A synthetic glycerol assimilation pathway demonstrates biochemical constraints of cellular metabolism

The engineering of synthetic metabolic routes can provide valuable lessons on the roles of different biochemical constraints in shaping pathway activity. In this study, we designed and engineered a novel glycerol assimilation pathway in Escherichia coli. While the synthetic pathway was based only on well‐characterized endogenous reactions, we were not able to establish robust growth using standard concentrations of glycerol. Long‐term evolution failed to improve growth via the pathway, indicating that this limitation was not regulatory but rather relates to fundamental aspects of cellular metabolism. We show that the activity of the synthetic pathway is fully controlled by three key physicochemical constraints: thermodynamics, kinetics and metabolite toxicity. Overcoming a thermodynamic barrier at the beginning of the pathway requires high glycerol concentrations. A kinetic barrier leads to a Monod‐like growth dependency on substrate concentration, but with a very high substrate saturation constant. Finally, the flat thermodynamic profile of the pathway enforces a pseudoequilibrium between glycerol and the reactive intermediate dihydroxyacetone, which inhibits growth when the feedstock concentration surpasses 1000 mm. Overall, this study serves to demonstrate the use of synthetic biology to elucidate key design principles of cellular metabolism.


Introduction
The structure of metabolic pathways reflects a complex interplay between different selection pressures and biochemical constraints. Even the canonical structure of central metabolism shows major variations that reflect changing constraints under different environmental conditions. To give just few examples, nonphosphorylating glycolysis represents an adaptation to very high temperatures, where the instability of phosphorylated sugars becomes a major constraint [1][2][3]; the Entner-Doudoroff pathway is preferred over (Embden-Meyerhof-Parnas) glycolysis for obligatory aerobic metabolism in which the rate of glycolysis is more important than its ATP yield [4]; the methylglyoxal bypass, converting dihydroxyacetone phosphate into the highly reactive compound methylglyoxal, is utilized when phosphate starvation constrains glycolytic flux [5,6]; and the use of pyruvate formate-lyase, supporting 50% increase in ATP yield during sugar fermentation, reflects adaptation to microaerobic and anaerobic conditions [7,8].
Yet, many of the constraints imposed on the structure and function of central metabolism are hard to derive from natural pathways, as these have specifically evolved to overcome such constraints. For example most assimilation routes start with an irreversible reaction (e.g. phosphorylation) to provide a thermodynamic driving force; hence, using natural pathways to study how a weak thermodynamic driving force limits pathway activity is challenging. Engineering and testing synthetic pathways can address this limitation and provide important lessons about key biochemical constraints that shape metabolism. Here, we demonstrate this approach by establishing a synthetic glycerol assimilation pathway and characterizing how different factors dictate its function.
Two routes are known to support glycerol assimilation [9] (Fig. 1A). In the first, glycerol is phosphorylated and oxidized to dihydroxyacetone phosphate, using quinones as electron acceptors. In the second, glycerol is first oxidized, donating its electrons to NAD + , and then phosphorylated, either by ATP or phosphoenolpyruvate, to give dihydroxyacetone phosphate. Yet, many other metabolic configurations can potentially support glycerol assimilation into central metabolism. It is not clear whether the natural pathways represent an optimal structure, and, if so, which constraints might limit the activity of alternative routes. To address this, we engineer a third alternative to the two canonical routes, based solely on known reactions (Fig. 1B): dihydroxyacetone, generated by the NAD-dependent oxidation of glycerol, is assimilated via the activity of fructose 6-phosphate aldolase (F6PA, encoded by the fsaA gene). The other reactant, glyceraldehyde 3-phosphate (GAP), is 'borrowed' by F6PA and then returned by the subsequent activity of 6-phosphofructose kinase and fructose bisphosphate aldolase (Fig. 1B). As for the natural routes, the net product is dihydroxyacetone phosphate from which all biomass building blocks can be derived.
While the enzymatic components of the synthetic pathwaywhich we term the fructose 6-phosphate aldolase pathwayare all metabolically characterized and can channel high flux by themselves, we show that the activity of the pathway as a whole is highly constrained. As long-term evolution fails to evolve high pathway activity, we deduce that this limitation is not regulatory but rather relates to fundamental aspects of cellular metabolism. Using a series of experiments, we show that the activity of the F6PA pathway is dictated by three physiochemical constraints: thermodynamics, kinetics and metabolite toxicity. The interplay between these constraints results in a unique growth phenotype, which sheds light on the biochemical logic of metabolic pathways. Specifically, a thermodynamic barrier at the pathway start necessitates high glycerol concentration to enable growth. A kinetic barrier, which stems at least partially from the thermodynamic constraint, results in a Monod curve shifted towards high concentrations, where growth rate keeps increasing up to 700 mM glycerol. Finally, the flat thermodynamic profile of the pathway results in pseudoequilibrium between glycerol and the metabolic intermediate dihydroxyacetone, such that at glycerol concentrations > 1 M, the deleterious accumulation of this reactive metabolite inhibits growth.

Results
The activity of the F6PA-pathway is limited at standard concentrations of glycerol We first tested whether the F6PA pathway can support growth on glycerol at the 'standard' (i.e. commonly used for laboratory-scale cultivation of Escherichia coli) concentration of 20 mM. We constructed an E. coli strain deleted in both glycerol kinase and dihydroxyacetone kinase (DglpK DdhaK), which was thus incapable of assimilating glycerol via either of the natural routes. We tested whether overexpression of the native glycerol dehydrogenase (encoded by gldA) and fructose-6-phosphate aldolase (encoded by fsaA) could rescue growth on glycerol (Fig. 1B). However, as shown in Fig. 2A,B, we did not observe growth.
We therefore decided to reduce the selection pressure for glycerol assimilation via F6PA. For this aim, we deleted, within the DglpK DdhaK strain, the two isozymes of fructose bisphosphatase [10], resulting in a DglpK DdhaK Dfbp DglpX strain. This strain was expected to grow on glucose, but growth on pyruvate or any other 'lower metabolism' carbon sourceshould not be possible, since gluconeogenesis is blocked at the level of fructose 1,6-bisphosphate and the cell cannot synthesize essential sugar phosphates: fructose 6-phosphate (F6P), glucose 6-phosphate, erythrose 4-phosphate and ribose 5-phosphate. Indeed, as shown in Fig. 2C, growth of the DglpK DdhaK Dfbp DglpX strain was possible on glucose but not on pyruvate.
We expected that the overexpression of gldA and fsaA in the DglpK DdhaK Dfbp DglpX strain would enable growth on pyruvate and glycerol: pyruvate metabolism provides GAP while glycerol oxidation provides dihydroxyacetone, which together can be condensed, by F6PA, to give F6P. The pentose phosphate pathway can subsequently metabolize F6P to provide all required sugar phosphates (Fig. 3A). As expected, only upon overexpression of gldA and fsaA did we observe growth on 20 mM pyruvate and 20 mM glycerol (Fig. 2D), where glycerol alone resulted in negligible growth only. Importantly, while gldA and fsaA are endogenous in E. coli, their native expression levels were too low to support growth: overexpression of gldA alone or fsaA alone did not support growth on pyruvate and glycerol.
To confirm that the growth on pyruvate and glycerol follows the metabolic pattern we predicted, we cultivated the strain on completely 13 C-labelled glycerol and unlabelled pyruvate and measured the labelling pattern in different proteinogenic amino acids. As shown in Fig. 3B, both alanine and serinewhich can be derived solely from pyruvateare negligibly labelled, whereas histidinetwo carbons of which are expected to be derived from glycerol (Fig. 3A)is almost completely doubly labelled. These results demonstrate that the combined activities of GldA and F6PA can support flux in the desired assimilatory direction. (We note the different labelling of the amino acids within a WT strain, where the labelling of serine and alanine indicates that glycerol is oxidized by glycolysis and the labelling of histidine indicates that it is mainly produced from glycerol.) Long-term evolution fails to achieve growth on low concentration of glycerol via the F6PA pathway At this point, we were wondering whether the F6PA pathway fails to support growth on glycerol as sole carbon source due to regulatory limitations. For example low induction of the glycerol channel (GlpF, upregulated by glycerol 3-phosphate [9]) or inhibition of a downstream enzyme, such as phosphofructokinase, could potentially limit glycerol assimilation. We anticipated that a long-term continuous cultivation under selective conditions would optimize the regulatory network and therefore lead to the emergence of a strain capable of growing on glycerol as sole carbon source via the F6PA pathway. Towards this aim we cultivated the DglpK DdhaK Dfbp DglpX strain, overexpressing gldA and fsaA, in a chemostat mode with limiting amounts of pyruvate (5 mM) and nonlimiting amounts of glycerol (40 mM). At the end of each week, we took a sample from the culture and tried to cultivate it on glycerol as a sole carbon source. However, even after 75 days of evolution (% 100 generations) we failed to see sustained growth on glycerol (Fig. 4). We then decided to reduce the level of pyruvate to 3 mM, and see whether the lower availability of pyruvate would assist in selecting for growth on glycerol. As shown in Fig. 4, upon reduction in pyruvate concentration, the OD within the bioreactor initially dropped, but regained its previous level after 1 month of further cultivation. Still, even after completing 120 days of chemostat cultivation (% 160 generations), we were not able to isolate a strain capable of growing on glycerol as sole carbon source. This failure prompted us to think that the problem might not be regulatory, but rather fundamental, that is, related to basic  properties of the process, and especially its thermodynamics and kinetics.
Unfavourable thermodynamics, poor kinetics and metabolite toxicity shape pathway activity Oxidation of glycerol to dihydroxyacetone, using NAD as an electron acceptor, is unfavourable, having D r G 0m~+ 20 kJÁmol À1 at physiological pH 7.5 and ionic strength of 0.25 mM (D r G 0m refers to change in Gibbs energy when all reactants are at a concentration of 1 mM) [11,12]. The following aldol condensation reaction has D r G 0m~0 , which does not help to resolve the thermodynamic bottleneck. If we assume that the concentration of GAP is~30 lM (in equilibrium with that of dihydroxyacetone phosphate,~300 lM [13]) and that [NAD]/[NADH]~10 [13], then the concentration of F6P needs to be~4 orders of magnitude lower than that of glycerol to enable assimilatory flux, i.e., achieve D r G 0 < 0. In other words, to maintain a reasonable cellular level of F6P, very high concentrations of glycerol are needed.   DdhaK strain overexpressing gldA and fsaA within a 200 mL bioreactor (overexpression of gldA alone or fsaA alone did not support growth). When we analysed the medium, we identified dihydroxyacetone which accumulated at the beginning of the fermentation and then remained rather constant throughout the growth phase. No other organic compound, besides glycerol, could be detected in the medium (Methods). We confirmed that the increase in OD corresponds to cell growth, rather than a change in the absorption property of the medium, by measuring the biomass concentration at different OD values. We wanted to confirm that growth on glycerol indeed takes place via the F6PA pathway as shown in   1B such that fructose 6-phosphate is further metabolized by 6-phosphofructokinaserather than by an alternative route (e.g. Entner-Doudoroff pathway). Towards this aim, we introduced further gene deletions to the DglpK DdhaK strain: we deleted either the two isozymes of 6-phosphofructokinase (DpfkA DpfkB) or glucose 6-phosphate dehydrogenase (Dzwf, entry point of the Entner-Doudoroff pathway). We found that while the DglpK DdhaK DpfkA DpfkB strain (overexpressing gldA and fsaA) could not grow on glycerol (regardless of the concentration), the growth of the DglpK DdhaK Dzwf strain was identical to that of the parent DglpK DdhaK strain. This indicates that F6P is indeed assimilated via its phosphorylation to fructose 1,6-bisphosphate and downstream glycolysis, rather than via the oxidative pentose phosphate pathway and/or the Entner-Doudoroff pathway.
Next, we asked how different concentrations of glycerol affect growth. While the growth rate of a WT strain remained high, above 0.4/h, at all glycerol concentrations (Fig. 6A), the growth rate of the engineered DglpK DdhaK strain (overexpressing gldA and fsaA) was strongly dependent on glycerol concentration (Fig. 6B). Similar to the classical Monod model [14], at 'low' glycerol concentrations there was a linear relationship between concentrations and growth rate (despite a small drop in growth rate at 200-300 mM glycerol). The observed Monod relationship, however, displayed a low maximal growth rate and a very high substrate saturation constant, indicating a   Discussion). At high glycerol concentrations, > 700 mM, growth started to slow and was effectively abolished at 1400 mM glycerol (Fig. 6C). We interpret this to indicate that dihydroxyacetonethe concentration of which is expected to be close to equilibrium with that of glycerolhas reached a toxic level. Indeed, dihydroxyacetone and its metabolic derivative methylglyoxal are well known to be highly reactive and to strongly inhibit cell growth [15][16][17].
To confirm that the growth collapse observed at high glycerol concentration relates to the accumulation of dihydroxyacetone, we measured the steady-state concentration of this reactive metabolite upon feeding the engineered strain with varying levels of glycerol. As shown in Fig. 7, we found that dihydroxyacetone concentrations increased with the concentration of glycerol, following the thermodynamically expected trend, which assumes a pseudoequilibrium of the two compounds (marked with a light blue line). At glycerol concentrations approaching 1000 mM, the concentration of dihydroxyacetone surpassed 10 mM. To confirm that such dihydroxyacetone concentration is indeed toxic, we added different levels of dihydroxyacetone to 500 mM glycerol (which otherwise supports growth). As shown in Fig. 8, growth of the engineered strain was abolished when the supplemented dihydroxyacetone exceeded a concentration of 10 mM. These experiments thus support our interpretation that growth at high glycerol concentrations become impossible due to the deleterious accumulation of dihydroxyacetone. We note that since the growth of a WT strain does not involve the production of dihydroxyacetone, this strain can tolerate high concentrations of glycerol.

Discussion
The synthetic F6PA pathway provides a unique opportunity to observe the interplay between different metabolic constraints, an opportunity rarely available with natural pathways evolved to overcome such barriers. Three key elements explain the observed growth via the F6PA pathway: thermodynamics, kinetics, and metabolite toxicity. As discussed above, the thermodynamic barrier of the F6PA pathway requires a high level of glycerol to support assimilatory flux while keeping the concentration of the key metabolic intermediate F6P at a physiologically relevant concentration. This explains why growth at a 'standard' concentration of 20 mM was unsuccessful without the addition of another carbon source, pyruvate, which substantially reduced the metabolic requirement of the F6PA pathway and provided GAP to thermodynamically 'push' dihydroxyacetone assimilation. We note that the native glycerol assimilation pathway in which glycerol is also initially reduced to dihydroxyacetone using NAD as an acceptor (Fig. 1A) does not suffer from the same thermodynamic barrier since the subsequent reaction, dihydroxyacetone phosphorylation, is thermodynamically and kinetically favourable. Hence, the concentration of dihydroxyacetone is kept low enough as to enable glycerol oxidation to operate far from equilibrium .
Even at high glycerol concentrations, the observed growth via the synthetic F6PA pathway was rather unusual. Specifically, the growth generally corresponds to a Monod model, but with a low maximal growth rate and a very high substrate saturation constant [14,18,19]. We argue that these poor Monod parameters can be explained by the very harsh kinetic constraints that underlie the activity of the F6PA pathway. This kinetic constraint is directly related to the thermodynamic barrier: the overall unfavourable nature of the pathway indicates that it operates at a low energetic driving force, i.e., close to equilibrium, which substantially lowers flux [20,21]. On top of this main barrier, poor enzyme kinetics further lowers pathway rate. Specifically, while alcohol dehydrogenase enzymes, such as GldA, are characterized by a high k cat in the carbonyl reduction direction, the Haldane relationshiplinking the kinetics of a reaction in both directions to its thermodynamics [22]dictates poor kinetics for glycerol oxidation. Moreover, the low affinity of F6PA towards dihydroxyacetone (K M~3 2 mM) and low cellular concentration of GAP further constrains the rate of F6P production. Finally, several of the enzymes downstream of F6P are characterized by a relatively low affinityfor example transaldolase with K M > 1 mM for F6P [23] such that their rate is expected to drop due to a low F6P concentration. The immense kinetic barrier that results from the combination of these factors accounts for the observed growth phenotype.
In contrast to the standard Monod model, however, growth on glycerol comes to an abrupt stop at high concentrations of glycerol (> 1000 mM). This is attributed to another key constraintaccumulation of a reactive intermediate, dihydroxyacetone in this case. As dihydroxyacetone lies at the centre of the thermodynamic and kinetic barriers, its concentration is close to equilibrium with that of the glycerol. This is not the case in the natural glycerol assimilation pathway in which glycerol is first oxidized to dihydroxyacetone. In the latter case, the downstream phosphorylation of dihydroxyacetone is highly favourable both thermodynamically and kinetically, and thus expected to enable much lower steady-state concentration of dihydroxyacetone.
As a further confirmation of the accumulation of dihydroxyacetone via the activity of the F6PA pathway, we noticed that the growth medium became brownish during the growth experiments, indicative of a Maillard reaction between reactive dihydroxyacetone and amines [24]. Interestingly, we observed this brownish colouring also at concentrations of glycerol which were too high to support growth, indicating that the nongrowing cells were metabolically active and produced dihydroxyacetone at high levels.
While the F6PA pathway is most definitely inferior to its natural counterparts in assimilating glycerol, it represents a unique metabolic biosensor strain. Specifically, working within the linear regime of the Monod curve enables one to quantify the concentration of the feedstock compound by measuring the growth rate. Notably, this is a special case, as the linear regime associated with most carbon sources and metabolic pathways is limited to vanishingly low feedstock concentrations, such that microbial growth cannot be easily detected. As the linear regime associated with the F6PA pathway is shifted to high feedstock concentrations that support detectable growth, the strain can be easily used to analyse glycerol concentration within the medium.
To conclude, this study serves to demonstrate the capability of synthetic biology to illustrate principle physicochemical constraints in vivo. We showed that the F6PA pathway provides a valuable lesson regarding the interplay between three key aspects of metabolismthermodynamics, kinetics, and reactivity. It is therefore evident that besides addressing societal challenges, synthetic biology offers effective tools to uncover the design principles and biochemical logic of cellular metabolism.

Strains and plasmids
All strains used in this study are listed in Table 1. E. coli strain SIJ488 [25], was used for the generation of deletion strains. SIJ488 carries the gene deletion machinery in its genome (inducible recombinase and flippase). All gene deletions were carried out by successive rounds of k-Red recombineering [25] using kanamycin cassettes (Gene Bridges, Heidelberg, Germany) with homologous extensions generated by PCR using primers listed in Table 2.

Cultivation conditions
For strain maintenance, generation of deletion strains, and for growth during cloning we used LB medium (1% NaCl, 0.5% yeast extract, 1% tryptone). For selection, antibiotics were used in following concentrations: kanamycin, 25 lgÁmL À1 ; ampicillin, 100 lgÁmL À1 ; streptomycin, 100 lgÁmL À1 . M9 minimal media were used for growth experiments (50 mM Na 2 HPO 4 , 20 mM KH 2 PO 4 , 1 mM NaCl, 20 mM NH 4 Cl, 2 mM MgSO 4 and 100 lM CaCl 2 , 134 lM EDTA, 13 lM For growth experiments over-night cultures were incubated in 4 mL M9 medium containing 10 mM glucose. Before inoculation of the experiment, cultures were harvested and washed three times in M9 medium without carbon source by centrifugation (4000 g, 3 min) to wash away residual carbon sources. For plate reader and tube experiments cultures were inoculated with a starting OD 600 of 0.01. Plate reader experiments were carried out in 96-well microtitre plates (Nunclon Delta Surface; Thermo Scientific, Waltham, MA, USA) at 37°C. Each well contained 150 lL of cell culture covered by 50 lL mineral oil (Merck KGaA, Darmstadt, Germany), to avoid evaporation. Infinite M200 pro plate reader (Tecan, M€ annedorf, Switzerland) was used for incubation, shaking and OD 600 measurements. Three cycles of 4 shaking phases, each of 1 min were run [(a) linear shaking at an amplitude of 3 mm, (b) orbital shaking at an amplitude of 3 mm, (c) linear shaking at an amplitude of 2 mm, and (d) orbital shaking at an amplitude of 2 mm]. Optical density (OD 600 nm) was measured after each round of shaking (~12.5 min). Plate reader OD measurements were converted to cuvette values according to OD cuvette = OD plate /0. 23 [27]) containing 50 mM glycerol, 5 mM sodium pyruvate and 100 lgÁmL À1 Streptomycin. The bioreactor was equipped with Rushton-type-impeller, an Lsparger for gasing, pH probe, sensor for dissolved oxygen (DO), OD metre and off-gas condenser. The temperature was set to 37°C and flow rate of air for gassing according to DO sensor to maintain 95% DO. After 2 days of cultivation in batch mode conditions were changed to chemostat. An inflow rate of fresh medium of 8.66 mLÁh À1 was set according to an estimated doubling time of 24 h. Seven days after inoculation flow rate of fresh medium was changed to 11.55 mLÁh À1 giving a doubling time of 18 h, which remained till the end of the cultivation. Seventy-five days after cultivation at a stable OD (600 nm) of 0.295, sodium pyruvate concentration was reduced from 5 to 3 mM.

Batch cultivation in bioreactor
Precultures for the bioreactor cultivations of E. coli DglpK DdhaK pZ-ASS-gldA-fsaA were prepared by thawing a glycerol stock (30% w/v glycerol) and streaking cell solution on a 29YT [28] agar plate which was incubated at 37°C for 24 h. A single colony was used to inoculate 5 mL 29YT complex medium in a glass test tube, which was incubated at 37°C on a rotary shaker at 120 r.p.m. for 7 h. The suspension was then used to inoculate 50 mL of M9 minimal medium with 10 mM glucose in a 500 mL baffled shaking flask to an optical density at 600 nm (OD 600 ) of 0.1, which was incubated at 37°C on a rotary shaker at 120 r.p.m. overnight. To inoculate the main culture in the bioreactor, cells were harvested by centrifugation (4500 g, 10 min, 4°C), the pellet was resuspended in 0.9% w/v NaCl solution and used to inoculate M9 minimal medium with 500 mM glycerol to an OD 600 of 0.2. All cultures contained 100 lg streptomycinÁmL À1 . The aerobic bioreactor cultivations of E. coli DglpK DdhaK pZ-ASS-gldA-fsaA were performed at 37°C as 200 mL cultures in glass vessels. The pH was maintained at 7.0 by online measurement using a standard pH probe (Mettler Toledo, Giessen, Germany) and addition of 12.5% NH 4 OH and 10% H 3 PO 4 . DO was measured online using a polarometric oxygen electrode (Mettler Toledo) and adjusted to ≥ 30% of saturation by stirring at 300 r.p.m. with aeration at 0.25 volume per volume per minute. Three independent cultivations were performed.

Detection of extracellular compounds
To analyse the presence of secreted products in the culture fluid of the bioreactor, 2 mL of the culture was harvested by centrifugation (12 100 g, 5 min, RT) at the given time points. The supernatant was analysed by HPLC using an Agilent 1200 Series apparatus (Agilent, Santa Clara, CA, USA) with a refractive index detector equipped with a Rezex ROA organic acid column H + (8%) column (300 9 7.8 mm, 8 lm; Phenomenex, Torrance, CA, USA), protected by a Phenomenex security guard column carbo-H (4 9 3.0 mm ID) (Phenomenex, Torrance, CA, USA) as described in Ref. [29]. Dihydroxyacetone concentrations in

Isotopic-labelling experiments
To deduce carbon source contribution to central metabolism stationary isotope tracing of proteinogenic amino acids was performed. Strains were grown in M9 with 20 mM pyruvate and 20 mM either unlabelled or completely (triply) labelled glycerol (Merck). Experiments were performed in duplicate. One millilitre of culture per OD 600 of 1 was harvested by centrifugation and washed twice in water. Protein biomass was hydrolysed by incubation at 95°C in 6 M HCl for 24 h [31]. After drying the samples under an air-stream at 95°C, they were resuspended in water and masses of amino acids were analysed with UPLC-ESI-MS as previously described [32]. Chromatography was performed with a Waters Acquity UPLC system (Waters, Eschborn, Germany), using an HSS T3 C 18 reversed phase column (100 mm 9 2.1 mm, 1.8 lm; Waters). 0.1% formic acid in H 2 O (A) and 0.1% formic acid in acetonitrile (B) were the mobile phases. Flow rate was 0.4 mLÁmin À1 and the gradient was: 0-1 min -99% A; 1-5 minlinear gradient from 99% A to 82%; 5-6 minlinear gradient from 82% A to 1% A; 6-8 minkept at 1% A; 8-8.5 minlinear gradient to 99% A; 8.5-11 minre-equilibrate. Mass spectra were acquired using an Exactive mass spectrometer (Thermo Scientific) in positive ionization mode, with a scan range of 50.0-300.0 m/z. Spectra were recorded during the first 5 min of the LC gradients. Data analysis was performed using Xcalibur (Thermo Scientific). Determination of retention times was performed by analysing amino acid standards (Merck) under the same conditions.