The acetone-butanol-ethanol (ABE) process is the production of the eponymous solvents by fermentation with solvent-forming Clostridia. In 1862 Louis Pasteur discovered motile, rod-shaped bacteria to produce butanol under anaerobic conditions [1
]. In 1915, Chaim Weizmann established an industrial process for the biotechnological production of acetone (and butanol), which was used all around the world in the following decades before being replaced by a synthesis of these solvents from mineral oil [3
]. Nowadays, we await a comeback of the industrial ABE process as not only the demand for sustainable produced solvents and bulk chemicals is increasing, but butanol is further discussed as a promising biofuel. Beside the exploration of new non-food and inexpensive waste materials as substrates for the fermentation process [5
], an important setscrew to improve the economics of the process is the development of new fermentation technologies [7
]. These should allow effective raw material conversion with a high yield, accelerate the onset of solvent formation in a biphasic ABE process and stabilize the bacteria in this metabolic phase. While in the first phase of the ABE fermentation, called acidogenesis, acetic and butyric acid are the main fermentation products, the formation of butanol, acetone and ethanol under partial re-assimilation of the organic acids occurs in the second phase, called solventogenesis. The research of the past decades has shown that low pH and/or the high concentration of the undissociated organic acids trigger the metabolic switch from acidogenesis to solventogenesis [9
]. Investigations with external electron donors, such as methyl viologen or neutral red, have further shown that the redox state of the cells influences the bias for acidogenesis and solventogenesis, respectively [11
]. Moreover, it has been proposed that small concentrations of butanol might promote the metabolic shift [14
]. Clostridium acetobutylicum
was used as a model organism in the majority of the studies investigating the metabolic switch. However, under certain conditions C. acetobutylicum
as well as other clostridial species fails to switch its metabolism to the solventogenic state, resulting in a hyper-acidification of the fermentation broth (pH ≤ 3.8) and a complete arrest of metabolic activity. The molecular mechanism behind this phenomenon often described as “acid crash” is still open. In some cases the spontaneous loss of the mega-plasmid pSOL1, encoding genes essential for solvent production, is a sound explanation. However, how this event can occur repeatedly under certain conditions, i.e., low buffer concentrations, while it rarely occurs under other conditions, i.e., high buffer concentrations, remains to be clarified. Explaining and controlling the metabolic shift has been the aim of numerous studies in the past [15
]. In batch-experiments, controlling the pH at 4.5 to 5.0 has been described as a successful strategy for a stable solventogenic phase achieving high butanol concentrations [10
]. However, onset of solvent formation is observed not earlier than 15 h after start of fermentation under such conditions. The maximal product concentrations are only reached after about 36 h [10
], which limits the productivity of the batch fermentation processes. Additions of acetate and butyrate have been shown to increase the productivity of batch processes [17
]. However, determining the optimal fermentation phase for acetate (or butyrate) co-feeding is challenging and laborious in a batch process as precise reproducibility of the time profile is limited. We here developed a model system that captures the metabolic phases of the ABE fermentation in different stages of a continuously operated bioreactor cascade at steady state. On the one hand this system allows analyzing the physiological characteristics of cells in the different metabolic phases, including the otherwise only transient occurring phases. On the other hand we aim at accelerating the solvent production and hence increasing the productivity of the fermentation process by varying the operating conditions of the cascade. In order to reduce the experimental work for this second task, we started developing a mathematical model that describes the continuous, multi-stage ABE fermentation process taking into account the presence of different subpopulation types in each bioreactor tank. Using experimental data from four operating conditions of the linear cascade, the unknown kinetic parameters of the model were estimated and the predictive capacity of the steady state simulations was validated. To further validate the model and to test a potential strategy for faster solvent formation, the simulator as well as the experimental set-up was modified by introducing a feedback loop from bioreactor 4 to bioreactor 2 of the cascade. In that way the fermentation broth with solventogenic cells and a higher butanol concentration is introduced into an early phase of the ABE process.
Here we analyzed continuous ABE fermentations with suspended cells in a six-stage bioreactor cascade. Under the investigated operating conditions, a good compromise of high product concentrations and high volumetric productivity was found for the 3-stage (truncated) cascade, operated at pHbr1 of 4.3 and Fin of 0.1 L h−1 as well as for the 6-staged cascade, operated at pHbr1 of 4.3 and Fin of 0.22 L h−1, i.e., for total mean residence times in the cascade of 12 h and 10.9 h, respectively. Under these conditions, volumetric productivities and final butanol concentrations of 0.81 ± 0.04 g butanol L−1 h−1/9.7 ± 0.5 g butanol L−1 and 0.75 ± 0.23 g butanol L−1 h−1/8.2 ± 2.6 g butanol L−1 were reached, respectively.
These productivities are more than twice times higher than those observed in a two-stage ABE process [23
], being the prototype of a multi-stage, continuous ABE fermentation. In that former work, residence times of 8 h in the first bioreactor (pH 4.3) and 33–42 h in the second bioreactor (pH 4.3) yielded final butanol concentrations of 10.5–12.6 g butanol L−1
, but productivities of only 0.30–0.32 g butanol L−1
]. Since then a high number of continuous ABE fermentation strategies has been explored and published, being summarized in a matrix of final butanol concentrations and volumetric butanol productivities by Setlhaku et al. [24
] This matrix shows that product concentrations above 13 g L−1
can only be reached in systems with integrated downstream processing, i.e., with multiple stages separating biological solvent production and technical product concentration (e.g., by gas stripping or pervaporation) [25
]. Furthermore, this matrix illustrates that butanol productivities of 1 g L−1
and above are exclusively reached in systems with immobilized cells or cell retention [30
]. A four-stage bioreactor cascade with Clostridium cells immobilized in a biofilm has been reported to yield up to 10.8 g butanol L−1
with a productivity of 9.2 g butanol L−1
However, the focus of our work was the establishment of a system that captures the temporal development of a batch process in a spatial dimension. Such an approximation of a plug flow bioreactor has been realized in the here-presented six-stage CCSTR. In contrast to biofilm or other fixed-bed bioreactors, cells from a certain metabolic phase can be retrieved from the CCSTR for further characterization or for a transfer into another environment. The characterization of metabolic phases in the ABE fermentation based on batch processes is linked to sophisticated prearrangements and poor reproducibility [16
]. In the future, experimental characterization of individual cell populations in the different metabolic phases using flow cytometry [34
], electrooptical measurements [38
] or single cell RNA sequencing [39
] can be applied to steady state samples of the CCSTR. Already, the here-presented model, showing good agreement with the experimental data, delivers prospects of the composition and the characteristics of the individual cell populations.
For instance, it is worth mentioning that the option for growth of intermediate and solventogenic cells, which had been newly introduced as an extension to an earlier model [20
], resulted in maximal growth rate parameters of 0.46 h−1
and 0.41 h−1
for intermediate and solventogenic cells, respectively, after parameter estimation. These are lower than the maximal growth rate parameters of acidogenic cells, being 0.73 h−1
. This result is in good agreement with the experimental analyses from other groups [22
], showing that also under solventogenic conditions a stable steady state can be reached in single-stage continuous fermentation. However, this is characterized by a lower optical density than under acidogenic conditions [9
], which might result from a lower maximal specific growth rate of the solventogenic cells.
In the model of Millat [9
], simulating the pH-induced metabolic shift in a one-stage chemostat, two subpopulations—acidogenic and solventogenic cells—had been considered. However, in that model the dimensions of the subpopulations develop independently from each other, i.e., the model functions of two subpopulations are not linked to each other. In the here-presented model the development of the subpopulations is a differentiation process from acidogenic to intermediate and further to solventogenic cells, which depends on the concentration of undissociated acids. Implementing the mechanism of cell differentiation in the model, we followed an example of modeling the segregation of dihydroxyacetone-producing Gluconobacter axydans
]. One result of our simulation was that only a comparatively small proportion of cells, i.e., 26–40%, differentiated to solventogenic cells in the last bioreactor stage, while 29–54% remained in the acidogenic state. This contrasts with the model of Millat [9
] where 100% of the acidogenic cells were transformed to 100% of solventogenic cells during the metabolic switch. However, shifting of subpopulation proportions rather than homogeneous differentiation agrees well with the recent insights into metabolic heterogeneity of bacterial communities, gained by now available single-cell analysis methods, e.g., single-cell RNA sequencing [40
]. For instance, in antibiotic stress situations many bacterial populations tend to segregate into subpopulations of metabolic-active, antibiotic-sensitive cells and of persistent, metabolic-inactive, but antibiotic-resident cells. The latter group ensures the survival of the species and may profit from nutrients released by cell lysis of the first group once the antibiotic pressure is withdrawn [45
]. Heterocyst formation of Nostoc punctiforme
is another example for beneficial differentiation processes in bacterial populations [46
Our model does not consider back differentiation from solventogenic cells to acidogenic cells, though shifting experiments in the single-stage continuous ABE fermentation propose that those are possible [47
]. However, under the experimental conditions applied here we did not expect a back differentiation to occur. Furthermore, the remaining part of acidogenic cells even under externally solventogenic conditions, proposed by our model, offers the perspective that not a back differentiation, but rather overgrowing of the solventogenic by the acidogenic subpopulation leads to a backward shift from overall solventogenic to overall acidogenic conditions.
Validation and improvement of the here presented model predictions of the subpopulation composition will be enabled by future experimental investigations, e.g., by flow cytometry. So far, the model kept its predictive capacity even when switching operating conditions from the linear cascade to a cascade with feedback loop, which is another validation of the presented model.
The hypothesis from pulse experiments by Junne [14
] proposes that butanol has an auto-inducing effect on the solvent production by promoting the differentiation process. Using the potential of our model, we simulated a feedback loop introducing solventogenic fermentation broth from bioreactor 4 (including ~21% of solventogenic cells and 5.3 g of butanol) into an earlier phase of the process in bioreactor 2. On one hand, one would expect a beneficial effect of the recirculated butanol on differentiation to solventogenic cells, which should even amplify itself after bioreactor 2. On the other hand there should be a beneficial effect, resulting from the recirculation of already differentiated solventogenic cells, which is expected to be rather constant in all bioreactors after entry from the feedback loop. However, here we observed significant higher solvent production in bioreactors 2 and 3 in the feedback loop cascade, but the beneficial effect declined in the later bioreactors, resulting in final butanol concentrations only 0.2 g L−1
higher than in the linear cascade. So there is no evidence for an amplification of solventogenesis triggered by butanol. However, the model prediction for the feedback loop cascade, matching the experimental results, indicates that a positive effect of a higher percentage of solventogenic cells in bioreactors 2 and 3 is compensated mainly due to reduced differentiation of new solventogenic cells caused by a decrease in the acetic and butyric acid concentrations.
As a new hypothesis, adding a feed of acetic and/or butyric acid to bioreactor 2 or 3 of the feedback loop cascade is one promising configuration that will be tested in the future with the here-introduced modeling tool and experimental setup for studies of the ABE fermentation.
Using the mathematical model not only cascade set-ups with additional feeding points, but also variations of the cascade with variable number of bioreactor stages, differently sized stages or cell recycling loops, can be evaluated and selected for experimental investigations. Thus, this work opens the door to a model-supported optimization of a continuous, multi-stage ABE process.
During the ABE process Clostridia go through different metabolic phases. With the here-presented cascade of continuous stirred tank reactors (CCSTR), it is possible to separate these phases in a spatial manner. Investigation and application of this reactor system is supported by a mathematical model, which combines two interacting levels of the process:
Reactor system model, depending on the configuration of the reactors, describing residence time distributions and their influence on microbial population fractions and metabolite concentrations;
kinetic model for microbial metabolism, allowing for populations with different metabolic activities, depending on the bioreactor environment.
The mathematical model, which fits the experimental steady state data of all operating conditions under investigation, indicates that even in the later bioreactor tanks of the cascade only a relatively small portion of the cells (about one third) is in the economically interesting solvent-forming state. Our attempt to increase the proportion of the solventogenic cells by introducing a feedback loop into the cascade from bioreactor 4 to bioreactor 2 resulted in a final butanol concentration of 8.3 g L−1 and a productivity of 0.76 g butanol L−1 h−1, which were only slightly higher than in the corresponding linear cascade. Achieving only this small improvement may be caused by the limited availability of substrate for conversion to solvents. The proposed addition of acetic and/or butyric acid to bioreactor 2 could overcome this limitation.
In the future the here-established multi-stage, ABE laboratory process with its corresponding mathematical description will serve as a tool for predicting and testing further fermentation strategies such as co-feeding of organic acids to different phases of the process. A collection of possible configuration modifications is shown in Figure 8
The modifications in Figure 8
serve different purposes:
Adapting individual mean residence time in bioreactors by using different working volumes.
Moving cells and metabolites between bioreactor environments by feed-back and feed-forward loops.
Feeding additional substrates along the cascade, e.g., organic acids.
Biomass retention by separation and recycle.
Modeling metabolism in the cascade led to a more profound understanding of the shift from acidogenic to solventogenic state of the cells and the metabolic activity of physiologically different subpopulations. Together with the reactor model, allowing for simulation of any combination of configuration modifications, the performance of a cascade set-up with its operating conditions can be predicted.