HUNGARIAN JOURNAL OF INDUSTRY AND CHEMISTRY Vol. 47(2) pp. 1–4 (2019) hjic.mk.uni-pannon.hu DOI: 10.33927/hjic-2019-13 INVESTIGATIONS INTO SUCCINIC ACID FERMENTATION ÁRON NÉMETH *1 1Department of Applied Biotechnology and Food Science, Budapest University of Technology and Economics, Műegyetem rkp. 3, Budapest, 1111, HUNGARY Succinic acid (SA) is an important chemical intermediate from which fine chemicals ( e.g. detergents), additives (for pharmaceuticals, food (taste), plant growth stimulants) as well as other important intermediates (maleic anhydride, suc- cinimide, 2-pyrrolidinone, dimethyl succinate) can be manufactured. Since SA is involved in the central metabolism of cells (in the tricarboxylic acid (TCA) cycle), it is a key player in the biochemistry of life, which has the potential of biotech- nological production. Since SA is formed in the “middle” of the TCA cycle it can be formed by both CO2 production and fixation. The significance of the latter is that the amount of the product can be controlled by the availability of CO2, since stoichiometrically one molecule of CO2 is fixed by one molecule of SA. In our studies of compositions of Actinobacillus succinogenes media, the role and effect of pH regulator compounds as well as the effect of an inert atmosphere were investigated in terms of the yield. Furthermore, in fermentation experiments, the application of higher sugar concentra- tions was also studied. On the basis of different fermentations, a neural network for modelling and describing how factors influence SA production was established. Keywords: succinic acid, Actinobacillus succinogenes, neural network, modelling 1. Introduction Succinic acid (SA), as an intermediate of the tricar- boxylic acid (TCA) cycle, plays an essential role in the metabolism of microorganisms. SA can be produced by many anaerobes or facultative anaerobes as a metabolic product, thus can be used as an important platform chem- ical, a precursor of many pharmaceuticals, feed additives, green solvents or biodegradable polymers. SA itself is a colourless, odourless and crystal-forming compound. Since this metabolite is bifunctional (pKa1 = 4.21, pKa2 = 5.72 [1]), it is very reactive so has many potential applications, e.g. it plays important roles in the synthesis of γ-butyrolactone, maleic anhydryde, succinimide, 1,4- butanediol, dimethyl succinate, succinonitrile and 1,4- diaminobutane. It is industrially produced, mainly syn- thetically, in a complex way from maleic anhydride found in crude oil, which is both economically and environ- mentally unfavourable [2]. Therefore, its biotechnologi- cal production is a current research topic to find an alter- native method to avoid the above-mentioned side effects [3]. Furthermore, a great advantage of the microbial pro- duction of SA is that one of the initial biochemical re- actions is the carboxylation of phosphoenolpyruvate [4] which is regulated in the case of anaerobic bacteria by the availability of CO2 [2], thus the elevation of the CO2 con- centration can shift product portfolio from formate and ethanol towards SA (Fig. 1). *Correspondence: naron@f-labor.mkt.bme.hu Despite this fact, the process can help to de- crease the CO2 emissions of the human population [5]. Both fungi and bacteria can be found among the SA- producing microorganisms, but their ability to produce SA differs significantly: fungi ∼45 g/L (Aspergillus niger, rec. Yarrowia lipolytica), Gram-negative bacte- ria (wild-type Actinobacillus succinogenes 98.7 g/L, Mannheimia succiniciproducens 90 g/L) and Gram- positive bacteria (Clostridium thermosuccinogenes 82.5 g/L, rec. Corynebacterium glutamicum 145 g/L) [5]. The most frequently applied strains in the industry are Aspergillus niger, Aspergillus fumigatus, Byssochlamys nivea, Lentinus degener, Paecilomyces variotii, Peni- cillium viniferum, Saccharomyces cerevisiae and Acti- nobacillus succinogenes [4]. The latter bacterium is one of the most prominent producer strains that is isolated from bovine rumen and has been identified as a member of the genus Pas- teurella, which is a facultative anaerobic, non-motile, Gram-negative pleomorphic, rod-shaped bacterium [2]. It has great potential in terms of SA production, because of its higher yield and wide range of applicable sub- strates, e.g. glucose, cellobiose, maltose, lactose, saccha- rose, fructose and sorbitol. Furthermore, it can tolerate high initial concentrations of glucose, therefore, is suit- able for simple batch fermentation instead of the more complex and costly fed-batch culture technique [5]. The most widely applied strains are Actinobacillus succino- https://doi.org/10.33927/hjic-2019-13 mailto:naron@f-labor.mkt.bme.hu 2 NÉMETH Figure 1: SA metabolism with CO2 regulation genes 130Z and its variants (FZ6, 9, 21, 45 and 53) [2], which can tolerate both high glucose and SA concentra- tions as well as achieve high SA yields [5]. The aim of this paper is to compare fermentations of Actinobacillus succinogenes under different experimental conditions, then – due to the many influential factors – to set up a neural network-based model which can be used to predict high SA titres. 2. Materials and Methods 2.1 Cultivation of the bacteria Actinobacillus succinogenes 130Z (DSM22257) was cul- tivated in 10 ml of tryptic soy broth (TSB) (Sigma) in an impedimetric BacTrac 4100 (SY-LAB, Austria) anaero- bic cell. For a 1 L fermentation with a working volume of 0.8 L, an AS medium was applied according to Liu et al. [5] as follows: 62.5 g/L total sugar including 44.9 g/L saccharose, 9.8 g/L glucose and 7.2 g/L fructose sup- plemented with 15 g/L yeast extract, 1.5 g/L NaHPO4, 1g/L Na2HPO4, 1 g/L NaCl, 0.2 g/L MgCl2 and 0.2 g/L CaCl2. During preliminary tests, a temperature of 37◦C was not successful for SA production, therefore, 34◦C was applied and the pH regulated by 3M Na2CO3. Af- ter each fermentation, 200 ml of broth remained in the reactor and 600 ml was extracted, while 600 ml of fresh AS media was introduced. The 5 different fermentations are compared in the Results section. An innovative solution was to apply a CO2 economi- cal gas supply through the oxygen enrichment system of a Biostat Q DCU3 fermentor system in the absence of air and oxygen. The gas mix oxygen enrichment regulator was set at 4 %, therefore, periodically a small amount of CO2 was introduced into the fermentation broth. 2.2 Analysis During fermentations, samples were taken periodically and the optical density (OD) determined by a spectropho- tometer (Ultrospec Plus, Pharmacia LKB) at a wave- length of 600 nm (OD600) against the supernatant of a centrifuged sample by applying the same dilution factor (5×) as in the case of the samples. The cell dry weight (CDW) was obtained by a multiplication factor of 2 from OD600. Substrate consumption and product as well as by-product formation were detected by the Waters Breeze HPLC System by applying 5 mM H2SO4 in deionized water at a flow rate of 0.5 ml/min through a BioRad Aminex HPX87H column at 65◦C in a refractive index (RI) detector at 40◦C. 2.3 Neural networking For model building and evaluation, Neural Designer v2.9.5 was used by applying the following 4 steps: 1. Fermentation data were combined into a single MS Excel spreadsheet and exported to a tab-delimited text file, which could be imported into the modelling software. 9 variables, i.e. 7 inputs (time, lactic acid, acetic acid, propionic acid, glycerol, ethanol, total sugar) and 2 outputs (succinic acid, CDW), were applied to 58 fermentation samples from which 36 were used for training the network, 11 were selected and 11 were used for testing the behaviour of the model. 2. From among the many options, a model was defined (Fig. 2) by automatic scaling, without any principle component, with 2 layers and 3 neurons/hidden lay- ers that exhibit a logistic activation function in the absence of a bounding layer. In terms of a training strategy, the normalized mean squared error method was selected using a Quasi-Newton algorithm and a maximum of 1000 iterations. Incremental order was chosen as the order selection algorithm together with the growing inputs. 3. Model fit, i.e. performance training, was conducted. 4. Output of the model: impact figures of factors were determined (the other parameters were fixed), model equations obtained and predictions made by imple- menting input data into an input data matrix. 3. Results and Discussion Since Actinobacillus succinogenes is a facultative anaer- obic microorganism, the first fermentation experiment (Fig. 3A) was started in the absence of any specific at- mosphere. However, it ran very slowly, therefore, around 48 h (denoted by a red arrow) of continuous 5 % CO2 enrich- ment was applied via a zero flow rate gas inlet. The exper- iment confirmed that the application of CO2 is essential to form SA, therefore, finally 6 g/L SA was achieved and the model fitted very well for both CDW and SA mea- surements. Hungarian Journal of Industry and Chemistry INVESTIGATIONS INTO SUCCINIC ACID FERMENTATION 3 Figure 2: The used neural network for describing SA fer- mentations Therefore, the next fermentation (Fig. 3B) was con- ducted under 5 % CO2 enrichment of zero flow rate gas inlet and resulted in the highest 16 g/L SA (besides 11 g/L residual sugar) corresponding to a yield of 37 %. To increase the economic feasibility, the amount of CO2 was reduced by 50 % to 2.5 % in the following experi- ment (Fig. 3C). To avoid a fall in the SA concentration, detected by-products (such as lactic acid 3.5 g/L, acetic acid 5.7 g/L, propionic acid 1.7 g/L and glycerol 1.3 g/L) should be repelled as a result of the addition of 20 g/L calcium lactate. The model once again fitted well to the measured CDW and SA values. Unfortunately, the carbon flux shifted towards propionic acid formation, therefore, the subsequent experiment (Fig. 3D) was both supplemented with 20 g/L lactic acid and 11 g/L propionic acid, but this resulted in a low level of SA and a high level of lactic acid (52 g/L). The model yet again fitted well to the measured CSW and SA values. Independent fermentation results were also checked by the model (Fig. 3E) and resulted in very good fits. It can be concluded that the artificial neural network model constructed described well the SA fermentations, which is in line with the results of Li et al. [7], namely that ar- tificial neural network models can describe succinic acid fermentation better than response surface methodology. The presented results revealed that for SA fermenta- tion, 5% CO2 enrichment is essential and cannot be fully or partially replaced with the addition of lactic acid or propionic acid. After validating the model by conducting 5 different fermentations, it was used to optimize influential factors via impact figures (directional output plots) (Fig. 4). These trends show that all the presented factors cor- relate with SA concentration, i.e. any increment in their amounts resulted in an increment in SA, with the excep- tion of propionic acid which exhibited a negative corre- lation. These suggest that the addition of propionic acid can decrease the concentration of SA obtained and the addition of lactic acid can increase it. (A) (B) (C) (D) (E) Figure 3: Actinobacillus succinogenes fermentations with different degrees of CO2 enrichment in combination with lactic acid (LA) and propionic acid (PA). Dots indicate the measured values and lines indicate the model prediction. 47(2) pp. 1–4 (2019) 4 NÉMETH Figure 4: Factors impact plots: PrOH – propionic acid (g/L), Total sugar (g/L), LA – lactic acid (g/L) and CO2 – carbon dioxide (%) 4. Conclusion Five Actinobacillus succinogenes fermentations were run under different conditions: semi-anaerobic method, con- trolled introduction of CO2, and introduction of CO2 combined with either the addition of lactic acid or both lactic acid and propionic acid. All together 58 samples were taken and analysed, the results of which were en- tered into Neural Designer modelling software and used for training and testing the model. While the fermenta- tions resulted in very different final SA concentrations, the established model fitted well to all of the fermen- tations, even to the one which was not used for model building, testing and validation. While economical CO2 enrichment was successfully applied and resulted in the highest SA yield (37 %), the addition of lactic acid and propionic acid was not successful in terms of SA concen- tration. Acknowledgement This research was conducted within the framework of and in cooperation with PROGRESSIO Engineering Bureau Ltd., MÉL Biotech K+F Kft. and Budapest University of Technology and Economics. REFERENCES [1] Tonova, K.: State-of-the-art recovery of fermenta- tive organic acids by ionic liquids: An overview Hung. J. Ind. Chem. 2017 45(2), 41–44 DOI: 10.1515/hjic-2017-0019 [2] Song, H.; Lee, S. Y.: Production of succinic acid by bacterial fermentation Enzyme Microb. Tech. 2006 39(3), 352–361 DOI: 10.1016/j.enzmictec.2005.11.043 [3] Sauer, M.; Porro, D.; Mattanovich, D.; Branduardi, P.: Microbial production of organic acids: Expand- ing the markets Trends Biotechnol. 2008 26(2), 100–108 DOI: 10.1016/j.tibtech.2007.11.006 [4] Zeikus, J. G.; Jain,·M. K.; Elankovan, P.: Biotech- nology of succinic acid production and markets for derived industrial products Appl. Microbiol Biot. 1999 51(5), 545–552 DOI: 10.1007/s002530051431 [5] Liu, Y.-P.; Zheng, P.; Sun, Z.-H.; Ni, Y.; Dong, J.- J.; Zhu, L.-L.: Economical succinic acid production from cane molasses by Actinobacillus succinogenes Bioresource Technol. 2008 99(6), 1736–1742 DOI: 10.1016/j.biortech.2007.03.044 [6] Mika, L.T.; Cséfalvay, E.; Németh, Á.: Catalytic conversion of carbohydrates to initial chemicals: Chemistry and sustainability Chem. Rev. 2018 118(2), 505–613 DOI: 10.1021/acs.chemrev.7b00395 [7] Li, X.; Jiang, S.; Pan, L.; Wei, Z.: Optimization for the bioconversion of succinic acid based on re- sponse surface methodology and back-propagation artificial neural network in: Wang, H; Low, K. S.; Wei, K.; Sun, J. (eds.): Fifth International Con- ference on Natural Computation 2009 3, 392– 398 (IEEE Computer Society, Los Alamitos, USA) ISBN: 978-0-7695-3736-8 DOI: 10.1109/ICNC.2009.20 Hungarian Journal of Industry and Chemistry https://doi.org/10.1515/hjic-2017-0019 https://doi.org/10.1515/hjic-2017-0019 https://doi.org/10.1016/j.enzmictec.2005.11.043 https://doi.org/10.1016/j.tibtech.2007.11.006 https://doi.org/10.1007/s002530051431 https://doi.org/10.1016/j.biortech.2007.03.044 https://doi.org/10.1016/j.biortech.2007.03.044 https://doi.org/10.1021/acs.chemrev.7b00395 https://doi.org/10.1109/ICNC.2009.20 Introduction Materials and Methods Cultivation of the bacteria Analysis Neural networking Results and Discussion Conclusion