Starting from the design parameters presented in the methodology section, we estimate an annual enzyme production rate of 88 t of enzyme/year in the baseline scenario, as shown in Fig. 1. As described previously (“Methods” section), we considered that the annual production of β-glucosidase proposed here would be sufficient to be used on the hydrolysis of approximately 39% of all the sugarcane bagasse produced annually by a Brazilian sugarcane plant that processes 2 million tons of sugarcane/year.
The process requires a seed train composed of two smaller fermenters, with volumes of approximately 0.2 and 4 m3. During each process cycle, 80 t of culture medium is fed to the main fermenter, and approximately the same amount of cell broth is found at the end of the microbial culture. This cell broth is collected in a storage tank and then lysed in a pressure homogenizer (throughput of 216 L/min). Next, the cell debris generated is removed using six disk-stack centrifuges (throughput of 36 L/min), and the remaining solids are removed by dead-end microfiltration (filter area of 50 m2). Finally, the enzyme solution is concentrated to 15 g/L and stabilized in a citrate buffer using a diafiltration system composed of eight ultrafilters (each with 74 m2 filter area) such that 22 tons of concentrated enzyme solution is produced per cycle, with a total of 264 cycles performed per year.
Economic assessment
The overall unit production cost obtained for the baseline scenario was approximately 316 US$/kg of enzyme. This value is approximately 32 times higher than the estimated cost of the fungal enzyme mixture (10 US$/kg protein), as provided by Klein-Marcuschamer et al. [3], and also higher than other similar enzymes available in the literature (Additional file 3: Table S7). In the scenario proposed here, an on-site production of β-glucosidase for supplementation of fungal cellulases, such costly BGL would increase the final cost of the fungal cocktail by 137%. We estimated this value considering a fungal enzymatic cocktail with a specific filter-paper activity of 0.5 FPU/mg of protein, a BGL-specific activity of 2.3 CBU/mg, and that BGL is supplemented at a ratio of 0.2 CBU/FPU. This cost increase would not be justifiable in view of the observed effect of BGL supplementation on the enzymatic hydrolysis yield [10]. However, the baseline scenario was intentionally constructed based on conservative assumptions in order to better identify the advantages and limitations of the E. coli recombinant system in this context. Moreover, some simulated scenarios that are described in this work achieved substantially lower values of enzyme cost than the baseline scenario. Thus, in this section, we investigate all the simulated scenarios and the main drivers of enzyme cost with an eye toward possible cost reduction measures.
Cost composition
Figure 2a shows that in the baseline case, facility-dependent costs, which include plant maintenance, depreciation, insurance, local taxes, and overhead costs not directly associated with the process (such as accounting, payroll, fire protection, and security), make up 45% of the unit production costs of the enzyme. Raw materials and consumables (filter cartridges and membranes) account for 25 and 23%, respectively. Since the facility-dependent cost is proportional to the purchased equipment cost in practice, as estimated by SuperPro Designer pricing models based on the US market, the facility-dependent cost may be somewhat overestimated. In fact, for example, Macrelli et al. [15] applied a factor of 0.82 to adjust the fixed capital cost of a bioethanol plant in the US Gulf Coast to Brazilian conditions. Given that the real/dollar currency exchange rate has increased dramatically in 2016 and that the proportion of equipment that would be imported is unknown, we decided not to apply any adjustment factor to the equipment cost provided by the software.
Of the costs of the raw materials, glucose and IPTG account for approximately 47 and 41%, respectively, and nitrogen- and phosphorus-rich compounds are together responsible for 10%. The costs of trace elements and, rather surprisingly, kanamycin seem to be negligible. These results confirm the common-sense idea that the use of less expensive carbon sources and induction strategies are important to reduce the enzyme cost.
Regarding the carbon source, it should be stressed that the cost of the glucose used was the market price of the compound. Therefore, it is reasonable to consider whether a glucose-rich liquor generated in the same (2G) plant could, at least in part, replace the purchased glucose, considerably reducing the cost of the carbon source. Similarly, one may envision replacing the purchased glucose with a glucose-fructose syrup obtained by inverting (hydrolyzing) sucrose in a 1G-plant setting or with glycerol, as suggested by Horn et al. [28]. The cost of glycerol, in particular, has decreased dramatically during the past decade, mainly because glycerol is a by-product of biodiesel production, which has greatly increased during the same period. Xylose-rich liquors generated from the hemicellulose hydrolysis process are also low-cost carbon sources that are not well utilized by the conventional ethanol-producing organism S. cerevisiae. Naturally, the use of these alternative and raw carbon sources could negatively impact the biomass and/or enzyme yields, since these carbon sources usually contain inhibitors of microbial metabolism.
Regarding the cost of induction, IPTG is widely considered to be too costly for the production of inexpensive recombinant proteins, especially at the concentrations at which IPTG is typically used in the laboratory (such as 1 mM). Our results confirm this perception. In fact, the cost contribution of IPTG is comparable to that of the main carbon source, which is 3 orders of magnitude less expensive. However, there are indications that lower IPTG concentrations may give rise to similar or sometimes better volumetric productivity of recombinant proteins, depending on the specific culture and expression conditions [32]. Since IPTG alone accounts for 10% of the unit production cost, reducing the amount of IPTG by one order of magnitude could, theoretically, reduce the enzyme cost by 9%. Alternative induction methods could also be explored, such as replacing IPTG with lactose (while keeping the lac operator) or employing a thermal induction system, which may be particularly convenient in cases in which the quantity and quality of the recombinant enzyme are not affected by this additional stress.
The effects of reducing the amount of IPTG (by tenfold), eliminating kanamycin from the process, and replacing glucose with glycerol were evaluated, assuming in the first two cases that biomass yield and protein productivity are unaffected. According to the simulations, kanamycin elimination and glycerol substitution make little difference in terms of cost, whereas IPTG reduction has a significant positive impact, reducing the enzyme cost by approximately 10%.
Additionally, the cost of consumables, which is quite significant (23%), is mainly due to the cost of the ultrafiltration membranes (80%) used in the diafiltration system and also the cost of the dead-end microfiltration cartridges (20%). Since the reason to use the dead-end filter is to avoid the fouling of the ultrafiltration membrane, one can conclude that, in our proposed process, the operation of the diafiltration unit has a large direct and indirect economic impact on the cost. In the pursuit of alternative units that are less expensive to operate, it might be interesting to concentrate the enzyme using different methods, such as by precipitation followed by centrifugation. However, the choice of the precipitation agent and the impact of this agent on enzyme activity, recovery yields, process complexity, and the environment should be experimentally evaluated. We have previously evaluated the potential of glycosyl hydrolase precipitation using ethanol under different temperature and pH conditions. In our experience, β-glucosidase activity can be almost fully recovered using 90% (v/v) ethanol at 25 °C and pH 6.5, indicating the potential of this solvent for use in a 2G ethanol plant [36].
The cost composition presented above refers to the overall process. However, this composition is far from uniform along the process, as seen in Fig. 2b. The facility-dependent cost, for example, is almost evenly distributed among the three main process sections, whereas the cost of raw materials is almost entirely due to the main fermenter feed, and the cost of consumables is entirely due to the downstream section (because the consumables are associated with the operation of filtration units). Overall, the fermentation section is the costliest section, followed by the downstream section. Nonetheless, the costs associated with the upstream section are significant (≈ 11%).
Effects of scale and operating time
The presented enzyme costs were calculated considering a process scale that corresponds to a main fermenter volume of 100 m3. However, it is useful to evaluate the change in enzyme costs associated with a change in the scale of the process, especially considering that the percentage of bagasse set apart for 2G ethanol production in a plant would also depend on the relative prices of ethanol and electricity.
As shown in Fig. 3a, the amount of enzyme produced grows almost linearly with the scale of the process (represented by the main fermenter volume) throughout the range analyzed (from 25 to 150 m3). The unit production cost of the enzyme, in contrast, decreases in a non-linear manner as the process scale increases from 25 to 150 m3, decreasing drastically at the lower end of the scale and becoming almost flat at the higher end of the scale. The shape of this curve is typical of the phenomenon of economy of scale, largely because the facility-dependent cost becomes relatively smaller as the scale of the process increases. It should be mentioned, however, that the scale-down and scale-up of the process were performed using the software by simply adjusting the process throughput without considering any variations in biomass or enzyme yield that might arise from problems with oxygen transfer or other transport phenomena. Similar to the production scale, the annual operating time of the process strongly affects the unit production cost of the enzyme and is a particularly relevant parameter in the case of an on-site enzyme production plant dedicated to the hydrolysis of lignocellulosic biomass because the harvest of sugarcane does not occur throughout the calendar year but only between April and November, for approximately 7 months. Since sugarcane bagasse contains a high degree of moisture (50%), this material cannot be stored for long periods. In this context, Santos et al. [37] reported that natural bagasse loses approximately 30% of its calorific content in 150 days. Consequently, if the enzyme production unit were used only for bagasse hydrolysis and if there was no bagasse storage, the enzyme plant would remain idle for approximately 5 months/year [38].
Figure 3b shows how the enzyme cost increases, markedly, from approximately 316 to 393 US$/kg (for a 100 m3 fermenter), depending on the annual operating time of the plant. In contrast, if the moisture content of the biomass is decreased to approximately 20% or less, the biomass becomes essentially stable with respect to microbial activity [39, 40]. Therefore, it seems evident that a strategy for the safe, long-term storage of bagasse would have to be developed.
Seed train
Frequently overlooked in the literature, the seed train is a key part of any industrial bioprocess, since it is responsible for the propagation of the microorganism from small volumes to large bioreactors. Ideally, the seed train should preserve the desirable characteristics and the viability of the microorganism while avoiding any contamination. Here, the effect of inoculum volume on the cost of the recombinant enzyme for different process scales was simulated, and the result is presented in Fig. 4, as represented by the volumes of the main fermenter. For all three scales evaluated, an inoculum volume of 5% led to the lowest cost. At the lowest scale (50 m3), an inoculum volume of 10% led to the highest cost, whereas at the 100 and 150 m3 scales, an inoculum volume of 1% led to the highest cost. These results demonstrate the countervailing effects of inoculum size on enzyme cost: on one hand, larger inoculum volumes reduce the duration of the main culture, thereby increasing the number of batches per year and reducing the enzyme production cost. On the other hand, larger inoculum volumes require larger and more numerous seed bioreactors, thereby increasing the facility-dependent cost and the enzyme production cost. Either way, strategies to reuse a minor fraction of the main fermenter biomass instead of a seed train should be evaluated despite the potential decrease in productivity. In particular, the plasmid stability and transgene expression during long-term fermentation using E. coli would be critical for the production of low-cost enzymes. Hägg et al. [41] warned that the loss of plasmid vectors during bacterial cell division, leading to an increasing proportion of plasmid-free cells during growth, is a major industrial problem that results in reduced product yields and increased production costs during large-scale cultivation. It is worth noting that the seed train is often neglected in models of microbial processes.
Fermentation section
The effect of the biomass concentration at the end of the fermentation and the effect of the recombinant protein (rEnzyme) content are presented in Fig. 5a. Clearly, both variables have a dramatic influence on enzyme cost; the case where the lowest biomass is coupled with the lowest rEnzyme content (1926 US$/kg) and the case where the highest biomass is coupled with the highest rEnzyme content (135 US$/kg) are 14 orders of magnitude apart. These results confirm the importance that has generally been ascribed to the volumetric productivity of fermentation processes, which is defined as the mass of the product at the end of the process divided by the final broth volume and the duration of fermentation. To better visualize the effect of this parameter, the biomass concentration and rEnzyme content data were combined and converted into volumetric productivity and plotted against enzyme cost, as shown in Fig. 5b. The chart shows that the enzyme cost does indeed decrease rapidly with volumetric productivity, and the data are very well approximated by a power law in which the cost is inversely proportional to the volumetric productivity. However, it is well known that rEnzyme content may negatively correlate with biomass concentration as a result of the so-called metabolic burden of recombinant protein synthesis and as a result of the toxic properties of inducer molecules such as IPTG [28]. Consequently, these findings indicate the need for a better understanding of the tradeoffs involved in recombinant protein production and, in particular, for experimentally identifying optimum induction conditions and the final biomass concentration range.
A parameter frequently overlooked in techno-economic analyses of bioprocesses is the material of the fermenter, which must resist the frequently corrosive products and by-products of fermentation. Grade 316 stainless steel is considered to be the standard bioreactor material for most biotechnological processes [42]. In fact, grade 316 steel was the material used in the simulations of fungal cocktail production performed by Humbird et al. [2]. However, the authors proposed the use of a lower grade of stainless steel, 304, for a Zymomonas fermenter. Furthermore, carbon steel has a precedent in the American corn ethanol industry [2] and the Brazilian sugarcane ethanol industry [43]. Thus, simulations were conducted assuming that the fermenter and seed fermenters were made of stainless steel of higher grade (SS316), stainless steel of lower grade (SS304), or a carbon-steel alloy (CS); the simulations were conducted using cost models from SuperPro Designer. The results are shown in Fig. 5c. The use of SS304 had a negligible effect on the enzyme cost, whereas the substitution of carbon steel for stainless steel was somewhat significant, decreasing the cost from 316 to 292 US$/kg on a process scale of 100 m3. However, these results do not account for possible requirements for corrosion prevention, such as the application of special coatings [43] or the need to design a thicker fermenter [2]. Therefore, the replacement of SS316 with less expensive materials is not warranted by these results.
Downstream processing
Despite being extremely streamlined, the downstream section accounts for nearly half of the enzyme cost, as discussed above. This can be partially attributed to the contribution of the downstream section to the facility-dependent cost (43% of the total). At this point, the cytoplasmic production of proteins by E. coli introduces a significant, but not critical, increase in costs.
The replacement of centrifugation by microfiltration was also evaluated. The choice between these unit operations is commonly encountered in the biotechnology industry, and although centrifugation is generally considered to be more cost effective at smaller scales [44], the choice must be evaluated on a case-by-case basis. The results presented in Fig. 5d indicate that centrifugation is the best option to separate cell debris at both the baseline scale (100 m3) and the smaller (50 m3) scale, whereas microfiltration was less costly at the larger scale (150 m3). Nevertheless, the difference in cost between these two separation methods was negligible at the smaller scale and quite modest (approximately 5%) at the largest scale.
Alternative downstream operations, such as precipitation using ethanol, might be used to concentrate enzymes from the clarified lysate [45]. However, additional unit operations for precipitate recovery (centrifugation or filtration) and resuspension have to be considered. Aqueous-two phase systems (ATPS) have also been effectively used for separation and purification of industrial proteins. An elegant review on this topic was published by Ansejo and Andrews [46]. The production and purification of chymosin from recombinant Aspergillus supernatant is the most successful industrial application of this technology. Silvério et al. [47] studied the separation and purification of laccase from a complex fermented medium using an ATPS system with a thermo-separating polymer. Despite the possible recovery and reutilization of the polymer, a large loss of activity was observed (88%) when compared with the classical PEG-Salt systems. In general, precipitation and ATPS are adequate if some increase in the purity level of the protein is needed [33]. Although a more comprehensive study of the possible downstream process designs would be very interesting, here we have focused on the most common downstream process configurations described in similar studies, such as those listed in the Additional file 3: Table S7 [3, 5, 48].
Finally, an important factor that can be attributed to the low cost of fungal cellulases is the simplification of the downstream process as a result of the extracellular secretion of the enzymes. The use of cheaper unit operations for cell removal, such as the vacuum drum filter (well suited to fungal biomass separation), may also contribute to the reduced fungal enzyme cost. Although we have no information on β-glucosidase secretion by E. coli, it was interesting to simulate scenarios in which this enzyme could be secreted at the same expression levels as those assumed for intracellular expression. In these cases, centrifugation was employed to separate the cells from the liquid phase containing the enzyme. The impact of enzyme secretion on cost reduction was confirmed because of the simplification of the downstream process (Table S8, Additional file: 3). Moreover, the cost of the enzyme was strongly dependent on the solids content of the sludge. Considering a solid content of 580 g/L (~ 24% of dry cell weight) in the sludge, an enzyme cost reduction of 9% was achieved in relation to the baseline scenario.
Searching for optimized processes
On the search for a better understanding of the factors contributing to the technical and economic feasibility of the process under study, we generated a comprehensive set of simulated scenarios of recombinant BGL production, by varying many of the process characteristics and parameters discussed so far, one by one. The results are compiled in Additional file 3: Table S8, which shows the individual impact on enzyme cost (relatively to the baseline scenario) from each parameter. Additional simulated scenarios combining changes in several parameters and process conditions can also be seen in Additional file 3: Table S8.
As expected, enzyme titer presented the most important impact on cost. The reduction of the cost of glucose, the reduction of the amount of IPTG added, and the use of less expensive bioreactor materials (carbon steel) also proved important (11, 9, and 10% of cost reduction, respectively) (rows #2, and 4). A large cost reduction was also achieved when combining the factors presented above and eliminating unit operations used to concentrate and stabilize the enzyme, thus strengthening the decision of on-site production of enzymes in the context of biorefineries. In this case, a production cost of 63 US$/kg of enzyme was found (row #15). Finally, the last (and best) scenario presented in Table S8 shows that enzyme secretion, combined with the previous efforts to optimize the protein production process, could ultimately reduce the enzyme cost to 37 US$/kg (row #25). Although still higher than the values found in some published studies for fungal cellulases, this value indicates that, with improvements on the enzyme expression level and adequate choices in process design, enzyme costs in the range of 40–70 US$/kg could be reached using recombinant E. coli.