the protein of interest into the culture supernatant, thereby avoiding the necessity for outer membrane disruption, which might result in a decreased cell viability. 2 Among the different secretion systems present in bacteria 4 the Type I secretion system (T1SS) is common in a large number of gramnegative bacteria (for reviews covering different aspects of the system see Delepelaire 5 and Holland et al. 6 ). The secretion system of the hemolysin A (HlyA) toxin in uropathogenic Escherichia coli belongs to this class of secretion systems and is one of the most well studied systems. 7,8 Secretion takes place in an unfolded state 9,10 in one step across both membranes. [11][12][13] Folding into the biologically active conformation is induced in the extracellular space through the binding of calcium ions. 14 The transport complex itself consists of three membrane proteins 15,16 : A membrane fusion protein, HlyD, and an ABCtransporter, HlyB, are both located in the inner membrane while an outer membrane protein, TolC, is located in the outer membrane. 17 A so-called secretion signal HlyA1, consisting of the C-terminal 218 amino acids of the full-length toxin, contains all information necessary for secretion and is secreted as efficient as the full-length toxin. 18 Consequently, HlyA1 has to be fused to a target protein to achieve its secretion. 9 Most of the studies addressing the HlyA secretion system focus on examining the functionality of the transport complex and a proof-ofconcept for heterologous protein secretion. [19][20][21] Consequently, most of the experiments in this context have been carried out in complex medium 9,20-25 using shake flasks and verification of the product and its concentration are usually determined by sodium dodecylsulfatepolyacrylamide gel electrophoresis (SDS-PAGE). However, utilization of a mineral medium is crucial to facilitate process development. First, the valuable advantage of rather "clean" supernatants of E. coli with low impurity and foreign protein content would be exploited, because no foreign proteins and amino acids (originating, e.g., from yeast extract) are present in a mineral medium. Second, product purification and, thus, downstream processing, which is a major cost determinant, 26 would be simplified. In addition, foam formation during fermentation in a bubbleaerated system (stirred tank reactor) is reduced, if less foreign proteins and peptides are present in the medium. 27 Utilization of a mineral medium enables the cultivation under defined conditions, thereby avoiding lot-to-lot deviations of complex medium components, which might influence the heterologous protein yield. 28 Furthermore, quantification of the product (e.g., Bradford measurement) would be possible, which is important for bioprocess development and optimization.
One important aspect associated with recombinant protein production is the so-called metabolic burden (also referred to as "metabolic load" or "metabolic drain") that arises from introducing foreign DNA into a host cell. 29 It is defined as the draining of metabolic resources (building blocks such as amino acids and ATP) from the host cell for the production of a recombinant protein. 29,30 By applying the respiration activity monitoring system (RAMOS), the metabolic state of a culture can be assessed. 31,32 The system measures the respiration activity in shake flasks using oxygen partial pressure sensors in the gas headspace to calculate the oxygen transfer rate (OTR). Differential pressure sensors are used to calculate the respiratory quotient and the carbon dioxide transfer rate. 31,32 RAMOS can be used to evaluate the metabolic burden of a host cell. 33 Previous successes in biochemical characterization, improvement of titers, and engineering of the HlyA secretion system have layed the foundation for an industrial application of the system. This study focuses on the scale-up of a fermentation process for the HlyA1 secretion system in E. coli utilizing a defined mineral medium. First, cultivation in a suitable mineral medium is established. Afterward, a first improvement of HlyA1 titers is carried out in shake flasks by online monitoring of the OTR. This way, the metabolic burden of the host cell during secretion is evaluated.
Next, scale-up from shake flask to stirred tank is performed. Bradford assay is established for product quantification.

| MATERIALS AND METHODS
2.1 | Microorganisms, expression vectors, and target protein E. coli Tuner strains were purchased from Novagen and transformed with the plasmids depicted in Figure 1. The secretion signal HlyA1 (~23 kDa) is used as a model product in this study and coded on an arabinose-inducible plasmid ("pSOI") as described by Bakkes et al. 9 ("product") ( Figure 1, left). The transport complex proteins HlyB and HlyD are coded on a second plasmid with pK184 backbone as described by Bakkes et al. 9 ("transporter") ( Figure 1, right). TolC is endogenously expressed in E. coli 34 with an estimated copy number of 1,500 per cell 35 and thus, not additionally expressed on a plasmid.

| Cultivation media and cultivation conditions
All cultivation media were supplemented with the appropriate antibiotics to ensure plasmid stability. The antibiotics were added from sterile filtered stock solutions. 100 μg/ml of ampicillin and 30 μg/ml of kanamycin were used as selection markers for the product plasmid and the transporter plasmid, respectively. All cultivations in shake flasks were performed in 250 ml shake flasks with a filling volume of 10 ml. An incubator shaker (model ISFX-1), purchased from Kuhner AG, Switzerland, with a shaking frequency of 350 rpm, and a shaking diameter of 50 mm was used for all experiments.

| Complex cultivation media
The cultivation in complex medium was carried out in 2xYT-medium containing 16 g/L tryptone, 10 g/L yeast extract, 5 g/L NaCl, and 10 mM calcium (supplemented as CaCl 2 • 2 H 2 O). The pH was adjusted to 7.0 ± 0.2 using 5 M sodium hydroxide solution. Afterward, the medium was sterilized at 121 C for 20 min.

| Mineral cultivation media
The cultivation in mineral medium was carried out in modified Wilms-MOPS mineral medium according to Wilms et al. 36 The medium consisted of 7.5 g/L glycerol, 6.98 g/L (NH 4 ) 2 SO 4 , 3 g/L K 2 HPO 4 , 2 g/L Na 2 SO 4 , 41.

| Cultivation with online monitoring and offline sampling
The first pre-culture was prepared in 2xYT-medium (see above). Ten milliliters of medium were inoculated with 100 μl from a cryo culture (OD 600 nm, Start = 0.035) that had been stored at −80 C. The preculture was incubated for 16 hr (for inoculation of the main culture in complex medium) or for 3-4 hr (for inoculation of the second preculture in mineral medium) at 37 C.
For experiments in complex medium, the main culture in 2xYTmedium was inoculated from the first pre-culture with an initial OD 600 nm of 0.1. For experiments in mineral medium, the second preculture in Wilms-MOPS-medium was inoculated from the first pre-culture with an initial OD 600 nm 0.01 and cultivated at 30 C for 16 hr. The main culture in Wilms-MOPS-medium was inoculated from the second pre-culture with an initial OD 600nm of 0.1. For the inoculation of the main culture in stirred tank fermentation, two flasks of the second pre-culture were run in parallel and pooled prior to measurement of the optical density.
The OTR of the main culture (complex or mineral) was monitored by applying the RAMOS built in-house. For monitoring of the OTR, modified 250 ml shake flasks with a filling volume of 10 ml were used. Measurement and calculation of the OTR was carried out as described by Anderlei et al. 31,32 In brief, the partial pressure of oxygen in the headspace is monitored by an oxygen sensor. The oxygen partial pressure in the head space decreases due to respiration of the microorganisms, because the flask is closed by valves during the measurement phase.
Afterward, the flask is flushed with air until the next measurement phase begins. From the decrease of the oxygen partial pressure, the OTR is calculated. Due to the very low changes in dissolved oxygen during batch cultivation, the OTR can be assumed to be equal to the oxygen uptake rate. 37 For parallel offline sampling, conventional Erlenmeyer flasks, equipped with a cotton plug, were cultivated at the same conditions as the RAMOS-flasks. At each sampling point, the whole volume of one flask was used for offline determination of the parameters described under "Offline analysis."

| Fermentation in stirred tank reactor
For the fermentation in a stirred tank reactor, the pre-culture and media preparation were carried out as described in Sections 2.4-2.6. F I G U R E 1 Schematic overview of the plasmid system used in this study. Left: Product plasmid (pSOI) with ampicillin resistance gene (amp r ) and secretion signal HlyA1 as model product under control of the L-arabinose operon (araC and P BAD ). Right: Transporter plasmid (pK184) with kanamycin resistance gene (kan r ) and transport complex (HlyB and HlyD) under control of the lac promotor. Promotors are inducible with arabinose and IPTG, as indicated by the arrows. IPTG, isopropyl-β-D-thiogalactoside 2.7 | Offline analysis

| Optical density
The optical density was determined with a spectrophotometer (Genesys 20, Thermo Fisher Scientific, Waltham, MA) at a wavelength of 600 nm in 1 cm cuvettes. The sample was appropriately diluted using 0.9% NaCl (mineral medium) or 2xYT-medium (complex medium). NaCl and 2xYT-medium in the respective dilution were used as blanks. The measurement was carried out in duplicates.

| pH-value
The pH-value was determined with an InLab Easy pH electrode (Mettler Toledo, Germany) that was calibrated using the appropriate calibration buffers and a CyberScan pH 510 meter (Eutech Instruments, Thermo Scientific, Germany). The gels were scanned using a scanner (Epson Perfection V700 Photo) equipped with the appropriate software (Epson, Japan) and standard settings for highlight (246), shadow (18), and gamma (1.00) levels.

| Sodium dodecylsulfate polyacrylamide gel electrophoresis
Whenever possible and as stated above, the same sample amount and the same scanning options were chosen for all gels to enable a qualitative comparison of the gels and to avoid distortion of the results.

| Bradford assay
The protein concentration in the supernatant was determined in triplicates by Bradford assay 38 using bovine serum albumin (Lot-No. BAH65-882, VWR) as a standard for calibration. The assay was carried out as suggested by the manufacturer (Thermo Fisher Scientific) in 96-well micro titer plates using Coomassie protein assay reagent (Merck, Germany). The values were corrected for evaporation (in shake flasks) and the value for pure medium was subtracted.

| Carbon sources
The concentration of the different carbon sources and media ingredients (glycerol, arabinose, and citrate) was determined via HPLC analysis. An organic acid resin column (250 × 8 mm, CS Chromatographie Services, Germany) was used for separation at a temperature of 80 C and a flow rate of 0.8 ml/min. As a mobile phase, 5 mM H 2 SO 4 was used.

| Protein secretion in complex medium
Since for many applications a tightly controlled and well inducible protein expression is desired, an arabinose-inducible product plasmid (p BAD -promotor) was investigated. The transporter (HlyB and HlyD) remained on a different plasmid under control of a lac promotor (see Figure 1). Both promotors are subject to "all-or-none behavior." 39,40 This means, that a fraction of cells is fully induced, while another fraction might not be induced at all. Therefore, E. coli Tuner was used as host strain. In contrast to the commonly used BL21(DE3) strain, the Tuner strain carries a mutation in the lac-permease, 41  though the supernatant appears to be very "clean" in both cases, it should be noted that complex media contain many small peptides and amino acids originating from yeast extract. Since the average size of these compounds is well below 10 kDa 43 they do not appear on the SDS-gel. The experiment was replicated (see Figure A1). With the constraint of a deviation in the absolute values of the OTR and differences in the length of the lag-phase, the results are well reproducible, especially with regard to the secreted amount and localization of HlyA1.

| Mineral medium for protein secretion in E. coli
In a next step, a defined mineral medium was adapted for the cultivation with this secretion system. To the author's knowledge, no cultivations have been performed in defined mineral medium with this secretion system, yet. A reason for this might be that previous studies focused on the biochemical characterization of the system (see Section 1) with a minor focus on a future industrial application.
Another reason might be that high calcium concentrations are required for product stabilization, which tend to precipitate as calcium phosphate or calcium sulfate for example.
The mineral medium is based on a medium developed by Wilms et al. 36 for the cultivation of E. coli and was adapted as mentioned in Section 2.4. Glycerol was chosen as a carbon source that, in contrast to glucose, neither inhibits the transcription from the lac-nor the araBAD-promoter. Since 10 mM calcium were added in complex medium, the same concentration was chosen for the mineral medium. The reason for this observation (more or less constant OTR after induction) remains unclear and was observed for a single amino-acid exchange, 33 different host strains expressing the same product 45 as well as a single nucleotide exchange. 44 In all cases, biomass formation declined upon induction, which is in agreement with data from literature and a result of a higher ATP demand. 46 The increased ATP demand is, however, not fulfilled by means of oxidative phosphorylation as the oxygen demand is constant after induction. Thus, another explanation might be that the cells use substrate level phosphorylation to supply ATP instead of oxidative phosphorylation. As the metabolic burden response affects a broad range of cellular functions, a combination of different regulatory mechanisms might also be responsible for this observation. Overall, it is rather difficult to give one overall explanation for the observed phenomenon without a thorough biochemical analysis.
This was, however, not part of this study, but should be investigated in more detail in the future.
As soon as arabinose is depleted, a distinct kink can be identified in the OTR (Figure 3a, Phase III). Most likely it appears due to changes in the cell metabolism (diauxic rearrangement of metabolism). 33 Afterward, cell growth continues, as indicated by an exponential increase in optical density and OTR. The remaining glycerol is consumed in a sec-  Figure A6. IPTG, isopropyl-β-D-thiogalactoside; SDS-PAGE, sodium dodecylsulfate-polyacrylamide gel electrophoresis be affected by many parameters like ammonium concentration in the medium, pH-value (see Kramer et al. 47 for a comprehensive review) and exposure to oxygen radicals. 48 The process of protein precipitation is still poorly understood. Since the reason for the instability of the product in shake flasks remains unclear, it is subject of further investigations. Furthermore, a degradation band appears after around 11 hr with a molecular weight of~20 kDa (Figure 3b). Since the induction phase ends after about 9.5 hr at the end of Phase III (Figure 3a Figure A3b). This is also reflected in the respiration profile (no constant OTR after addition of arabinose) ( Figure A3a). In contrast to arabinose, the addition of IPTG itself does not place a metabolic load  Figure A8a,b. IPTG, isopropyl-β-D-thiogalactoside; SDS-PAGE, sodium dodecylsulfate-polyacrylamide gel electrophoresis on the cells (Figure 3a, closed triangles and closed diamonds). From an industrial point of view, omitting IPTG for induction is an advantage, which reduces costs. However, to enable comparison with previous studies and data presented within this work, all other experiments were performed with addition of IPTG. In future work, experiments will be performed without the initial addition of IPTG for a comparison with the ones with IPTG.

| Variation of the time of induction
Time of induction and inducer concentration is known to have a high impact on recombinant protein expression. 49,50 Since secretion of . This is also represented by a higher OTR. Consequently, the induction phase is shorter (more cells are consuming the arabinose). Thus, the non-induced second growth phase is less distinct, because more glycerol had been consumed before the induction was performed and less glycerol remained for the second non-induced growth phase. Therefore, the overall cultivation time is shorter for the later induced culture. Nevertheless, the same phases that have already been identified and described in Figure 3 can be assigned to this culture. The course of the OTR is well reproducible ( Figure A4). Figure 4b shows the SDS-gel at the point where arabinose was depleted (9 hr). While there is again some protein present in the cell extract (closed square and triangle on the right), there clearly is a higher amount of product secreted when the induction was performed after 5 hr (closed square and triangle on the left). Measurement by Bradford assay shows that the protein concentration was increased from 190 (±11) mg/L to 621 (±67) mg/L. This corresponds to a more than threefold increase, which is also reflected (qualitatively) in the SDS-PAGE ( Figure 4b). In addition, a sample was also taken from a RAMOS flask for the culture induced at an OTR of 17 mmol/L/hr. It demonstrated that the amount of HlyA1 in the supernatant is reproducible (Table S1).
For both cultures presented in Figure 4, a slight degradation band is visible below the target protein band. Depending on the target application of the product, this might necessitate further purification (if the product is, e.g., intended for medical applications) or is tolerable (if the product is, e.g., a technical enzyme). However, compared to an intracellularly produced protein, which necessitates opening of the cell envelope and releasing all host proteins, the purification process is much simplified.

| Scale-up into stirred tank bioreactor
A protein that is secreted into the supernatant might experience a different environment in a stirred tank reactor, in comparison to a shake flask.
Due to different aeration (surface aeration vs. bubble aeration) and agitation (shaking vs. stirring), scale-up from shake flask to stirred tank reactor is a crucial step for process development. It is also a prerequisite for an industrial application. To prove that the production of the secretion signal and model product HlyA1 is also possible in a stirred tank bioreactor, a scale-up was performed using the adapted mineral medium. The pH and temperature were kept constant and the DOT was kept above Both aspects are especially important in an industrial production process.
In contrast to the cultivation in shake flasks (Figure 3), no protein accumulates in the cell extract during cultivation in stirred tank. Only after arabinose is already depleted, a slight decrease of HlyA1 is observed in the supernatant (Figure 5a, open circles). Since the course of the OTR is similar in shake flasks and stirred tank, the respiration activity is not affected by the different localization of the product. This supports the hypothesis that HlyA1 was secreted, but subsequently precipitated from the supernatant (see Section 3.2). The reason for the partly different localization in shake flask and stirred tank is unclear and subject to further investigations. Besides differences in aeration and agitation (see above), it might be caused by differences in hydromechanical stress. 51 The full potential of the system will now be investigated by secretion of different products. Furthermore, additional plasmid systems (e.g., T7-system) will be tested. A mathematical method was recently developed to predict the expression performance of different E. coli strains cultured under different conditions, based on the course of the scattered light signal over time. 53 This approach might in the future be used to evaluate the secretion performance of different products (e.g., enzymes) in the presented secretion system.

ACKNOWLEDGMENTS
The scientific activities of the Bioeconomy Science Center were financially supported by the Ministry of Culture and Science within the framework of the NRW Strategieprojekt BioSC (No. 313/323-400-002 13).