Online Monitoring of Chip-Based Microscale Perfusion Fermentations

Chemicals

Working solutions and eluents were prepared with ultrapure water supplied by a Milli-Q IQ 7000 water system equipped with a Bio-Pak polisher cartridge from Merck Chemicals and Life Science GmbH (Vienna, Austria). LC-MS grade methanol, ethanol (EtOH), and formic acid were acquired from Sigma-Aldrich (Vienna, Austria). l-phenylalanine and trans-aconitic acid were from Sigma-Aldrich. For medium preparation, ethanol 96% was from Merck Millipore (Germany), urea and D­(+)-glucose monohydrate from Carl Roth GmbH + Co. KG (Germany), and yeast nitrogen base without amino acids and ammonium sulfate from Becton, Dickinson and Company (France). For pH control in the microbioreactor, citric acid and trisodium citrate dihydrate (Carl Roth GmbH, Germany) were used to prepare buffered medium. All chemicals used had purities greater than 99% unless otherwise stated.

Yeast Strains and Cultivation Conditions Used in This Work

All strains are based onSaccharomyces cerevisiaein-house yeast strains producing L-lactic acid by overexpression of Lactobacillus plantarum lactate dehydrogenase. As previously described by Totaro et al. a low producing strain (1a) and a high producing strain (1e) were used.

For biomass formation, YNB + E medium containing 10 g/L ethanol, 4.54 g/L urea, and 3.4 g/L yeast nitrogen base (YNB) was used. YNB + G medium containing 5 g/L ethanol, 4.54 g/L urea, 3.4 g/L YNB, and 200 g/L glucose was used for lactic acid bioconversion.

The cells cultivated on Petri dishes were transferred into 10 mL YNB + E medium and incubated at 30 °C, 180 min^–1 and a relative humidity of 70%. A 1-to-10 volume ratio was kept constant for every preculture. On the fourth day, cells were centrifuged (2,000 g, 10 min, 20 °C) and resuspended in YNB + G at a biomass concentration of OD600 = 40 for lactate production experiments. The OD was measured by a Biochrom WPACO8000 Cell Density Meter.

Cells were loaded into 20 μL hydrophilized reaction chamber Topas microchips from microfluidic ChipShop (Jena, Germany). The microchips were cleaned the day before use by triple rinsing with water and 70% (v/v) EtOH and allowed to dry. Immediately prior to use, the chambers were flushed once with 70% EtOH and then triply with YNB + G medium. Durapore membrane filters (0.45 μm, hydrophilic PVDF) from Merck Chemicals and Life Science GmbH (Vienna, Austria) were punched out to size, rinsed with 70% EtOH before use, dried, and placed in the outlets of the microchip. During inoculation, the 20 μL microchip chambers were filled with the OD40 cell suspension by pipetting. Excess YNB + G medium was pipetted into the inlet and outlet ports of the chip to remove the potential for air bubbles when placing the chip in the incubator.

Instrumentation

The primary 2DLC hardware comprised of a 1290 Infinity II 2D-LC System from Agilent Technologies (Waldbronn, Germany), equipped with one 1290 high-pressure binary pump and one 1260 quaternary pump, and a Multicolumn Thermostat set at 30 °C. The ¹D detector was removed from the system and the ¹D pump replaced with an external Harvard Apparatus Pump Controller (Holliston, MA, USA) connected to three Pump 11 Pico Plus Elite dual syringe pump modules. An Agilent Ultivo triple quadrupole mass spectrometer (Agilent Technologies, Singapore) equipped with a Jetstream ESI source was used for online ²D detection.

The first and second dimensions were interfaced by a 10-port/4-position active solvent modulation (ASM) valve (configured in countercurrent mode) connected via two 1.9 μL MP35N transfer capillaries (170 × 0.12 mm) to two 14-port/2-position deck valves carrying six stainless steel sample loops with a volume of 40 μL each (420 × 0.35 mm) or six MP35N 10 μL loops (104 × 0.35 mm). The standard pressure release kit installed between the ¹D detector and the ASM valve was removed.

A 10-port, 5-position flow-through stream selector (FSS) valve from VICI Valco Instruments with 1/32” PEEK fittings, 0.25 mm bore size, and wetted parts made of biocompatible materials (PAEK stator and Valcon E rotor) was controlled through a high-speed universal electric actuator. The actuator was connected to an Agilent Infinity 1200 Universal Interface Box II with an external contact cable (Agilent General Purpose Cable GPIO-Open End) and automatically controlled by contact closure in ChemStation (Table S10).

Two computers, one with Agilent MassHunter (Ultivo LC/TQ 1.2 (Build 1.2.23)) installed and the other with OpenLab CDS ChemStation (Rev. C.01.10 [314]) software and Agilent 1290 Infinity 2D-LC software add-on (version A.01.43) installed, were used to operate the system, control the valves, and acquire all LC-MS data.

Microchip Interface

The microchips were placed into a Lab-on-a-Chip Cell Culture Incubator from microfluidic ChipShop with a temperature control unit set to 37 °C. For the on-chip perfusion cultures, the inlet ports of the incubator were connected through PEEK tubing (1/32” outer diameter, 63.5 μm inner diameter) to 1 mL Omnifix F Luer Lock Solo plastic syringes from B. Braun (Melsungen, Germany) filled with YNB + G medium placed in the infusion pump modules. The YNB + G medium was delivered to fermentations in microchips at flow rates of 1 μL/min or less. The outlets of the incubator were connected to the FSS valve with 1/32” PEEK tubing through the directly integrated fluidic interfaces on the incubator. An additional Pico Plus Elite syringe pump module was used for washing the sample loops in between fractions collected from different fermentations with water and another for flushing the system in between analyses. A wash solution of 10 μmol/L trans-aconitate was used between independent fermentations as an internal standard to ensure the stability of the platform remained consistent throughout the duration of the fermentations.

LC-MS/MS and 2DLC Settings for Monitoring Online Microperfusion Fermentations

Separations were performed with an ACQUITY Premier HSS T3 RPLC column (2.1 × 50 mm; 1.8 μm d _p ) equipped with a VanGuard FIT column from Waters (Wexford, Ireland) at a temperature of 30 °C. The following gradient program was used with eluent A (99.9% v/v water, 0.1% v/v formic acid) and eluent B (100% v/v methanol) at a flow rate of 0.1 mL/min. See Table S1 for the gradient program. As active solvent modulation was not used, a 4 min gradient time and 4 min postgradient time resulted in a ²D-cycle time of 8 min. All 2DLC method parameters are listed in Table S4. Source conditions for MS detection are listed in Table S2 and all MS/MS transitions are listed in Table S3. A fully detailed schematic of the platform is shown in Figure S3.

Data Analysis

Data analysis was performed in Agilent MassHunter Workstation Qualitative Analysis and Quantitative Analysis (11.1) and Microsoft Excel. R Studio (2024.12.1 Build 563) was used for statistical analysis.

Enabling Compatibility of Microperfusion Flow Regime with Analytical Platform

Adapting an analytical LC-MS platform to be fully compatible with microscale fermentations posed some additional technical challenges primarily due to the low perfusion flow rates of only a few μL/min or less. Thus, a balance between the viability of microperfusion conditions and stable performance of the ESI source flow requirements of several tens of μL/min (minimum) must be sought. The required flow rates for microperfusion for microbial fermentations usingS. cerevisiaecan be readily estimated from previous work highlighting the impact of perfusion rate (P) on lactic acid production, with a maximum growth rate achieved at P = 1 h^–1 in 15 μL microbioreactors. This perfusion rate is equivalent to a flow rate of 333 nL/min for a 20 μL microbioreactor. However, working at this flow rate would greatly increase the time needed to perform an analysis on this platform, with each collected fraction requiring 24 min on the standard installed 2DLC hardware (Figure ). Accordingly, a perfusion rate of P = 3 h^–1 was selected as a compromise, reducing the time required to collect one cut to 8 min on the standard platform (ideal for online LC-MS), but still maintaining representative perfusion conditions for yeast based on previous work.

Necessary Modifications to Standard 2DLC Hardware and Operation

In conventional 2DLC operation, the transfer of the effluent from the first dimension to the second dimension occurs via the sample loops on the deck valves. There are six loops on each deck valve into which the ¹D effluent flows during a set sampling time that determines the volume of each collected fraction. In the present work, each of these cuts represents a snapshot of the perfusion fermentation integrated across the sampling time. For the analysis, the 2DLC was operated in high-resolution sampling mode in which the valve switches directly before and after each cut allowing for precise fractions to be collected. A maximum five back-to-back cuts can be taken per deck valve as the first loop is used as a bypass during flow. Successive fractions over the duration of the fermentation are collected in the sample loops and transferred (injected) onto the LC column which is serviced by the higher flow rate of the ²D pump upon a valve switch (Figure ). To reduce carryover, the valve was operated in countercurrent mode where the fraction is collected in one flow direction and then displaced from the capillary in the opposite direction to which it was collected. After each measurement, the sample loops and transfer capillary are automatically flushed by the 2DLC software while uninterrupted perfusion fermentations continue in the microchip chambers. Lastly, a syringe pump module was used to flush the path from the FSS valve to the deck valve with water (10 μL/min), removing traces of the previous sample.

Due to the increased time constraints imposed upon the system via the lower absolute flow rates needed for maintaining fermentation viability, several parameters were investigated for optimization including the filling percentage of the sample loops used to collect the fractions, the volume of the sample loops on the deck valves, and the timing of valve switches.

In contrast to traditional 2DLC, the slow microperfusion flow rates employed in this work greatly increase the time required to collect one cut. The minimum sampling time is limited by the volume of the transfer capillary installed between the ASM valve and deck valve. This time is automatically determined by the software through the 2DLC configuration settings and the ¹D pump flow rate. To this end, the quaternary pump was used as a ¹D pump solely for software compliance purposes and its flow rate set to the pump minimum of 1 μL/min with flow directed to waste.

As a minimum of 20% loop filling is recommended by the manufacturer, the standard 40 μL sample loops on the deck valves were replaced with 10 μL sample loops to reduce the time per cut (Table S9). The sampling time t _s was calculated according to eq :

where f is the fraction of loop filling, F ₁ is the ¹D flow rate, and V ₁ is the sample loop volume.

However, the volume of the installed transfer loop capillary practically limits the theoretical gains from switching to lower volume sample loops as the last cut sampled is stored in the transfer capillary (Figure B). To address this experimentally, three different sample loop filling percentages were tested: 20%, 10%, and 5%. This was achieved by lowering the ¹D pump module flow rate and collecting four cuts of a solution containing 8 μmol/L of phenylalanine. A 20% loop filling of the 10-μL sample loop was obtained by taking one cut of 2 min at a flow rate of 1 μL/min, a 10% loop filling with a flow rate of 500 nL/min, and a 5% loop filling with a flow rate of 250 nL/min. Finally, a sample loop filling percentage of 20% with 10-μL loops was chosen for all subsequent experiments as it yielded the lowest RSD (3.9%) across four cuts (Figure S5).

For standard 2DLC instrumentation, the transfer capillary between the ASM valve and deck valves represents the entirety of the path volume and corresponds to the minimum sampling time required to stop contamination from one cut to another. However, the aim in this work of assessing multiplexed fermentations required alteration of the flow path for using the syringe pump and FSS valve. The increased path volume between the FSS and ASM valve required a corresponding adjustment to the minimum sampling time applied. As the software automatically determines this value, an additional sampling delay time to allow for complete filling of the flow path was calculated and experimentally confirmed. This delay time was entered into the sampling table as the start time in the 2DLC method (Table S5).

The volumes in the system along the transfer path are much more relevant when the perfusion flow rate is ≤1 μL/min compared to the high flow rates normally used with these 2DLC valves. Considering the volume of the transfer capillary (1.92 μL, 170 × 0.12 mm) and the capillary added between the FSS and ASM valve (0.60 μL, 190 × 0.0635 mm), a theoretical total volume of 2.52 μL was expected (Table S8). However, when using a perfusion flow rate of 1 μL/min, an additional sampling delay time of 0.6 min was found to be inadequate to prevent carryover. The true path volume was determined experimentally by taking three cuts and prefilling the total path with a blank or a sample. The sampling delay time needed to be increased by 0.38 min to reduce the RSD across the peak area of the cuts and reduce the signal (peak area) seen when flushing the capillaries at the end of every measurement. This resulted in a practical total path volume of 2.90 μL, with the additional 0.38 μL likely occurring from unaccounted for volume inside the valves that is only relevant at such low flow rates (Figure S6).

Additionally, the timing of the switches of the FSS valve needed to be set so that the sampling delay time allows the cuts to be preloaded into the capillaries while other cuts are being collected (Figure S7). The total analysis time for one collection and corresponding measurement is the sum of the analysis times of each cut and the time needed to flush the capillaries in the decks (which is automatically programmed by the software and equal to the time of one ²D-cycle), plus the sampling delay time. The resulting timing tables are shown in Tables S6 and S7.

Despite taking these steps, the carryover between independent perfusion fermentations was not eliminated when setting the sampling delay time to 0.98 min (with the automatic delay of 1.92 min implemented by the software). This contamination cannot be analytically corrected for without increasing the complexity of the system and thus decreasing robustness, so a washing step was incorporated between independent fermentations to flush the path between the FSS and deck valve. As the wash flow rate must match the ¹D flow rate of the first pump module to avoid washing the samples out of the sampling loops, a third pump module operating at 1 μL/min was used for this purpose. This allows fractions from microbioreactors operating in parallel to be infused at low perfusion flows, collected in sampling loops, transferred to the LC-MS/MS for measurement with washing steps to preventing cross-contamination. It is noted that stepping backward with the FSS valve is not recommended as the transition through the different streams will introduce sources of contamination. Thus, for use with the currently employed FSS valve, five cuts are collected per time point as follows: two cuts of the first fermentation, the wash, and then two cuts of the second fermentation (see Figure S8).

Application of Platform to Multiplexed Monitoring of Lactate-Producing Yeast Strains

To demonstrate the capabilities of the full system for its intended use, both single and multiplexed microscale fermentations of lactic-acid-producing strains ofS. cerevisiaewere performed in 20 μL bioreactors focusing on online monitoring of the product and other metabolites. For all experiments, the engineered lactic-acid-producing strains were cultivated in the microfluidic device, where a continuous 1 μL/min flow of cell medium was pumped through the 20 μL cultivation chambers within which cells were retained by placing a 0.45 μm PVDF filter in the outlet ports before inoculation (Figure S1). The addition of a filter was found to be necessary for ensuring cell retention within the microchip chambers during perfusion operation (Figure S2).

To demonstrate the long-term stability of the system at time scales of relevance for fermentation biotechnology (i.e., up to 24 h), the production of lactate was initially monitored in a single fermentation of the high-producing strain 1e. Lactate was detected in all successive cuts over a 5 h fermentation and revealed that the production quickly stabilized after approximately 2 h of monitoring with the peak area of lactate remaining consistent thereafter with 4.0% RSD determined across three cuts at the 5 h mark (Figure S8A). Furthermore, lactate could still be reliably detected after 24 h of continuous operation without any disruptions to the system (e.g., blockages), demonstrating the stability of the entire system for typical fermentation time scales in biotechnology. With the FSS valve used in the current setup, such experimental comparisons are possible for multiplexed monitoring of a maximum of five different flowing streams. With one flow-through stream dedicated to the postmeasurement wash flow, and one other stream for intermediate washing between independent fermentations, a total of three perfusion experiments can be monitored. The stable performance observed in single-fermentation assessments could also be demonstrated for monitoring of lactate in two multiplexed fermentations performed in parallel chambers on a single microchip.

To further demonstrate the capability of the platform for measuring additional extracellular metabolites covered by analytical LC-MS/MS, the panel of monitored metabolites was expanded to include phenylalanine, isoleucine/leucine, tyrosine, and methionine (Figure S10). These cellular metabolites are routinely quantified for yeast biotechnology metabolomics with RPLC-MS and are relevant for the strains of lactate-producing yeast considered in this work as indicators of bioprocess stability under perfusion growth conditions. The peak area repeatability of the entire panel was assessed for biological replicates. Initially, two biological replicates of the 1e strain were monitored concurrently across an 8 h fermentation time. The measured lactate peak area remained within expected biological variation after allowing 2 h for stabilization, with 9.1% to 11% RSD determined across four cuts per time point.

Additionally, these metabolites were also monitored in multiplexed fermentations with the 1e and 1a strain over 5 h. As expected, the fermentations quickly reached a ready state, with greater production of lactate in the fermentation with the 1e strain than with the 1a strain observed (Figure ). After initial rapid changes in metabolite abundances in the first 2 h, the abundances of product (lactate) and amino acids reach steady levels that are maintained across several hours of lactate production, which is in good agreement with previous results. An RSD of 2.9% for the internal standard (trans-aconitate) peak areas from the wash sampling cuts between the independent fermentations indicated that the platform remained stable during the entire 5 h fermentation, which is critical for robust comparison of different strains assessed within a single experiment. All other metabolites were found to be detected in similar concentrations across both strains and with much higher concentrations than present in the cell media background (Figure S9).

Finally, to test the potential of this platform as an effective tool for rapid screening of different fermentation conditions and strains, two different media conditions were examined for growth ofS. cerevisiaestrains. The standard medium used has an initial pH of 4.3, but as lactic acid accumulates in the culture medium, the pH lowers, decreasing cell metabolism and the production of lactic acid by the cells. Citrate-buffered media was prepared at pH 3 and 5, to monitor the differences in production under a harsh and beneficial environment. The buffering capacity of the media during the fermentation was monitored via pH test strips by testing the flow-through effluent from the FSS valve outlet ports. Five cuts were collected and analyzed online over 8 h fermentations for the four conditions assessed (i.e., strains 1e and 1a with fermentation media buffered at pH 3 and pH 5). As expected, since a pH 5 environment is less stressful to the cells than pH 3, both strains struggled more at the lower pH and produced significantly less lactate at pH 3 than at pH 5 (p < 0.0001, Figure ). These results are in-line with previous results reported withS. cerevisiaemicrobioreactor fermentations and demonstrate that the new platform can monitor different strains and bioprocess conditions at relevant time scales and with minimal consumption of biomass and reagents.