Citric acid fermentation occurs as part of a diauxic growth response

To investigate citric acid production by the parent citric acid-producing ATCC 1015 strain, empirical time-course data were obtained from fermentation performed in shake flasks. Biomass and citric acid production were monitored with samples taken at 24-h time-points. Diauxic growth behaviour was observed, with a drop in growth rate at day 3 (Fig. 1a). Citric acid production commenced at day 3, coinciding with the diauxic growth shift (Fig. 1b). 60 g/L citric acid was produced.

In order to better understand the basis of this growth behaviour, we developed a dynamic flux balance analysis (dFBA) model based on the previously published FBA model [13, 14]. To validate the model and further investigate the diauxic growth behaviour, empirical data were obtained for citric acid fermentation under a range of phosphate levels (0.05, 0.09, and 0.17 g/L). Samples of the cultures were taken every 24 h to produce a time-course of biomass dry weight, phosphate depletion, citric acid production, and glucose consumption (Fig. 2). Phosphate was rapidly taken up and depleted by day 2 (Fig. 2b), yet growth continued (Fig. 2a). Phosphate was therefore clearly stored internally to enable growth during the absence of external phosphate. Diauxic growth was observed, with growth becoming phosphate-limited. The diauxic growth shift was synchronous with depletion of external phosphate. The phosphate-limited growth rate was a function of the initial phosphate concentration, with increased growth rate at higher phosphate. The timing of citric acid production was observed to coincide with the onset of phosphate-limited growth and external phosphate depletion (Fig. 2c). Up to 50 g/L citric acid was produced, with the culture at 0.17 g/L phosphate producing the most. Glucose uptake was relatively slow for the lower phosphate cultures and a limiting factor in citric acid production (Fig. 2d).

From these observations, we hypothesised that the diauxic growth shift is caused by a switch to phosphate-limited growth, resulting in citric acid production. This hypothesis was motivated by examination of our data, existing knowledge of A. niger [10] and also the ecological evidence that organic acids are released extracellularly in order to facilitate the mobilisation of phosphate, especially in soil [15]. We decided to examine the plausibility of this hypothesis using dFBA modelling.

Simulating citric acid fermentation by dynamic flux balance analysis

To create time-course simulations comparable to the citric acid fermentation empirical data, dynamic flux balance analysis (dFBA) was used with the iMA871 metabolic model [13]. Citric acid production was modelled by incorporating kinetic acid-dissociation reactions into the dFBA schema for the organic acids in iMA871 and setting the objective to proton production. This explicit inclusion leads to an acid hierarchy [14], which suggested that citric acid production was the most efficient means of acidification with oxalic acid production switched off.

In the standard setting for the metabolic model, citric acid secretion is included as a part of the external constraints during growth [13]; however, this is not supported by our observations. Therefore, a novel modelling approach was designed to simulate the diauxic growth behaviour with citric acid production commencing upon a diauxic growth shift coupled to phosphate intake. To achieve this, a double optimisation dFBA setup was designed (Fig. 3). The objective is first set to biomass production, with the maximised growth rate then used in the second optimisation. The second objective is dependent on the growth-limiting condition of the first optimisation. The decision process uses a boolean expression. If the external phosphate flux is lower than its flux constraint, the second objective is set to phosphate storage to store excess phosphate not used for growth. Otherwise, external phosphate flux is equal to its flux constraint and the second objective is set to proton production to make use of the carbon not used for growth (phosphate-limited growth).

The dynamic modelling approach, dFBA, therefore includes a number of metabolite pools that are tracked outside of the FBA, including external glucose, external phosphate, external pH, organic acids as well as the hypothesised stored phosphate. These metabolite pools are linked to the FBA simulations at each step via first order differential equations describing transport processes. These differential equations are solved at each time-step to provide flux constraints for the FBA optimisations occurring in a tandem fashion and assuming the metabolic system remains at a steady-state despite the small changes in the external constraints. All equations used are detailed in the methods, but are essentially either linear diffusion or Michaelis–Menten transport equations across the membrane as described below and mathematically in the methods section. Literature sources were used to parameterise the model wherever available as described below.

Phosphate uptake and release of stored phosphate were modelled according to Michaelis–Menten kinetics. As no characterised phosphate transporters could be found for A. niger in the literature, kinetic parameters were fitted to empirical data on phosphate uptake (Table 2).

Fitting of model parameters to empirical data and model validation

The empirical data obtained from the experiment varying phosphate were used to fit model parameters and validate the model. A total of eight parameters were fitted to a data-set containing 84 data-points. As each data-point was in quadruplicate with very low error margins, we decided to use the data-set for both model training and validation. The trained model was later applied to independent data-sets (Fig. 4), which gave further validation. Using the trained model, citric acid fermentation was simulated for each of the phosphate levels tested, and model predictions plotted alongside empirical data (Fig. 2). The modelled diauxic growth behaviour gave good fits to empirical data, with external phosphate depletion being the trigger that results in phosphate-limited growth and citric acid production. All the model outputs showed a strong qualitative comparison to the empirical data with unfitted parameters taken directly from the literature. Notably, the modelled glucose uptake fitted empirical data closely (Fig. 2d) with unadjusted literature values for transport-mediated uptake rate and affinity.

Our model initially overestimated citric acid production. This may be due to the many internal constraints imposed on the internal metabolism by the intracellular accumulation of, or simply high throughputs of citrate that are not accounted for by the steady-state methodology of flux balance analysis. For example, the citrate sensitivity of 6-phosphofructo-1-kinase is a target of attempts to increase citrate production [21] and the rates of mitochondrial citrate export [22] and citrate secretion may be limiting. To reflect these constraints, a limit to the citric acid output rate, v _CIT, was added and fitted as a parameter to more closely reflect empirical data (Table 2). Carbon uptake was decreased slightly as a result of the constraint on citric acid output, but still gave close fits to empirical data. The new model was assessed by calculating the AIC (see “Methods”), which showed a significant improvement (Table 3).

Citric acid production on other carbon sources

To further investigate the diauxic growth behaviour, we tested citric acid fermentation using d-xylose as a substrate at an initial concentration of 160 g/L. The same diauxic growth shift coupled citric acid response was seen with xylose (Fig. 4) as seen with glucose. We applied our model, with previously fitted parameters unchanged. The empirical data from this experiment were not used in previous model training, and served to provide further validation with glucose as substrate and at a different phosphate level. The uptake rate of xylose was modelled similarly to glucose as the sum of passive and facilitated diffusion. The kinetic parameters for xylose uptake were fitted to our empirical data (Table 2). Close fits were achieved for biomass production and carbon source consumption, demonstrating the wide applicability of the dynamic model. Citric acid production was overestimated by the model, which may suggest a further limiting factor with xylose as the carbon source. The constraint applied to citric acid output rate, v _CIT, was the same as for glucose (Table 2). The discrepancy may be due to differing morphology as we observed decreased biomass pellet sizes and higher viscosity in cultures grown on xylose.

Investigating the role of phosphate during citric acid fermentation

As growth on glucose continued beyond external phosphate depletion (Fig. 2b), it became clear that A. niger has a phosphate storage mechanism, possibly via accumulation of polyphosphate as previously reported [23]. To investigate this, polyphosphate was extracted from biomass grown under citric acid-producing conditions and quantified. Polyphosphate levels were observed to rise early on in fermentation, peaking at day 2 at the point of external phosphate depletion (Fig. 5). Polyphosphate levels dropped rapidly from day 2 to day 4, with a more gradual decrease later in fermentation coinciding with phosphate-limited growth and citric acid production.

To further investigate the importance of phosphate, we searched for the genes encoding phosphate transporters in A. niger ATCC 1015. A total of eight putative genes were found (based on similarity to known transporters), suggesting that A. niger has evolved a range of phosphate uptake mechanisms as adaptation to different environmental conditions (Additional file 1: Table S2). It may be that only a subset of these genes encode phosphate transporters while others encode phosphate sensors. One of the genes (Accession Number EHA22558) has clear homologues in other species (Additional file 2: Figure S1), but none of these have been characterised or parameterised at the level of protein activity. The other gene annotations are more speculative so may not encode phosphate transporters [24].

The dFBA model provides a platform for predictive metabolic engineering

A prediction of the model is that oxalic acid production is the most efficient means of acidification at initial pH 7, followed by citric acid. It is well known that A. niger predominantly secretes oxalic and gluconic acid at higher initial pH and that by imposing a low initial pH during fermentation, production of these competing organic acids is prevented and citric acid production is increased [14]. Our model suggests that by switching off oxalic acid production by deletion of oxaloacetate hydrolase (oah), citric acid will solely be produced. The model does not predict gluconic acid production suggesting that this may be decoupled from proton production and is instead a means of quickly sequestering glucose, through the action of extracellular glucose oxidase, early in fermentation.

To investigate this phenomenon, we engineered the ATCC 1015 strain by targeted gene deletion strategies to knockout oah and the gene encoding glucose oxidase (gox) responsible for gluconic acid production. We created two single knockouts (Δoah and Δgox) and a double knockout (Δoah Δgox), and characterised citric acid fermentation by these knockout strains at initial pH 7 (Fig. 6). Citric acid yield was significantly increased in the Δoah strain with a further marginal improvement in Δoah Δgox. This was not the case for the Δgox strain suggesting gluconic acid production occurs independently of proton production without impacting citric acid fermentation. Gluconic acid was produced early in fermentation while oxalic and citric acid production occurred later. The synchronicity of oxalic and citric acid production suggests that they are part of the same proton production response. In this experiment, the Mn²⁺ concentration was increased to 1000 ppb. Citric acid production usually requires Mn²⁺-deficient media, though was previously reported insensitive to Mn²⁺ in an oah and gox double negative mutant strain at pH 5 [25]. The presence of Mn²⁺ did not prevent citric acid production at initial pH 7, suggesting that its effect is limited to low pH conditions.

We applied our dFBA model to the Δoah and Δgox knockouts at initial pH 7, which gave close fits for oxalic and gluconic production. The differences in predicted citric acid production between the knockout strains showed a qualitative fit with empirical data (Fig. 6). However, constraints on oxalic and citric acid output rates, v _OXAL and v _CIT, respectively (Table 2), were required to achieve the close fits. The constraint on citric acid output rate, v _CIT, was different to that applied at initial pH 2. This may be due to morphological differences as we observed increased biomass pellet sizes when A. niger was grown at higher initial pH. The impact of differing morphology on transport processes and on anaerobicity within pellets requires further investigation. The widely reported absence of oxalic acid production below pH 2 [10, 25] was implemented in the model to reflect empirical data. To simulate gluconic acid production, the flux of the extracellular GOX and gluconic acid-dissociation reactions were forced depending on GOX kinetic parameters and the ambient pH. The kinetic parameters V _max and K _M of GOX (Table 1) were taken from the literature [26, 27]. The concentration of GOX per gram biomass dry weight is unknown for these experimental conditions so was fitted to empirical data (Table 2). The proportion of active GOX was based on empirical data of GOX activity at varying pH [28].

Shake flask experiments

Citric acid fermentation experiments were performed in 250-mL DeLong neck baffled shake flasks (Bellco Glass Inc.; Vineland, NJ, USA) with 30 mL medium. Flasks were siliconized with 2% (v/v) dimethyldichlorosilane. Cultures were incubated at 30 °C with shaking at 250 rpm. The following medium was used: glucose (160 g/L), urea (3.6 g/L), (NH₄)₂SO₄ (0.52 g/L), K₂HPO₄ (0.5 g/L), CaCO₃ (0.03125 g/L), MgSO₄·7H₂O (0.275 g/L), ZnSO₄·7H₂O (0.00225 g/L), FeSO₄·7H₂O (0.0095 g/L), CuSO₄·5H₂O (0.0117 g/L), MnCl₂·(H₂O)₄ (0.0000108 g/L), citric acid monohydrate (3.3 g/L), Tween 80 (0.0094%). The Mn²⁺ concentration was confirmed as 7 ppb by ICP-MS (Biorenewables Development Centre, York, UK). The medium was autoclaved (121 °C 15 min) excluding glucose which was filter sterilised (0.22 µm). The pH of the medium was adjusted after autoclaving by the addition of sterile 2 M H₂SO₄. The medium included 10 mM uridine in experiments using pyrG negative strains. The medium was inoculated with 1 × 10⁶ spores/mL. Spores were harvested from potato dextrose agar slants incubated for 2 days at 37 °C. 2 mL saline Tween (0.1% Tween 80, 9 g/L NaCl) was added per slant and shaken to disperse spores. Spores were washed 3 times in saline Tween. 500 µL samples of cultures were taken every 24 h for determination of biomass, metabolites, and phosphate. Samples were collected in pre-dried, pre-weighed 1.5 mL Eppendorf tubes and centrifuged at 9000g for 5 min. The supernatant was retained for metabolite analysis and phosphate determination and stored at − 20 °C.

Biomass dry weight determination

Mycelia were washed 4 times in 1 mL dH₂O and centrifuged at 9000g for 5 min. Biomass was dried at 70 °C to constant weight. Biomass dry weight was determined by subtracting weight of the pre-dried 1.5 mL Eppendorf tube.

Metabolite analysis

Enzymatic assay kits were used to determine the level of metabolites. Glucose, citric acid, xylose, glycerol, and gluconic acid were determined using Megazyme assay kits (K-GLUC, K-CITR, K-XYLOSE, K-GCROLGK, and K-GATE, respectively) (Megazyme International Ireland Ltd., Wicklow, Ireland). Oxalic acid was determined using the LIBIOS oxalate assay kit (Oxalate-100; LIBIOS, France).

Phosphate determination

Phosphate was determined by the ammonium molybdate method, using an assay kit (ab65622; Abcam, Cambridge, UK).

Polyphosphate extraction and quantification

Mycelia were grown up in shake flasks using the same method as previously described. Mycelia were harvested at 8 time-points (days 1–8) in triplicate. To obtain sufficient biomass, one flask was harvested per sample. Day 1 samples required the pooling of four flasks per replicate. Mycelia were harvested using a double layer of Miracloth (Calbiochem) and washed in 300 mL ice-cold 100 mM Tris·HCl pH 7 followed by 600 mL ice-cold dH₂O. Washed mycelia were transferred to 15-mL Falcon tubes, flash frozen in liquid nitrogen, freeze dried, and stored at − 80 °C. Freeze dried mycelia were weighed out in 2 mL vials, approximately 50 mg per vial. Biomass was ground using the TissueLyser II (QIAGEN; Crawley, UK) at 30 Hz for 90 s 3 times. Each vial contained two beads. Powdered mycelia were lysed by adding 2 mL 10% (w/v) lysing enzymes from Trichoderma harzianum (Sigma, Dorset, UK) and incubating at 30 °C with shaking for 3 h. Samples were centrifuged and supernatant discarded. Polyphosphate was extracted following a previously described protocol [23]. All centrifuge steps were done at 13,000 rpm for 10 min at 4 °C and all shaking was done at 30 rpm. The polyphosphate fraction was dried in a Savant SPD131DDA SpeedVac Concentrator (Thermo Fisher Scientific). Polyphosphate was quantified by measuring free phosphate before and after acid hydrolysis using the previously described phosphate determination method. Acid hydrolysis was performed by adding 2 mL 0.5 M H₂SO₄ to the dry pellet and boiling at 100 °C for 3 h.

Dynamic modelling of organic acid fermentation

The kinetic parameters were either fitted to our empirical data (Table 2) or set to empirical values from the literature if available (Table 1).

The iMA871 model was adapted to include proton production as an objective function and acid-dissociation reactions for seven acids (citric, oxalic, gluconic, acetic, malic, succinic, lactic) but as a function of a dynamic external pH rather than a fixed pH [14]. The number of protons released in each acid-dissociation reaction was calculated at each time-step according to the following equation based on ambient pH and pKa values.

An output reaction was added for external protons (Hpe <==>), which was set as the objective when maximising proton production. An explicit phosphate storage reaction was also included in the dFBA. An input reaction for internal phosphate (PI <==>) was added to the metabolic model, and the dynamic pool of internal phosphate was tracked in the dFBA. This new reaction was set as the objective when maximising phosphate storage.

When plotting alongside empirical data, the dFBA start time was taken as the spore germination time, 18 h after inoculation. The initial biomass dry weight was set to 0.3125 g/L following empirical data.

Model parameterisation

Glucose transport-mediated uptake [16, 17] and glucose oxidase [26, 27] kinetic parameters were calculated from empirical data in the literature (Table 1). The concentration of active GOX enzyme [GOX] was fitted to empirical data (Table 2). The other kinetic parameters in the model were fitted to empirical data via a manual fitting routine (Table 2).

Quality of fit assessment and model selection

Targeted gene deletion of oah and gox