Cost of Goods Modeling and Quality by Design for Developing Cost-Effective Processes

June 1, 2010

Combine cost analyses with QbD to improve operations and lower costs.


Increases in recombinant protein titers expressed in mammalian cells and cell culture process volumes have shifted the economic paradigm from upstream processing to downstream processing for the manufacturing of recombinant proteins, especially monoclonal antibodies (MAbs). Moreover, the pressure exerted by authorities on the healthcare industry to decrease the costs associated with treatments, and the recent entry of biosimilars into the market, have forced the biopharmaceutical community to find new solutions and strategies to reduce production costs. MAb process development currently is focused on the optimization of yield and throughput to deliver cost-effective processes. In this article, we describe the use of a cost of goods model in combination with a Design of Experiment approach for optimizing the capture step by affinity chromatography. The most economical operating conditions identified through this approach imply sacrificing some protein without affecting the quality of the final product.

The market for therapeutic proteins, particularly monoclonal antibodies (MAbs), is rapidly expanding. Indeed, it represents the industry segment with the highest growth rate over the last decade.1,2,3 To meet the increasing market demand, the biopharmaceutical industry is looking for new strategies to increase bioprocessing productivity while minimizing production costs. Bioprocesses have to be carefully fine-tuned to ensure the consistent quality of the material produced, and initiatives such as Quality by Design (QbD) should result in the implementation of robust and flexible processes from the start. Additionally, biotechnology companies must develop processes that are cost effective. The pressure from competition—molecules being approved for the same indications, patent expiry, and the arrival of biosimilars—will put increased pressure on sale prices and as a result, the minimization of production costs will be essential for maintaining economically viable products.1,4,5,6

Merck Serono

Understanding manufacturing process costs is not trivial; many important factors are not directly related to the process itself—plant capacity, equipment depreciation, allocation, and other fixed costs—can make cost estimation extremely difficult.7 For an accurate evaluation of expenses, a detailed process description with facility and equipment depreciation is needed. An exhaustive process simulation tool such as SuperPro Designer (Intelligen, Scotch Plains, NJ) can be used to model bioprocesses. This modeling includes a detailed description of the process steps, from vial thawing to final drug substance, with all related costs (direct and indirect). It is then possible to accurately estimate the final unit cost of the product generated at manufacturing scale.

This article describes the successful combination of a cost of goods (COGs) model (using SuperPro Designer) with a Design of Experiments (DOE) approach to reduce the cost associated with the affinity chromatography step for the capture of a MAb.

Figure 1


SuperPro Designer is a process simulation tool developed for modeling industrial recombinant protein production processes. SuperPro Designer takes into account all the process and subprocess steps and their associated costs. Additionally, fixed costs related to the infrastructure, such as building and equipment, also can be incorporated. The resulting COG models generated allow the user to predict the production costs of processes—performed at manufacturing scale—from cell thawing to final drug substance bulk.

Table 1. Cost of goods estimation

The MAb process modeled in this case study (Figure 1) was a typical three-step platform process adapted to fit into a facility with a maximum cell culture capacity of 120,000 L, a protein titer of 2 g/L, an overall yield of 80% on the downstream process (DSP), and an annual MAb throughput of 4,500 kg. The facility described by the model was assumed to be an entirely new plant used at full capacity. For simplicity, the upstream process (USP)—from inoculum to clarified harvest—was lumped into a single unit operation. To accurately calculate the upstream costs, the entire USP was modeled separately, and a cost per liter of clarified harvest was determined. This harvest cost was then used as an input for the DSP process calculations. The different cost categories used for calculations are listed in Table 1.

Figure 2

A more complex model also was developed to account for the progressive degradation of the capacity of the resin when increasing the number of cycles performed on the column, a well-documented phenomenon for Protein A affinity chromatography steps. This additional level of complexity, however, does not provide a better illustration of the concepts proposed in this case study.

Figure 3

The output of the MAb platform COG analysis is shown in Figure 2. Facility-associated costs account for almost 35% of the total costs, while raw materials and consumables represent 31% of the entire production costs. A further breakdown of the process-related costs for the different unit operations (Figure 3) shows that DSP accounts for more than 65% of the total production costs. Among the three DSP steps, the capture alone accounts for 25% of the total costs. By further breaking down the costs of this unit operation (Figure 4), you can see that the capture step contribution is driven by consumables, specifically the Protein A resin. Therefore, to reduce a MAb process cost, decreasing the consumption of the Protein A resin, by maximizing loading or increasing resin lifetime, for example, will have the greatest effect.

Figure 4


The Protein A intensification study was addressed by optimizing the loading conditions to minimize process costs. A DOE approach was applied, taking into consideration two main factors:

  • the percentage of the breakthrough curve (BTC%), which is a direct measure of the resin capacity corresponding to the amount of material that can be loaded per gram of resin, and

  • the residence time.

A design comprising 13 experiments was generated using a face-centered composite surface response DOE. The experiments are described in Table 2. For each set of experimental conditions, a production cost per gram of protein was calculated using SuperPro Designer. Then, as a response to the DOE, the difference, expressed in percent, between the calculated cost and the cost associated with a reference loading strategy, was determined. The reference used was loading at 80% of the maximum resin capacity, determined at 5% BTC at 3 min residence time.

Table 2. Design of experiments experimental plan

All the experiments were carried out on an AKTA system (GE Healthcare, Chalfont St. Giles, UK) using a MabSelect Extra resin (GE Healthcare) and purified protein. The BTC% was defined as the percentage of the protein concentration measured by the inline UV probe at the outlet of the column over the protein concentration of the initial feed, for a given residence time as described in Figure 5. For each experimental condition, the loaded volume, the loading flow rate, and the step yield were measured or determined and then entered into the COG model to calculate the resulting process costs.

Figure 5

The most important parameter for capture is the process capacity, which is directly linked to the maximum resin capacity for a given residence time. In other words, the greater the BTC% value, the higher the process capacity (mass of protein processed on 1 L of resin), but also the greater the mass of protein lost in the flow-through. Therefore a balance must be reached between high process capacity with a lowered amount of Protein A resin needed, and low protein recovery with a lowered amount of protein produced per batch.

The output of the study is summarized in Figure 6 and describes the DOE results modeled and analyzed using the Minitab 15.1 software. An R-sq value (the goodness of fit) of 90.6% was obtained, indicating the validity of the model. The P-values (the probability of obtaining a test statistic at least as extreme as the one that was actually observed, assuming that the null hypothesis is true) for BTC% and residence time were respectively 0.049 and 0.02, lower than the typically chosen α-level cutoff value of 0.05, indicating that both factors can significantly affect the process costs. Figure 6 shows that minimal costs, or optimal loading conditions, are obtained for a BTC between 1 and 8% and for residence time values between 3.5 and 5 min. Up to 5% of process cost reduction compared to the reference loading conditions (i.e., loading the resin at 80% of the maximum resin capacity as determined at 5% BTC) could be obtained, with a limited loss of product in the flow-through, but still assuring a step yield >95%, thus confirming the predictions made by Kelley.8

Figure 6

The DOE results are a contribution of different categories of costs, as shown in Figure 7. Facility- and raw materials-related costs have the strongest impact, accounting for almost 50% of the total process costs (as discussed above) and both tend to minimize the costs when decreasing the BTC% to maximize the product yield. On the other hand, the consumables-related cost, which is directly proportional to the Protein A resin cost, has an opposite effect: the costs are minimized for high BTC%. This is because at higher BTC% values, higher volumes of loading material can be processed with the same amount of resin, increasing process capacity.

Figure 7


Current publications often focus on the comparison between USP and DSP costs, as a function of the harvest titers.1 This focus is linked to the recent and dramatic improvements in cell culture processes that have led to an important increase in protein titers. It would be interesting to compare the costs as a function of the production titers variations. To simplify the modeling, we assumed that DSP process capacity and yield do not depend on the protein titer in the harvest material. The evolution of process costs with two additional product titers, 0.5 and 6 g/L, were evaluated with our model. The results are shown in Figure 8. As expected, the larger the titer, the larger the BTC% needed to minimize costs. At low protein titers (0.5 g/L), MAb production is more expensive and DSP is not considered a bottleneck. In this situation, maximizing DSP yields generally is the most effective strategy. In fact, when comparing the minimal costs obtained by the DOE analysis at this productivity with the respective reference run (loading at 80% of the maximum resin capacity), no savings were possible, indicating that yield is the driving factor affecting process costs. On the other hand, when high USP productivity is considered (6 g/L), an optimal point is found between 6 and 15% BTC, indicating that savings of up to 7% can expected because of the increased process capacity, even if some product is lost in the flow-through material. At this productivity level, producing the protein is less expensive, saving Protein A resin. Overloading the column while sacrificing product has a significant effect on process costs.

Figure 8


The lifespan of Protein A affinity resins is limited to a few hundred cycles. This lifespan directly affects the cost of the capture, because the higher the number of cycles one can perform on the resin, the less expensive the capture cost per resin cycle will be. Both the lifespan and the titer can easily be linked together by defining a cost per liter of resin and per cycle. Therefore, one way to reduce the resin price would be to maximize the lifespan of the Protein A. The expiration of Protein A patents also should reduce the resin price because it will open up the market to more Protein A vendors and reduce the resin price.

Because this study was focused on Protein A capture costs, it was logical to see how the Protein A resin price would influence the process costs. A sensitivity analysis on resin pricing was then generated using our COG model for a MAb concentration of 2 g/L. Three additional scenarios with resin prices of respectively twice, half, and one fourth the initial price were created. For each scenario, the same DOE approach (discussed above) was applied. The resulting costs of each scenario were then compared to the reference loading condition calculated specifically for each scenario: 2 g/L harvest is loaded at 80% of the maximum resin capacity, using a resin lifespan of 100 cycles and by varying the price of resin according to the specific scenario. The obtained costs, expressed in percentage of the reference loading condition, were then entered into our DOE and the results modeled and analyzed by Minitab. The R-sq values obtained were all >90%, indicating the validity of the model. As for the previous models, all the P values for BTC% and residence time for each set of conditions were below the selected α-level cutoff of 0.05, indicating that lifespan and titer significantly can affect the process costs, as expected. In Figure 9, the DOE output for each scenario is illustrated. For high resin prices (Figure 9A), a cost saving of >7.5% can be obtained by overloading the column (higher volumes can be processed for the same amount of resin) while some product is lost. However, for lower resin prices, half to ¼ of the initial price, the trend is the opposite (Figures 9B and 9C); no significant cost reduction was obtained by overloading the column because of the very low price of the resin compared to the protein costs. Therefore, maximizing yield would be more beneficial than maximizing resin use.

Figure 9

A similar conclusion can be drawn when resin lifespan is considered. In this case, lower maximum resin cycles can be compared to the high-price Protein A scenario, and higher maximum resin cycles to low-price Protein A resin. However, as mentioned earlier, the resin capacity also is a function of the number of cycles performed on the resin, so the capacity cannot be considered a constant. A more accurate experimental design considering this dependence has not yet been performed, and different assumptions must be considered for the cost calculations.


Carrying out a cost analysis of various protein production processes is becoming a necessary tool in the biotechnology industry because process development is not only concerned with traditional targets such as recovery, robustness, and product quality, but also is progressively influenced by economics.9 Sources of potential cost savings must be identified and implemented as soon as possible during process development to be competitive and beneficial at manufacturing scales. When applied with a QbD approach, a cost of goods analysis can help optimize process steps with respect to production costs, as in the case of the MAb capture presented in this article, while maintaining the required product quality. Finally, different strategies can be compared in terms of production costs and alternative unit operations aimed at improving processes and reducing costs can be better evaluated.

Matteo D. Costioli is a bioprocess and innovation downstream process specialist, Christine Mitchell-Logean is the head of bioprocess improvement, and Hervé Broly is the head of biotech process sciences and the vice president, all at the Center of Expertise, Merck Serono SA, Fenil-sur-Corsier, Switzerland, +41 21 923 2181, matteo.costioli@merckserono.netClémentine Guillemot-Potelle is an engineer at the Ecole Polytechnique and AgroParisTech, Thiverval-Grignon, France.


1. Farid S. Economics drivers and trade-offs in antibody purification processes. Supplement to BioPharm Int.: Downstream Processing 2010: Embracing Innovation. 2009 Oct.

2. Kelley B. Very large scale monoclonal antibody purification: the case for conventional unit operations. Biotechnol Prog. 2007;23(5):995–1008.

3. Farid S. Process economic drivers in industrial monoclonal antibody manufacture, In: Gottschalk U, editor. Process scale purification of antibodies, John Wiley & Sons. 2009. p. 239–77.

4. Suresh P, Basu PK. Improving pharmaceutical product development and manufacturing: impact on cost of drug development and cost of goods sold of pharmaceuticals. J Pharm Innov. 2008:3:175–87.

5. Basu P, Joglekar G, Rai S, Suresh P, Vernon J. Analysis of manufacturing costs in pharmaceutical companies. J Pharm Innov. 2008;3:30–40.

6. Rathore AS, Latham P, Levine H, Curling J, Kaltenbrunner O. Costing issues in the production of biopharmaceuticals. BioPharm Int. 2004;14(2).

7. Farid S. Process economics of industrial monoclonal antibody manufacture. J Chromatogr B. 2007;848:8–18.

8. Kelley B, Bender J. Monoclonal Antibody Purification at a Crossroads. IBC conference, Bioprocess international; 2008; Apr 22–23 Vienna, Austria.

9. Rathore AS, Karpen M. Production cost analysis: economic analysis as a tool for process development: harvest of a high cell-density fermentation. BioPharm Int. 2006;19(11):56–63.