Journal of Biotechnology 59 (1997) 19 – 23 Minireview Biochemical engineering and industry Anders Gram No6o Nordisk A/S, No6o Alle 2, DK-2880 Bags6aerd, Denmark Received 10 October 1996; received in revised form 7 January 1997; accepted 25 July 1997 Abstract The use of biochemical engineering in an industrial setting is presented by examples from the author’s own field of expertise: process development for industrial enzymes. Examples given cover the fields bioreactor performance, measurement and control on fermentation, downstream operations and development and optimization of production strains. The use of engineering principles is placed in the industrial context of a high focus on development speed, fast product turnover and competing ‘technologies’ such as empirical testing, parallel conclusions. Finally the overall applicability of engineering principles is reviewed in light of the industrial aspects of process development: existing production equipment, production facilities of varying age, choice of expression system and overall product economy. © 1997 Elsevier Science B.V. Keywords: Bioprocess development; Optimization; Enzyme production; Mass transfer; k1a; Regime analysis; Flux analysis; Statistical process control; Metabolic engineering; Downstream operations 1. Introduction It is in itself interesting that the use of biochemical engineering principles in industry is the title of a plenary lecture at the first European Symposium on Biochemical Engineering Science, and that a representative from the high-volume, lowcost end of the biotech industry has been asked to expand on the subject. It reflects the position of the research in biochemical engineering as an uncommonly collaborative effort between academic and private research. This position has come about as a consequence of the very fast development of biotechnological tools in the wake of genetic engineering and the very fast growing recognition by both established biotechnological industry and others, such as the pharmaceutical, chemical and food industries, that genetic engineering and it’s derived tools would have a tremendous impact on the basic production processes and the development of new products (van Suijdam and Roels, 1994). In hindsight it is very 0168-1656/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved. PII S 0 1 6 8 - 1 6 5 6 ( 9 7 ) 0 0 1 6 1 - 2 20 A. Gram / Journal of Biotechnology 59 (1997) 19–23 fortunate that both academia and industry jumped so fast and furiously into this field of research, the scientific level of research in both areas is high and increasing and the results achieved so far ensure a continued favorable investment climate in academia and industry. The results could not have been reached without the fruitful synergies between the two research areas. But the research field is maturing in these years and it becomes increasingly important to formulate common strategies for future research to ensure that the results achieved continue to be seen as value adding relative to the investment by government and private enterprise. If common strategies are not formulated and pursued the research field will very likely end up suffering from ‘volume sickness’, a common disease in industry when an industrial success is pursued into too many diverse fields. The result is invariably that the costs skyrocket and the results materialize less quickly. When the absolute size of the costs is large enough to attract attention, they are slashed regardless of the benefits in the pipeline. This paper will deal with one of the major weaknesses in biochemical research, which is also one of the major remaining opportunities. The synergy between molecular biology and biochemistry on one side and chemical engineering on the other is still not fully utilized. We may be starting to speak the same language, but we still have a way to go before the full advantage is seen. The necessity of ensuring a steady flow of applicable results from research is highly recognized in industry. Not so much so in the academic world, at least not on the same short time horizon as in industry. But when a research field is strongly dependent on collaboration the demand for results merge into a common paradigm for the two research areas. This fact needs to be taken into account when planning a strategy for future research. In the following a series of examples of the application of biochemical engineering science at Novo Nordisk will be given. Finally a suggestion for a common strategy to tie together biochemical engineering research will be given. 2. Bioreactor performance; aeration and agitation It is natural for a fermentation based company to work seriously with bioreactor performance. We have done so for many years and we have to admit that the concrete results have been limited. Consequently, we are investigating the underlying parameters thoroughly so that we are able to build a model or a skeleton of understanding into which we then put specific result. The benefit of this approach as compared to many previous attempts to increase mass transfer, mixing and heat transfer capacity at the same time in totally new bioreactor designs is that it is not only a tool in its own right, but also an important tool in allowing us to synthesize experience in this field to manageable single parameters. In order to separate mass transfer capacity from mixing effects, cooling capacity and process variation, we have implemented a specific measurement of k1a (oxygen transfer capacity) which we have used in all of our laboratory, pilot plant and production scales under different sets of parameters. The resulting overview is of course a valuable tool in up- and downscaling, it allows us to determine much more accurately the aeration and power input needed for a given process at a given scale. It also allows us to test new equipment at any scale and predict the performance of this new equipment at other scales. In fact in model media, we find approximately 25% better k1a at the same power input with the Scaba 6SRGT impellers than with Rushton turbines. The next step is, of course, to see the performance of these new impellers in a real process. Our preliminary conclusion is a clear tendency for the Scaba’s to deliver a better k1a in the last part of highly viscous Aspergillus fermentation, where the viscosity is peaking. We still need to work with the process to establish if the productivity can be improved, but if and when we run into problems with that, we know that we do not have to worry about the k1a. Consequently we can focus on other limitations such as mixing time, heat transfer, etc. A. Gram / Journal of Biotechnology 59 (1997) 19–23 3. Regime and flux analysis When analyzing a fermentation process, a critical question is ‘which partial process is limiting?’ Substrate components, biotransformation, mixing time, oxygen transfer, or maybe cooling. The answer to this question points to the optimal route for improving the process economy. At Novo Nordisk, process development has, to a large extend, used trial and error or to use a finer word, ‘experience’ to find which of these limitations we were dealing with. This has led us down some blind alleys and we have had to backtrack on some occasions, and it has meant that we have a series of processes which are rather different, because they are a result of trial and error. An empirical development strategy also gives poor documentation, not because of lack of research reports as such, but because of the lack of a framework to describe the development in a rational way. But we have harvested results fast and used a minimum of resources in getting competitive production economy. Recently, however, we have begun to use bioengineering tools to generate a framework for our process development, thus enabling a faster shift between products expressed in the same host and a more solid platform for the development of fermentation processes that resemble each other. Flux analysis has been much promoted in the literature as an important tool in the optimization of fermentation processes. The basic idea is to establish a quantitative picture of all of the major metabolic fluxes in the cell by analyzing the uptake and secretion of metabolites during the course of a fermentation (Holms, 1986). Our experience is that with enzyme production processes, a flux analysis as such, is of little value because it relates too specifically to the one process being scrutinized. Even more importantly, the results that make the large difference are found in the initial phases of the flux analysis, where the overall skeleton understanding of the process fluxes is generated. Regime analysis aims at determining the rate limiting mechanisms in large scale processes and thus provide important understanding of the ‘scale-down’ of processes. Further on, the knowl- 21 edge that, for example, the oxygen transfer rate is, in fact, the rate limiting mechanism becomes an important tool in reactor design (Kossen and Oosterhuis, 1985). We have performed a regime analysis on one of our Aspergillus fermentations at production scale. From the analysis it is evident that the dominating characteristic time is mixing time, leading to several related problems, e.g. Substrate consumption is fast and it is likely that there are substrate gradients and consequently starvation in large regions of the tank. Gas residence is short, the available oxygen is not consumed efficiently (this is also a consequence of substrate depletion or starvation). The consequence of these findings is that the scale down of this process is difficult, one has to simulate the gradients to get a representative small scale process. A further consequence is to investigate thoroughly the mixing patterns in the bioreactor with the aim of improving the impeller design, the morphology of the fungus or the rheology of the medium to decrease mixing times. 4. Measurement and control; statistical process control When a fermentation is running in pilot or production plants, data from 10–15 online measurement devices is collected. Sampling time ranges from 10 s to 30 min. A long fedbatch fermentation will result in more than 100 000 data points. At Novo Nordisk, the large products are produced in more than 300 single batches every year For each batch there are several discrete data points as well, e.g. batch numbers on raw materials, data from propagation steps, etc. On top of that comes traditional chemometrics, e.g. IR and fluorescence spectroscopy to further characterize the individual runs. The positive problem is how to deal with this wealth of data. The traditional answer is that process operators and supervisors occasionally look at a screen with the online data for the batch that is running. When looking at a complex picture we can concentrate on 3–4 elements at the same time. When remembering sets of complex 22 A. Gram / Journal of Biotechnology 59 (1997) 19–23 pictures, we can hold 6 – 7 images in the human RAM at the same time. The experienced operator or supervisor will have decided which 3 – 4 parameters are most important to keep track of, and in an unusual situation, he will be able to draw on 6 – 7 previous batches to explain the occurrence and decide on appropriate action. This is pretty good, but it is not enough. Consequently we are looking at ways to better utilize the huge experience which is being collected. Currently we are focusing on statistical tools, for example PLS (projection to latent structures) (McGregor et al. 1994). All the data, for example, for 50 batches are collected in a 3-dimensional matrix. The matrix is unfolded for ease of handling and related to a matrix describing one or more quality variables (e.g. final product concentration, final yield or final product economy) for the same batches. The outcome is a model that, based on available variables, can predict the quality of the batch under surveillance. This model has immediate uses: In the control room, the screen will at all times show a best prediction of the quality of a batch in process. This will be a great help for the operators allowing them to concentrate on the batches with the most severe problems. The screen will also be able to show 2 – 3 derived variables (latent structures) for a given batch. These variables will be conglomerates of several online variables, and we know from the analysis that they are the most important for the final outcome. With a little bit of experience, the operator learns which physical actions to take to correct the derived variables. One may, for example, be dominated by agitation and aeration and the operator can increase the aeration rate and in a few hours see the derived variable back in control and the predicted outcome of the batch improving. This last part also allows us to do a causal analysis on the poorly performing batches. We can see which of the most important variables is the hardest to control and from that concentrate our process development on the underlying variables. 5. Metabolic engineering; isopenicillin N synthase When working with the biosynthetic pathway for penicillin and from analysis of metabolic flux it was possible to hypothesize that isopenicillin N synthase (or cyclase) activity was a limiting step (Nielsen, 1994). The gene encoding for the cyclase was put under control of the same promoter as the preceding step in the penicillin synthesis pathway, the ACV synthetase. We found that, indeed, the resulting strain produced higher levels of cyclase and that the penicillin production started earlier and proceeded to a higher end titer than the mother strain. There are many other examples available in the open literature, the most obvious being the work by several groups on amino acid production. Very significant productivity improvements have been achieved (Krämer, 1996). For Novo Nordisk, the primary target is obviously the expression of proteins, specifically enzymes. Metabolic engineering to increase specific production of proteins is a difficult discipline, and as it has been hinted above, we often find that the bottleneck for productivity is not the organism itself but our understanding of the needs of the organism. In other words, many of the enzyme production processes are not as well understood and optimized as the penicillin and amino acid production processes. We are, however, confident that optimization of the production organism in concert with the optimization of process and equipment is the optimal way of improving our processes. Metabolic engineering is seen as an increasingly important tool in the optimization of production organisms. 6. Downstream operations; drum filtration Fermentation and microbiology dominate the examples of use of engineering principles. This reflects very well the focus that Novo Nordisk has had and which is also predominant in academia. But when looking at the cost structure of enzyme production it should probably be revised. The full manufacturing cost of an enzyme is typically split as follows: 25% Fermentation 25% Recovery 25% Formulation 10% Packaging 15% Sales costs. A. Gram / Journal of Biotechnology 59 (1997) 19–23 The 25% recovery cost consists of many items, prominently: recovery yield, raw material usage, waste streams and manning. These important features are interconnected in a very complex way. A model describing this interconnection would be very valuable, but would also have to include more detailed models of the unit operations employed in a given operation. This is what slows down progress in this field. The unit operations that we work with are not described to the level necessary for a plausible model to be developed of their performance on the product of industrial fermentation. We are, however, working with some of the unit operations, e.g. vacuum drum filtration, were we have generated a macroscopic model of the capacity and recovery yield of a filter as a function of the controllable variables. These include rotation speed of the drum, precoat thickness, knife increment per rotation. But characteristically, the model does not include important features on the biological side, e.g. specific or unspecific binding of the enzyme to the filter precoat material, the flow effects of final medium rheology or the morphology of the production organism. The process engineers struggle to understand the consequences of enzyme chemistry, antifoam, substrate components, salts, pH, temperature, etc. on complex results such as drum filter capacity. In conclusion, there is a huge potential for increased understanding of the interaction between biochemistry and stainless steel for all the relevant unit operations in enzyme production. 7. Conclusion There are still huge gaps in our understanding of the synergies between biology/biochemistry and stainless steel. All the model work, be it black-box . 23 or structured, should be fitted into a common framework or skeleton so that it is more readily applicable. This will lead to much better work because the hypotheses of the models can be verified or discarded rapidly. If we can do this, then we will also see engineering tools being used more and more over the coming years, because the benefits will match the efforts. Industry will be able to get more products on the market and create totally new biochemical products, because production economy will be less of an obstacle. Academia will get valuable insights because of the rapid testing of hypotheses and will be able to formulate new areas of research which will be supported by industry. The quest is for understanding the complexities. The treasure chest holds faster product development and a strengthened biochemical industry through better process economy. It also holds a continued fruitful collaboration between industry and academia in biochemical engineering. References Holms, W. H., 1986. The central metabolic pathways of Eschericia coli: relationship between flux and control at a branch point, efficiency of conversion to biomass, and excretion of acetate. Curr. Top. Cell Regul. 28, 69 – 105. Kossen, N.W.F., Oosterhuis, N.M.G., 1985. Modelling and scaling-up of bioreactors. In: Brauer, H. (Ed.), Biotechnology, vol. 2. VCH, Weinheim. Krämer, R., 1996. Minireview, Genetic and physiological approaches for the production of amino acids. J. Biotechnol. 45, 1 – 21. McGregor, J.F., Jaeckle, C., Kipparides, C., Koutoudi, M., 1994. Process monitoring and diagnosis by multiblock PLS methods. AIChE J. 40 (5), 826 – 838. Nielsen, J., 1994. Physiological engineering aspects of Penicillium chrysogenum PhD Thesis, DTU. van Suijdam, J.C., Roels, J.A., 1994. Stratigies of biotechnological industries, biotechnology in the nineties: Back to business. ECB6: Proc. 6th European Congress on Biotechnology. Elsevier, Amsterdam, pp. 51 – 56.