Professional Documents
Culture Documents
Abstract: Uncertainty and variability affect economic and success of a process design and the analysis of existing
environmental performance in the production of biotech- processes. Thus, a methodology is needed that combines
nology and pharmaceutical products. However, commer- standard process simulation software with uncertainty anal-
cial process simulation software typically provides analysis
that assumes deterministic rather than stochastic process ysis. In this article, we use a well-characterized process, the
parameters and thus is not capable of dealing with the production of penicillin, to illustrate how this goal might
complexities created by variance that arise in the decision- be accomplished.
making process. Using the production of penicillin V as a Penicillins produced by Penicillium chrysogenum are
case study, this article shows how uncertainty can be still among the most important antibiotics. Penicillins be-
quantified and evaluated. The first step is construction of
a process model, as well as analysis of its cost structure long to a family of hydrophobic h-lactams. Each contains
and environmental impact. The second step is identifica- a different acyl side chain attached by an amide link-
tion of uncertain variables and determination of their proba- age to the amino group of the penicillin nucleus, the
bility distributions based on available process and literature 6-aminopenicillanic acid. Penicillin G and penicillin V
data. Finally, Monte Carlo simulations are run to see how are the main commercial penicillins. Most of the pen-
these uncertainties propagate through the model and af-
fect key economic and environmental outcomes. Thus, the icillin V (phenoxymethylpenicillin) is converted to 6-
overall variation of these objective functions are quantified, aminopenicillanic acid (6-APA), which in turn is used
the technical, supply chain, and market parameters that to make amoxicillin and ampicillin (McCoy, 2000). In
contribute most to the existing variance are identified and addition, penicillin V is used directly as an antibiotic
the differences between economic and ecological eval- (f1,600 tons per year) (Van Nistelrooij et al., 1998)
uation are analyzed. In our case study analysis, we show
that final penicillin and biomass concentrations in the fer- and ranks among the 100 top prescribed drugs in the
menter have the highest contribution to variance for both United States (American Druggist, http://www.rxlist.com/
unit production cost and environmental impact. The pen- top200a.htm, May 2004).
icillin selling price dominates return on investment vari- h-Lactam antibiotics amount to about 60% of the
ance as well as the variance for other revenue-dependent worldwide antibiotics market; this was approximately
parameters. B 2004 Wiley Periodicals, Inc.
$5 billion per year in sales in 1999 (Demain and Elander,
Keywords: Monte Carlo simulation; uncertainty; vari-
ability; penicillin; economic assessment; environmental 1999). The global demand for h-lactams grows by around
assessment 2% annually, mainly because of rising demand in countries
such as China and India (Milmo, 2003). Lowe (2001)
estimates that the world production of penicillin was
INTRODUCTION
65,000 tons in 2001. As a result of large overcapacity
Commercial process simulation software usually provides in the market, penicillin prices have been under continu-
analysis that assumes deterministic process parameters. ous pressure for several years. Prices have fallen signifi-
The software does not consider existing variations of tech- cantly during the last several years from $20/billion units
nical, supply chain, and market parameters that can sig- (BU) during the mid-1990s to $12/BU in 1997 to $9/BU
nificantly alter operating decisions and batch-to-batch in 2000 (McCoy, 2000). As of 2003, the price of peni-
expectations. However, an understanding of these param- cillin G was approximately $11/BU, which is $17–18/kg
eters and their uncertainty is essential for the economic (Milmo, 2003).
Improvements in the penicillin production process result
primarily from genetic-based strain improvements, while
Correspondence to: C. Cooney
the process flowsheet has changed very little (Van
This article includes supplementary material available via the Internet at Nistelrooij et al., 1998). Although some improvement has
http://www.interscience.wiley.com/jpages/0006-3592/suppmat. been realized from refinement in operating conditions,
168 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 90, NO. 2, APRIL 20, 2005
Table I. Parameter values of the fermentation model of penicillin V production.
1997; Lowe, 2001; Van Nistelrooij et al., 1998). Aspen 1997). We chose a facility with 11 fermenters, each with a
Batch Plus and Intelligen’s SuperPro Designer were the volume of 100 m3, optimizing the usage of the downstream
software packages considered for the implementation of the equipment. Penicillin V sodium salt is the final product.
process model. Although both packages are robust simu- The media components (pharmamedia, trace metals,
lation tools, SuperPro Designer was chosen based on the phenoxyacetate; S-102 to S-104) are blended in tank P-1
intuitive relationship between its process representation and sterilized in the continuous heat sterilizer P-4. The glu-
and the spreadsheet model that was constructed for cose solution is prepared in tank P-2. Medium and glucose
Monte Carlo analysis. solution are fed to the fermenter P-7 (glucose solution is
Figure 1 shows the process flow diagram created with fed continuously only during the production phase). The air
the software SuperPro Designer. Fermenters with a total (S-113) is compressed (P-5) and filter sterilized (P-6). The
capacity of 40–200 m3 are used for production (Ohno et al., exhaust air, containing mainly carbon dioxide, is filtered
2002; Lowe, 2001; Falbe and Regnitz, 1999; Perry et al., (P-8) to prevent release of by-products to the environment.
Figure 1. Process flow diagram of the penicillin V production model (SuperPro Designer, version 5.1).
170 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 90, NO. 2, APRIL 20, 2005
to be 108 MJ (30 kWh) electrical energy, 40 kg steam Based on an annual production of 2,090 tons, the unit
and 9 m3 cooling water. However, the demand of steam production costs in the model are $16.2/kg final prod-
is only 5% of that amount. In Ohno et al. (2002), most uct, which equals $10.3/BU. The selling price is set to be
of the steam is needed for batch sterilization. In our mod- $17.3/kg (Milmo, 2003). Hence, the annual revenue is
el, continuous sterilization is used, and it requires much $36 million. Since the cost of operations is $32 million per
less steam. year, this results in EBITDA of $4.0 million. Including
depreciation leads to negative EBIT of $350,000 and
the ROI is negative ( 0.5%). Note that the ROI number
Economic Assessment
assumes a 35% tax rate and no financial leverage for the
The base case model provides an estimate of the costs project (i.e., no interest payments).
involved in penicillin manufacture. Combined with the
current market price of penicillin, these estimates allow us
Environmental Assessment
to calculate a number of financial measures that indicate
the economic value of an investment in penicillin manu- Environmental aspects have become increasingly important
facture. Although in a rigorous capital budgeting process, in pharmaceutical production and should be considered
we would seek to determine the net present value of the together with the economic assessment (DeSimone and
estimated investment, we use the basic financial metrics of Popoff, 1997). The method used, which was recently
earnings before interest and taxes (EBIT), earnings before developed by Biwer and Heinzle (2004), aggregates the
interest, taxes, depreciation, and amortization (EBITDA), whole range of possible environmental impacts to two per-
and return on investment (ROI) in the analyses that follow. formance figures that enable an appropriate comparison
The justification for the use of these metrics is that our with the economic key figures and the examination of
primary objective is illustration of uncertainty analysis how uncertainty affects the environmental impact. They
rather than project valuation. are therefore more suitable for this purpose than complex
In the model, the total purchased equipment costs is methods such as the life cycle assessment.
$9.1 million. Based on data from Peters et al. (2003), it Wastewater is discharged mainly from the extraction
was assumed that the purchased equipment cost is 27.5% step (remaining broth after penicillin removal) and after the
of fixed capital investment (FCI) leading to a FCI of separation of the crystals in the basket centrifugation,
$44.1 million and a total capital investment (TCI) of where a mixture of butyl acetate, acetone, water and some
$51.4 million. The fermenters with $5.5 million dominate impurities are discharged. After a partial removal of the
the purchase costs. All other equipment costs less than butyl acetate (P-27) and acetone (not shown), both waste
$1 million; thus, the fermenters and related equipment dom- streams (S-171, S-173) are assumed to pass to a biological
inate the equipment costs, in agreement with the analysis wastewater treatment plant. Solid waste is produced in the
of Swartz (1979). biomass removal. The only relevant emission is the exhaust
The total operating costs are $33.8 million per year. The air of the fermenter, which includes a large amount of
raw material costs, mainly glucose, phenoxyacetic acid and carbon dioxide (S-117). We have not attempted to assess
butyl acetate (including recycling costs), have the biggest fugitive emissions from the process.
share (37%); this is consistent with results from Lowe Figure 2 shows the EI of the input and output
(2001). They are followed by equipment-dependent costs compounds. On the input side, the media components
(24%) (mainly depreciation and maintenance) and the labor
costs (14%). Utility costs (11%, mainly electricity) and
plant overhead cost (11%) also play an important role,
while laboratory/QC/QA, waste treatment, and consum-
ables (altogether 3%) have only a small impact. Seven
single operating cost parameters capture each by them-
selves more than 2% of the global operating costs. The
fermenter-related costs of glucose (6.3%), phenoxyacetic
acid (12.9%), and electricity for fermenter (2.0%) and
compressor (3.0%) constitute 25% of the total operating
costs. Furthermore, basic labor cost (11.5%), butyl acetate
(9.1%, including recycling cost), and chilled water demand
(3.1%) contribute considerably to the operating cost. Here
we note that the price of glucose and assumed hourly labor
rates play an important role. This explains why today most
penicillin producing plants are located in countries where
sugar and labor costs are low but are capable of supplying a Figure 2. Environmental indices (EI) of the input and output
stable source of energy given the high energy requirements components of the penicillin V production model. [IP/kg P] = index
of the process. points per kg final product.
172 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 90, NO. 2, APRIL 20, 2005
Table III. Parameters used for Monte Carlo simulation and their variation and probability distribution chosen.*
1. Technical Parameters
Yield biomass on 0.45 Internal estimate, based Normal V = 17.5%; min: 0.2 Industry data
glucose (g/g) on fermentation data
Maintenance coefficient 22 Internal estimate, based Normal V = 17.5%; min: 10 Industry data
(mg glucose/g dcw h) on fermentation data
Precursor utilization 92 De Tilly et al., 1982 Normal V = 15.0; 70–100 (min, max) Industry data
efficiency (%)
Final biomass 45.0 Internal estimate, based Normal V = 17.5%; min: 25 Industry data
concentration (g/L) on fermentation data
Final product 63.6 Demain and Elander, Normal V = 10%; 20–100 (min, max) Industry data
concentration (g/L) 1999; Van Nistelrooij
et al., 1998
Aeration rate (vvm) 0.8 Perry et al., 1997; Normal V = 10%; 0.5–1.0 (min, max) Perry et al., 1997;
Lowe, 2001 internal estimate
Agitator power (kW/m3) 2.5 Perry et al., 1997; Normal V = 20%; 1.5–3.5 (min, max) Perry et al., 1997;
Lowe, 2001 internal estimate
Yield downstream recovery (%) 90 Lowe, 2001; Normal Calculated for single step yields
Van Nistelrooij
et al., 1998
Yield biomass removal (%) 97 Internal estimate Normal F2 (SD)
Kj extraction 60 McCabe et al., 2001 Uniform 60–80
Industry data
Yield crystallization (%) 97 Internal estimate Normal F2 (SD)
(overall yield)
Yield basket centrifuge (%) 99 Internal estimate Normal F1 (SD)
Yield fluid bed dryer (%) 99 Internal estimate Normal F1 (SD)
Yield butyl acetate 80 Internal estimate Normal F5 (SD) Internal estimate
recycling (%)
Yield acetone recycling (%) 70 Internal estimate Normal F5 (SD) Internal estimate
2. Supply chain parameters
Price glucose [$/kg] 0.216 USDA Foreign h a = 3.49; h = 1.2; USDA Foreign
Agricultural Service, scale = 29.1 (for a normal Agricultural Service, 2001a
2001a distribution: V = 25%)
Price phenoxyacetic 3.80 Internal estimate; Normal V = F10% Internal estimate
acid ($/kg) supplier data
Electricity cost ($/kWh) 0.0468 U.S. Energy Information Weibull Loc: 4.13; Scale: 0.61; U.S. Energy Information
Administration, 2004b; Shape: 1.96 Administration, 2004b
Peters et al., 2003 (for a normal distr.: V = 6%)
3. Market parameters
Selling price final 17.30 Milmo, 2003 Normal V = F10% Internal estimate
product ($/kg)
used, fermentation or purification conditions). Supply chain The energy costs are dominated by the costs for
and market parameters are not influenced by the process electricity. Mean value and probability distribution for this
conditions and display variance that affects the economics parameter are derived from monthly average prices for the
of the process over an extended period. United States from 1995 to 2003. (Weibull distribution,
The raw material costs account for a large part of the unit values see Table III).
production costs. They are dominated by the costs for The price for penicillin V and penicillin in general has
glucose and phenoxyacetic acid. Therefore, the prices of varied dramatically over the last few years as noted earlier.
these materials are considered in the MCS. The mean As the mean value, the current price stated by Milmo (2003)
glucose price was calculated from monthly average prices is used, and a coefficient of variation of 10% is assumed.
on the world market from 1996 to 2001. Using the ‘‘batch
fit’’ function of Crystal Ball, a h distribution was identified
Best-Case and Worst-Case Scenario
as the best fit (values see Table III). h distributions are
often used to describe empirical data. For phenoxyacetic Based on the parameters chosen, a best-case and a worst-
acid, an average price is chosen that is realistic for the case scenario are calculated. For those parameters where a
annual demand of 1,600 tons. standard deviation is defined, 2 times the standard devi-
Table IV. Objective functions in the base case, worst-case, and best-
case scenario.
174 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 90, NO. 2, APRIL 20, 2005
and has a coefficient of variation of only 2%. As one might
expect, the MCS-AP displays the broadest variation. The
MCS-Pen, which includes substantial variation contributed
by penicillin concentration, is only slightly smaller; the
MCS-TPW distribution lies between those of MCS-S/MP
and MCS-Pen.
176 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 90, NO. 2, APRIL 20, 2005
These results show that the relevant parameters and how
strongly they contribute to the uncertainty differ to some
extent between the economic and environmental indica-
tors. However, the direction of change is the same for all
relevant parameters. The contributions of the variables to
the overall uncertainty indicate the sensitivity of the pro-
cess to these variables. Thus, there are parameters that can
be changed to improve the economic performance without
affecting the environmental performance. While for other
parameters, an economic improvement leads directly to an
environmental improvement. This represents an economic
and environmental (eco-efficiency) win-win scenario that is
contrary to the use of end-of-pipe technologies for en-
vironmental pollution control that lead to additional costs.
178 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 90, NO. 2, APRIL 20, 2005
of UPC and EI. Fermentation parameters such as yield, References
maintenance coefficient, and precursor utilization also
Biwer A, Heinzle E. 2004. Short-cut method for environmental assessment
have a high impact on the variance of the environmental in early phases of process development. J Chem Techn Biotechnol
impact, as well as the recycling rate of the organic sol- 79:597–609.
vents. The production costs are significantly affected by Castells F, Aelion V, Abeliotis K, Petrides D. 1994. Life cycle inventory
downstream yield and raw material costs. The selling analysis of energy loads in chemical processes. In: El-Hawagi M,
price dominates the variation of ROI and other revenue- Petrides D, editors. Pollution prevention via process and product
modifications. New York: AIChE. p 161–167.
dependent parameters. Chang Z, Wei X, Chen J. 2002. Simulated foam separation of butyl acetate
The additional use of an environmental assessment from wastewater discharged by solvent extraction operation in
method that aggregates the environmental performance of penicillin production. Separation Sci Tech 37:981–991.
the process to two indicators allows one to understand Demain A, Elander R. 1999. The h-lactam antibiotics: past, present, and
future. Antonie Leeuwenhoek 75:5–19.
how the uncertainty of the input variables affects not only
DeSimone L, Popoff F. 1997. Eco-efficiency: the business link to sus-
economic performance, but also impact on the environ- tainable development. Cambridge, MA: MIT Press.
ment and the differences between these two metrics. This DeTilly G, Mou D, Cooney C. 1982. Optimization and economics of
approach has been used infrequently so far in the evalua- antibiotic production. In: Smith J, editor. Filamentous fungi. London:
tion of chemical and pharmaceutical processes. Edward Arnold. p 190–209.
The sensitivity analysis shows how important it is to Falbe J, Regnitz M. 1999. Roempp Lexikon der Chemie. Vol 10. CD-Rom.
Stuttgart: Thieme.
define exactly the variance of key input parameters. The Lowe D. 2001. Antibiotics. In: Ratledge C, Kristiansen B, editors. Ba-
threshold for positive EBITDA has been graphically sic biotechnology. Cambridge, UK: Cambridge University Press.
derived and used to show that decreasing selling prices p 349–375.
over the past few years have drastically increased the McCabe W, Smith J, Harriott P. 2001. Unit operations of chemical
pressure for further strain improvements and cost reduction. engineering. New York: McGraw-Hill.
McCoy M. 2000. Antibiotic restructuring follows pricing woes. Chem Eng
We note, however, that the case presented is limited by the News 4:21–25.
fact that the base model is a generalized model of the Milmo S. 2003. Challenges for European antibiotics producers: competi-
penicillin V product process. Depending on the location, tion from China and a soaring euro are just two of the factors making
the cost structure of a manufacturer might vary. business more difficult. Chem Market Rep 263 (24).
This article presents the penicillin V production as a case Ohno M, Otsuka M, Yagisawa M, et al. 2002. Antibiotics. In: Ullmann’s
encyclopedia of industrial chemistry. Weinheim: Wiley-VCH.
study. It proves that combining calculation-based modeling Paradkar A, Jensen S, Mosher R. 1997. Comparative genetics and molec-
with Monte Carlo simulations can be used as a general ular biology of h-lactam biosynthesis. In: Strohl W, editor. Biotech-
methodology for multi-parameter uncertainty analysis. nology of antibiotics. New York: Dekker. p 241–277.
Thus, it goes beyond the usual case of economic assessment Perry R, Green D, Maloney J. 1997. Perry’s chemical engineers’
based on average value analysis that is standard practice handbook. New York: McGraw-Hill.
Peters M, Timmerhaus K, West R. 2003. Plant design and economics for
in process development using commercial simulation soft- chemical engineers. Boston: McGraw-Hill.
ware. The methodology presented is generally applicable Strohl W. 1997. Biotechnology of antibiotics. New York: Dekker.
to the analysis of any process where stochastic variables Strohl W. 1999. Secondary metabolites, antibiotics. In: Flickinger M, Drew
influence both within-batch and between-batch process S, editors. Encyclopedia of bioprocess technology—fermentation, bio-
catalysis, and bioseparation. New York: Wiley. p 2348–2365.
capability and the long-run economic performance upon
Swartz R. 1979. The use of economic analysis of penicillin G manu-
which development decisions are based. facturing costs in establishing priorities for fermentation process im-
provements. Annu Rep Ferment Proc 3:75–110.
The authors thank Intelligen, Inc., for their support and helpful Van Nistelrooij H, Krijgsman J, De Vroom E, Oldenhof C. 1998. Penicillin
discussions. This work was supported by a fellowship within the update: industrial. In: Mateles R, editor. Penicillin: a paradigm for
Postdoctorate Program of the German Academic Exchange Service biotechnology. Chicago: Candida. p 95–102.
(DAAD). Vose D. 2003. Risk analysis. Chichester: Wiley.