Establishing hot solvent quenching on an automated cultivation platform

Historically, many different quenching and extraction [32,33,34,35,36] methods have been developed but the most commonly utilized one today is cold methanol quenching [37, 38]. Even this single term encompasses a range of protocols with varying temperatures, incubation times and methanol solutions. Provided an ultra low temperature device is available, the transfer of such a method to a robotic system is theoretically feasible, albeit with some drawbacks. It would involve impractical pipetting steps with a solvent, inherit the general problem of leakage due to membrane damage during quenching [37, 39], and depend on the toxic solvent methanol.

By instead adapting hot ethanol quenching [40, 41] for an automated platform, the issue of solvent toxicity is sidestepped and additional pipetting actions prone to spilling are avoided since the method utilizes a so-called whole broth sampling approach. Here, the quenching and extraction of metabolites occurs in one step, thus prohibiting the differentiation between endo- and exo-metabolome [39] which on a positive note also renders the issue of metabolite leakage irrelevant. Besides this disadvantage, the method is quite convenient for our robotic setup as the integrated BioShake can heat up to a temperature of 99 °C, exceeding the boiling point of such alcohols as ethanol at 78 °C and isopropanol at 82.5 °C. The limiting factor regarding temperature proved to rather be the low heat transfer coefficient of the commonly used plate material polystyrene. In a pre-experiment, in which a microtiter plate with an isopropanol solution was heated on a BioShake, the liquid’s temperature plateaued at roughly 65 °C.

Due to the increased heat transfer of metals, an aluminum vial holder (Fig. 1) was designed in a 6 × 8 format to be used with 2 mL vitreous vials. Utilizing this plate and the BioShake, the temperature of isopropanol inside vitreous vials could be increased beyond 83 °C (Additional file 1: Fig. S1). Further preliminary tests were performed to investigate the effects of the choice of solvent (Additional file 1: Fig. S2) and of a prolonged incubation at high temperature (Additional file 1: Fig. S3) on the amount of extracted amino acids. In particular, ethanol and isopropanol were considered as commonly used solvents with a sufficiently low boiling point. Longer incubation periods of 20 min in the hot solvent proved to have a minimal effect on the detected amounts for most amino acids. The selection of ethanol or isopropanol did not seem to be of consequence, either, as the observed molar amounts were not consistently higher for one solvent regarding all or even most amino acids. Unlike isopropanol, however, ethanol may be given consideration as a substrate in ILE [42], in which case the addition of an ethanol solution for quenching would at the very least alter the labeling pattern of the ethanol pool. Thus, the setup with isopropanol was henceforth used as described in more detail in the “Methods” section.

Fig. 1
figure 1

Complete experimental workflow of the automated cultivation, sacrifice sampling, quenching, and sample processing. 1) A pre-heating step for the hot isopropanol quenching is initiated upon exceeding the cultivation time threshold (tthreshold), whereas the start of the sampling procedure is dependent on crossing a backscatter threshold (BSthreshold). 2) The time for heating the isopropanol solution and the residence time of cellular material inside the heated vials depend on the pipetting scheme of the particular variant of the workflow so the given values are to be understood as lower practical limits. Depending on the subsequent analyses, the extracts may be stored in a sealed deep well plate (a) or Eppendorf tubes (b).

Regarding software-wise process control, beyond the use of the vendor software for the robotic system (Freedom EVOware), every experimental run was governed by the in-house developed DigInBio control system (DCS) [43]. The DCS is connected to the laboratory network and devices, features an extensive logging framework, and executes each run within a fresh Docker container using a Python-based experiment control script (ECS). Variables such as the backscatter threshold for triggering sacrifice sampling events (Fig. 1) are defined in separate configuration files and can be adjusted even during the cultivation. In the ECS, the bletl Python package [44] is employed to parse current BioLector data for a given measurement cycle which may be plotted and sent to the practitioner via Slack, e.g., in the format of a heat map portraying the backscatter values of all wells or as line diagrams focusing specifically on selected wells and parameters.

Validating the quenching method via a spiking experiment

Performing a regular cultivation on unlabeled substrate and spiking the quenching reagent with uniformly labeled d-glucose is a reliable method to detect any residual enzyme activity during quenching and has been published previously, albeit in reverse—with fully labeled cells and unlabeled substrate [45, 46]. In the present case (Fig. 2 and Additional file 1: Fig. S4–S8), the amino acids in closest proximity to d-glucose’s entry into the metabolic network of C. glutamicuml-alanine (Ala) and l-serine (Ser)—showed a low single-digit fraction of their uniformly labeled molecular species as expressed in the M3_m2 mass traces of 1.4% and 3.5%, respectively. Separated by a few more reaction steps from pyruvate, for l-valine (Val), too, a 3.5% enrichment for its fully labeled mass trace was observed. This was expected insofar as these amino acids were the most likely to incorporate any labeled material in case of a delay in complete quenching. However, no label was detected for l-glycine (Gly) which is only one reaction removed from Ser. This particular reaction, catalyzed by the serine hydroxymethyltransferase (SHMT), is coupled in an equimolar manner to one carbon metabolism through the co-factor tetrahydrofolate (THF) and by extension to the methylene THF (MTHF) dependent l-methionine (Met) synthesis [47]. On the other end of the connection, Met belonged to the group of amino acids, which were detected with low intensity suggesting a relatively small pool size, but its most intense fragment Met133 showed no full 13C-label as well.

Fig. 2
figure 2

Combined results of the spiking experiment for validation (bar diagrams) and the isotopically instationary ILE as a proof of concept (line diagrams) accompanied by a simplified portrayal of the metabolic network of C. glutamicum ATCC 13032. The numbers in brackets pertain to each amino acid’s number of carbon atoms. In both experiments, the cultivation began on unlabeled d-glucose but in the former case, U13C d-glucose was added to the quenching reagent, and in the latter case, U13C d-glucose was pulsed before the transient sampling

Generally, there have been issues detecting some amino acids. Especially for l-tyrosine (Tyr), l-tryptophan (Trp), l-phenylalanine (Phe) and Met, detection could not be guaranteed in every sample. Intuitively, one may surmise that the amount of harvested biomass may be an issue at volume of 800 µL per well, but at the time of sampling biomass values ranged from 8.8 to 9.5 g L−1.

Another concern was constituted by the deamidation reactions of Gln to Glu and Asn to Asp which have been reported to occur even at physiological conditions [48]. Hence, one may suspect them to be accelerated under prolonged heat treatment. Since these reactions would influence the detected labeling distributions for Glu and Asp, the effect was investigated under experimental conditions but was found to be negligible (Additional file 1: Fig. S9).

Additionally, the HPLC method could not fully separate l-leucine (Leu) and l-isoleucine (Ile) resulting in a double peak at best. As the present results show, these amino acids must be omitted or at least treated with a great measure of uncertainty when further evaluating the data or using it for 13C-MFA.

Surprisingly, the more distant amino acid l-threonine (Thr) showed higher relative quantities of labeled carbon atoms than Ala and Ser. Due to the activity of the phosphotransferase system [49], a high flux towards the synthesis of pyruvate was expected and thus, a faster incorporation of label carbon into Ala than Thr. Therefore, it could be speculated that a combination of a comparatively large pyruvate pool acting as a buffer for Ala and higher flux rates of the enzymes towards Thr synthesis were responsible for these observations. Still, the label enrichment of its two measured fragments remained below 7%. Overall, more than half the measured amino acids and 26 out of 41 fragments were completely unlabeled—excluding natural isotopes. The label incorporation of a further five fragments merely amounted to less than 2% enrichment. This indicates a small bias the impact of which on reaction rates of a fluxome depends on the sensitivity of the reaction in question [21]. While this may affect the accuracy of the measurement, the precision at least can be improved by taking multiple fragments per amino acid into account.

Due to the evident remaining enzyme activity at the onset of quenching, it was hypothesized that the boiling of the solvent, which is inhibited inside the glass vials due to mounting pressure, may be critical to the procedure. Therefore, the same experiment was repeated without vitreous vials in the open wells of the aluminum plate. However, the results turned out nearly identical (Additional file 1: Fig. S10–S13) and the evaporation rates were subject to rather irreproducible position effects so the method with vials was continued.

Automated isotopically instationary ILE

For this experimental setup, the workflow had to be adapted to accommodate multiple pulsing and sampling events. The delay between the pulse and the onset of quenching had to be as short as possible in order to gain meaningful data of the relevant isotopically instationary period. Since the liquid handler preferably performed a washing step after every pipetting action to regenerate its system air gaps, said delay would have amounted to roughly 57 s meaning a loss of valuable data points particularly for amino acids in close proximity to the substrate d-glucose in the metabolic network. So, instead of harvesting a full column of six wells simultaneously, three liquid handler tips were used for the pulse and the neighboring three for an immediate sacrifice sampling event. This mode of operation decreased the minimal delay to circa 24.5 s. For the next measurements, increasingly long pauses were inserted between pulse and sampling to obtain dynamic labeling data. Performed this way, one may sample up to 16 time points per microtiter plate in biological triplicates.

The dynamic data alone is of limited use when discussing the cellular phenotype as it is influenced both by metabolic pool sizes and fluxes. For example, the slower label enrichment of Ala compared to Ser (Fig. 2) may be due to a larger Ala pool size or to a lower flux value or any combination of the two. Similarly, it seemed intuitive to expect a connection between the stoichiometric biomass requirement of an amino acid and the flux values in its synthesis pathway but the direct transfer to the labeling dynamics is not permissible as the present data shows. Instead, the requirement for Ala (YAla/CDW = 1.15 mmol g−1) is almost fivefold higher than Ser’s (YSer/CDW = 0.24 mmol g−1) [50] which is not reflected in their labeling dynamics. Nonetheless, there are still interesting conclusions to be drawn.

Despite observing a relatively fast time course of labeling for Ser, no transfer of labeled carbon atoms to Gly seemed to occur within one minute of the pulse despite both amino acids being separated merely by a single reaction (Fig. 2). This may hint at a comparatively large pool size of Gly or to a regulation via the abovementioned link to one carbon metabolism and by extension to Met. Unfortunately, Met was not detected in this experiment so this remains a matter of conjecture. It is, however, in accordance with the findings in the spiking experiment in which the mass trace pertaining to fully labeled Ser amounted to 3.5% while Gly remained unlabeled. The slower labeling dynamics of Ala relative to Ser, too, concurred with a lower labeled fraction of about 1.4%.

In the case of both Phe120 and Tyr136 (Additional file 1: Fig. S15), an increasing enrichment of numerous mass traces was detected over time hinting at a scramble of carbon atoms in the pentose phosphate pathway [51, 52]. Additionally, both showed a nearly constant signal in their respective fully labeled mass trace. Since such a fast incorporation of labeled material in less than 25 s seems unlikely judging by previous data [29] and their synthesis pathways encompass ten reactions from phosphoenolpyruvate, this might be an interfering signal which would explain the similarly unexpected observations in the spiking experiment. Even the enrichments of the fully labeled mass traces of Phe120 amounted to similar values of 18.3% in the spiking experiment and between 20 and 25% in the present data. However, only in the spectra of fully labeled Phe an unexpected signal with an m/z of 129.1025 was found aside from the expected fully labeled species of fragment Phe120 with an m/z ratio of about 128 [53].

Regarding the reconcilability of labeling dynamics of such amino acids, which are derived from others, the two nodes of l-aspartate (Asp) and l-glutamate (Glu) were of most interest (Fig. 2). For Asp, a relatively fast labeling time course was observed implying a smaller pool size compared to Ala and/or higher flux rates since the Asp synthesis required additional reaction steps. In accordance with expectations, its derived amino acids l-homoserine (Hom), Thr, and l-lysine (Lys) each showed—compliant with expectations—a less inclined variant of Asp’s labeling curve.

It was hypothesized before that “large buffer pools for transamination involving α-keto acids” [29], referring to the Glu pool in first and the l-glutamine (Gln) pool in second instance, were responsible for the delayed labeling of TCA cycle intermediates and the amino acids themselves. In the cited publication, the Glu data was not included due to measurement noise but the present data strongly supports the hypothesis. Judging by the Glu130 curve, a minor incorporation of label into Glu had barely occurred at the onset of measurements after 24.5 s but the decrease of M0_m0 seems more linear and slower in contrast to the initially parabolic one of e.g. Ser, Asp, Ala, and others (Fig. 2 and Additional file 1: Fig. S14–S18). Gln showed a similar trend, merely offset by more than two minutes. Even after five minutes, the enrichment of each Gln fragment’s unlabeled mass trace was only decreased by a few percent. Combined with the observation that the amino acids “downstream” from Glu and Gln, namely l-arginine (Arg), l-proline (Pro), and l-ornithine (Orn), remained unlabeled during the sampled time frame of 10 min, it stands to reason that the large Glu pool indeed acted as a buffer for several minutes and the Gln pool as a secondary one delaying the spread of 13C-isotopes.

In summary, the established automated workflow’s capacity to conduct ILEs has been demonstrated successfully for most observed free amino acids. Especially the aim of increasing the throughput while decreasing time and monetary investments was achieved. The automated quenching method was designed for the analysis of amino acids, but may be expanded to other metabolites such as sugar phosphates. In that case it has to be investigated whether the inactivation of enzymes occurs sufficiently fast for the expectable higher turnover rates and a faster onset of label incorporation in these metabolites [38] and whether the prolonged heat treatment decreases their concentration below the limit of detection.

Regarding the applicability of this method to 13C-metabolic flux analysis in particular, the brief window of residual activity observed during quenching may slightly alter the labeling pattern of some free amino acids, thereby possibly influencing the fluxome. In addition to labeling data, the determination of extracellular flux rates constitutes a prerequisite for MFA. Since backscatter and dissolved oxygen are measured online in the BioLector and at-line assays are well established for further analysis of transient samples [54], the specific growth rate as well as substrate and (by)-product formation rates are readily accessible via bioprocess modeling [55]. Due to the lack of an off-gas analysis in this setup, the carbon dioxide evolution rate (CER) is not known the impact of which on the sensitivity of a MFA and statistical identifiability of the network needs to be investigated. Depending on the cultivation conditions, though, even in a bioreactor the correlation between off-gas CO2 content and CER can be uncertain or misleading due to the pH-dependent equilibrium of bicarbonate and CO2 in the medium [56, 57].

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.


This article is autogenerated using RSS feeds and has not been created or edited by OA JF.

Click here for Source link (