Separation and Purication Technology 72 (2010) 3439

Contents lists available at ScienceDirect

Separation and Purication Technology

journal homepage: www.elsevier.com/locate/seppur

Purication performance index and separation cost indicator for experimentally based systematic downstream process development
T. Winkelnkemper, G. Schembecker
TU Dortmund, Department of Biochemical and Chemical Engineering, Laboratory of Plant and Process Design, Emil-Figge-Strasse 70, D-44227 Dortmund, Germany

a r t i c l e

i n f o

a b s t r a c t
Cost-driven downstream process development and optimization is only possible when a complete process concept is either modeled or set up in laboratory scale, which requires comprehensive data or respective time and effort. Before that stage the decisions to apply or discard alternative purication steps are based on practical feasibility, experimental yield and purity improvement, without estimates for their cost-efciency within the process. This article introduces two new parameters to overcome that problem. The purication performance index (PPI) weights the improvement in product purity achieved by a single step and relates it to the boundaries of the process. In this way the purication performance is assessed as a percentage of the required total purication. The separation cost indicator (SCI) combines purication rating and yield losses with cost data. Thereby cost-efciency of single purication steps or step combinations can be estimated without the need of a complete process concept. 2010 Elsevier B.V. All rights reserved.

Article history: Received 31 July 2009 Received in revised form 11 December 2009 Accepted 21 December 2009 Keywords: Process development Cost-estimation Purication rating PPI SCI

1. Introduction Process development and especially conceptual design is regarded as the most important design stage since the major decisions affecting the whole lifecycle of the process are done in this phase. The performance of the process is relied on the decisions made during process development [1]. Especially for pharmaceutical products the downstream process accounts for a major part of the production costs [2]. But even for production of bio-based bulk chemicals the cost of purication can mount up to over 50% [3]. On the other hand time and effort for developing efcient processes also cause signicant costs [4]. Furthermore, especially for biotechnological products the need for shorter time-to-market and the difculties to change an established process due to quality assurance and regulations like GMP lead to suboptimal processes [5]. These conicts motivate continuous research in the eld of systematic process design. Harjo et al. comprehensively present a systematic procedure for manufacturing phytochemicals [6]. Design strategies for, e.g. pro-

Abbreviations: ABx, mixed-mode ion exchange (resin); CF, clearance factor; CHOP, Chinese hamster ovary cell proteins; HIC, hydrophobic interaction (resin); IgG1, immunoglobulin G1; Log PI, logarithmic purication index; PI, purication index; PPI, purication performance index; ProA, Protein A (resin); Q, anionexchange (resin); S, cation-exchange (resin); SCI, separation cost indicator. Corresponding author. Tel.: +49 231 755 2339; fax: +49 231 755 2341. E-mail address: gerhard.schembecker@bci.tu-dortmund.de (G. Schembecker). 1383-5866/$ see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.seppur.2009.12.025

tein purication processes are well reviewed by Nfor et al. [7] who distinguish between heuristic, algorithmic, high throughput experimental and hybrid methods. While heuristics can establish process alternative trees and eliminate some alternatives on a qualitative level with little information [8], further quantitative evaluation of the remaining alternatives is inevitable. These steps can differ either by unit operation, mass separating agent and/or operating conditions. The choice between alternatives always requires rating, ultimately by costs [9]. For chemical processes Gadewar et al. state that a sound operating cost calculation is essential, but only available by comprehensive experimental and model-based investigation in the later phases of conceptual design [10]. Leser et al. introduced an expert system for protein purication based on economic separations coefcients (ESC) [11]. Although only four qualitative cost factors were employed, the employed algorithm allows a costbased design of chromatographic sequences. However, Asenjo et al. later found that using the purity after each separation as selection criterion may lead to better processes [12,13]. Steffens et al. demonstrated cost-driven development of bioprocesses by shortcut modeled efuent concentrations and operating costs of different unit operations [14]. The estimated product value was balanced against expenses, where only yield losses accounted for costs. The MINLP approach of Simeonidis et al. for protein purication with peptide tags, in which the target function is to minimize the number of chromatographic steps [15], is extended by Lienqueo et al. by a cost function [16]. This prot function is the difference between revenue and costs of a series of chromatographic steps.

T. Winkelnkemper, G. Schembecker / Separation and Purication Technology 72 (2010) 3439

35

It arises from the need to balance yields, purities and costs, but Lienqueo nds that the chosen function results in maximum recovery. Groep et al. employed an arctangent correlation between nal purity and revenue and estimated operating costs to be linearly proportional to the amount of material processed in each operation [17]. These approaches allow process synthesis and optimization in a system-wide, rather than unit-by-unit, sense. For a systematic downstream process development the evaluation of alternative process steps with regard to their performance within the total process is essential. The most advanced methods to quantify the inuence of single step performance on the total process costs are proposed by Huenupi et al. and Shene et al. For investigation of operational conditions of aqueous two-phase extraction [18] and ion exchange chromatographic separation [19] a cost function is set up which estimates the inuence of the single step yield, outlet purity and concentration on the operational costs of the total process relative to a base case. That allows, e.g. considering the impact of yield losses on production costs. However, the total process concept is known a priori and necessary to reason all terms in the process-specic function. Even the contribution of each term to the total costs must be estimated and framed in advance, which is questionable in view of the purpose of the method. But the decisions having most signicant consequences on total costs are made in earlier phases of process design [20]. Hence, it is desirable to base these decisions upon costs for the complete process, even before comprehensive process concepts are available. These costs are not directly deducible from investigation of single steps, especially not from few experiments, which focus on technical feasibility, achievable yields and purities. A quantitative relation between purication and cost-efciency of a process step is not yet available. For these reasons, new key performance indicators for rating purication and cost-efciency on basis of single step purity improvement, yield and specic costs are introduced. In contrast to previous indicator-based methods for process design [21] the proposed indicators do not require complete mass and energy balances and can be applied from the beginning of experimental investigation. 2. Theory 2.1. Purication rating Purity is the fraction of the target product in a mixture with contaminants, it is dened as: amount of product Purity = . amount of product + amount of total impurities (1)

Two ways of normalizing purication measures are: 1. Purication index (PIj ): xout,j xin,j xf x0 2. Logarithmic purication index (log PIj ): lg(xout,j ) lg(xin,j ) lg(xf ) lg(x0 ) (6) (5)

Log PI can be obtained as ratio of the logarithmized purication factors of the step and the total process. Though both normalized purication measures range from 0% to 100%, the weighting of single steps within a process can differ considerably. That is illustrated in Fig. 1 showing a downstream process designated to purify a crude mixture (x0 = 0.1%) into a high-purity nal product (xf = 99.9%). The different ways of rating purication are compared using three typical steps within this process. The rst step in the process (A) starts with very low initial purity and results in moderate purity according to a several-fold product concentration. An intermediate step (B) ranging from moderate to high purity represents another part of the process, but fails to achieve the target purity. Usually it takes much effort to reach very high purities [23], so the depicted nal step (C), which already starts from high purity, must still be considered difcult. These steps are chosen arbitrarily, so equal rating is implausible. But according to the considerations above, a balanced weighting should yield that none of these efcient steps is rated much lower than the others. The step ratings by the classical measures purity difference and purication factor as well as by clearance factor (CF) are illustrated on the left side of Fig. 1. PI and log PI rating on the right side are normalized, but very imbalanced over the process, causing the need for a new rating: 3. Purication performance index (PPIj ): tanh1 (2xout,j 1) tanh1 (2xin,j 1) tanh1 (2xf 1) tanh1 (2x0 1)

(7)

The area hyperbolic tangent weighting of the new purication measure PPI appears to be less straightforward, but with Eq. (8) it turns out, that PPI can similarly be obtained as ratio of the logarithmized clearance factors of the step and the total process. tanh1 (2x 1) = 1 ln 2 x 1x (8)

There are two common measures for purication achieved with a single step j: Purity difference : and Purication factor : xout,j xin,j . (3) x = xout,j xin,j (2)

As demonstrated in Fig. 1, the PPI appears to be most balanced over the whole purity range of a downstream process and can thus be used to link the purication performance with the expected effort. 2.2. Cost-estimation The inuence of single purication steps on the total process costs is easily assessable as soon as a complete downstream process concept is set up. The specic process costs (tot ) of a general process with one conversion step (complete upstream) and n purication steps can be calculated from the specic costs for all steps (conversion and pur,1 ,. . .,pur,n ) as: tot mprod = conversion +pur,2 mprod Y1 Ytot mprod Ytot + pur,1 mprod Ytot Ytot , (9)

Another measure is the clearance factor as dened, e.g. by Asenjo et al. [22]: Clearance factor(CF) : xout,j /xin,j (1 xout,j )/(1 xin,j ) (4)

In order to assess the purication of one step as percentage of the total process to be developed, the purity of the initial mixture x0 and the target purity xf must be considered as given boundaries of the downstream process.

+ . . . + pur,n

mprod Yn1

36

T. Winkelnkemper, G. Schembecker / Separation and Purication Technology 72 (2010) 3439

Fig. 1. Different ways of rating three exemplary downstream process steps unrelated and normalized to the total process boundaries (CF: clearance factor).

where conversion is related to the mass of product at the outlet of the conversion step, pur,j to the mass of product at the inlet of purication step j and tot to the mass of nal puried product mprod . In this context specic costs mean the costs per amount of product to be processed in a downstream process unit. Yi is the yield of the downstream process up to the purication step i and is calculated from the step yields of all preceding purication steps Ypur,j .
i n

the purication target of a designated process and leads to a measure of cost-efciency, the separation cost indicator (SCI), that only depends on normalized purication rating, yield and specic costs of the single step j, e.g. with PPIj : Ytot = Yjn = Yj1/PPIj SCIj = tot,pur,j (Yj , PPIj , pur,j , conversion ) = Yj1/PPIj conversion + pur,j 1 Yj1/PPIj 1 Yj (14) (13)

Yi =
j=1

Ypur,j ,

Ytot = Yn =
j=1

Ypur,j ,

(10)

Rating single steps prior to completion of the total process concept is less straightforward, but possible by relating yields and specic costs of a single step to the process boundaries. Thus, the cost-efciency of any single step can be estimated as the specic process costs for a hypothetical process, which only consists of steps with equal purication, yield and specic purication costs. tot,pur,j pur,j = conversion + (1 + Yj + Yj2 + . . . + Yjn ) Ytot Ytot (11)

The SCI-value is a step-specic economic rating in the context of a complete process which is unknown except for the boundary purities. Clearly, the intention of SCI is not to estimate the total process costs on basis of only one step, but to rate single steps in terms of costs per purication scaled up to 100% purication. 3. Application of PPI and SCI In order to demonstrate advantages and new capabilities of PPI and SCI an example process is evaluated. Follman and Farner published a case study about the development of alternative downstream processes for the purication of a CHO-derived recombinant monoclonal IgG1 antibody from fermentation broth [24]. They investigated experimentally several combinations of cationexchange (S), anion-exchange (Q), mixed-mode ion exchange (ABx), hydrophobic interaction (HIC) and other resins in a three-

The geometric series can be transformed to: tot,pur,j =

n conversion pur,j 1 Yj conversion pur,j 1 Ytot + = + (12) Ytot 1 Yj Ytot 1 Yj Ytot Ytot

A normalized purication rating as introduced in Section 2.1 can be used to calculate the number of steps required to reach

T. Winkelnkemper, G. Schembecker / Separation and Purication Technology 72 (2010) 3439

37

Fig. 2. Four different processes for purication of a CHO-derived monoclonal antibody with experimental purity and yield data according to [24].

Table 1 Calculation of specic purication costs for different chromatographic steps. Step Resin Resin pricea [D /l] Eluent consumptionb Specic costs [D /g IgG1 to process] Resinc ProA S Q Abx HIC
a b c d e

Eluentd 0.48 0.48 0.39 0.48 0.63

Non-consumablese 15.00 15.00 15.00 15.00 15.00

Total (pur ) 35.48 17.68 17.59 17.88 18.03

Protein A Sepharose SP-Sepharose fast ow Q-Sepharose fast ow Bakerbond Abx Phenyl-Seph. fast ow

10000 1100 1100 1200 1200

16 16 13 16 21

20.00 2.20 2.20 2.40 2.40

VWR International, online price 12/2009, 50% off for large scale. l eluent/l resin [24]. Assumptions: 50 cycles for each resin [27], load 10 g IgG1/l resin [24]. Assumption: 0.3 D /l costs for buffer preparation [27]. Estimated to be about 50% of downstream costs [28], equally distributed among three chromatographic steps (independent of resin).

step chromatographic purication process and compared the results with a traditional process applying Protein A (ProA). The measured concentrations of Chinese hamster ovary cell proteins (CHOP; in ng impurities/mg antibody) are used in the following for calculating the purities according to Eq. (1): x= 1 1 + CHOP/1000 (15)

Starting from an initial purity of 0.45%, three processes consisting of only non-afnity steps and the Protein A reference process reached the desired target purity of >99.8%, all of these are shown in Fig. 2 together with the purity and yield data provided. Most noticeable are the purity difference between the two alternative rst steps and the high purities after two steps each. A comparison of the PPI values calculated from these purities is shown in Fig. 3. A PPI rating of 57% for the rst step of process I appears to be low as Protein A removes 99.9% of the feed stream impurities. But this rating is correct, because the purity after this step is only 77% and considerable effort is required to achieve the nal target purity. Nevertheless, the performance of the Protein A step is very good and adding only one equally efcient step would lead to the nal product specication. But such an efcient process step does not exist. In fact, a second and a third step are needed for process I, the purication reached by which is rated very bad. As the purication rating of Protein A is much higher than that of the cation-exchange resin applied as a rst step, utilization of Protein A seems to be advisable. But the purication performance must be charged up against costs. Protein A resin is much more expensive than non-afnity resins, so the alternative rst step S (processes IIIV) could still be cost-efcient despite less than half purication performance. Therefore, the specic costs of all steps are estimated in the following enabling calculation of cost-efciency by SCI. Based on costs for production and purication of 175 D /g IgG1 (260 $/g IgG1) for industrial monoclonal antibody manufacturing [25], 55% of which are contributed by the production step [26], and

a total yield of the downstream process of 85% [24], the specic production costs are estimated to be conversion = 80 D /g IgG1. Emanating from equipment, materials and conditions described by Follman and Farner and actual prices for the applied chromatographic resins the specic purication costs for each resin are calculated as summarized in Table 1. With the given yields and cost data the manufacturing costs of all 4 processes can be calculated according to Eq. (9), the results are shown in Table 2. The manufacturing costs of all four processes differ moderately. The reference process with Protein A has the highest manufacturing costs; processes III and IV are more favorable. The SCI-values of all single steps of the four processes are presented in Fig. 4. The y-axis starts in all diagrams at the point of

Fig. 3. Purication rating of all steps of the four example processes by PPI.

38

T. Winkelnkemper, G. Schembecker / Separation and Purication Technology 72 (2010) 3439

Table 2 Calculated manufacturing costs of the example processes. Process Manufacturing costs [D /g puried IgG1] Production I II III IV 94.56 105.49 93.63 90.58 Pur. step 1 41.94 (ProA) 23.31 (S) 20.69 (S) 20.02 (S) Pur. step 2 20.27 (S) 22.27 (Q) 19.76 (Q) 19.43 (ABx) Pur. step 3 17.95 (Q) 22.63 (ABx) 20.26 (HIC) 17.59 (Q) Total 174.72 173.70 154.35 147.62

specic production costs conversion as values below are not possible. An SCI of 149.5 D /g puried IgG1 was calculated for purication with Protein A (step 1, process I). That means that the manufacturing costs of a downstream process consisting only of steps equally cost-efcient to Protein A would be 149.5 D /g puried IgG1. Fig. 4 shows that the SCI of Protein A is lower than that of cation-exchange resin S as a rst step (processes IIIV) and comparable to the total manufacturing costs of the most advantageous process IV. Thus, application of Protein A could lead to a competitive downstream process, if the following steps were equally or more efcient. In fact they are not. Especially the (low-cost) cation-exchange step

Fig. 5. Relative operating costs of a process consisting of protein production (fermentation), two-phase extraction, concentration and nal purication depending on yield and purity of the extraction step. Specic step costs and base case for rel. SCI calculation were chosen according to Huenupi et al. Base case: purity 45.9%, yield 88.3%, concentration 0.64 g/dm3 [18].

(S) is very inefcient downstream Protein A due to comparatively low purication and yield. As illustrated for processes II and III, the cost-efciency of cationexchange is better when applied as a rst step in the downstream process. The SCI of the following anion-exchange is very low indicating extraordinary cost-efciency. If a nal step with equal SCI could be found, the manufacturing cost would be low, but neither mixed-mode ion exchange (ABx, process II) nor hydrophobic interaction (HIC, process III) meets this demand. A comparison of process II and process IV demonstrates that ABx is much more efcient when applied second. Follman and Farner presented similar conclusions, but on a qualitative level. Rating by SCI has the advantage of quantitative assessment of cost-efciency. Further benet arises, when SCI is used from the beginning of experimentally based process design to guide the development process, which will be demonstrated in a following publication. 4. Discussion The new concept of SCI and PPI can be discussed best in comparison to the method of cost-estimation proposed by Huenupi et al. [18], as it was created with a very similar intention. The inuence of the ow rate ratio of an aqueous two-phase extraction on the total costs of a given protein purication process was investigated on basis of yield, outlet purity and concentration. Due to the comprehensive simulation data presented by Huenupi the results of both cost-estimations can be directly compared and show the same trend. That is not surprising, as simulation results show that yield and purity are not inuenced by the investigated ow rate ratio over a wide range and fall simultaneously above a certain value. Therefore, any reasonable cost function depending on yield and purity would show the same trend. Far more demonstrative is a comparison of both methods concerning the sensitivity towards changes in step yield and outlet purity as shown in Fig. 5. In contrast to the SCI, the cost function by Huenupi et al. is not sensitive to yield and purity changes over the wide range of 1090%. A part of this problem might be, that scaling indices were adopted from equipment scale-up, which have no bearing on operating costs. Shene et al. consequently set up a similar cost function without scaling indices, which represents the qualitative dependencies of process costs on yield and purity linearly [19]. SCI calculation is based on yield and purity (PPI). The concentration considered by Huenupi et al. is indirectly contained in the specic purication costs comprising solvent recycling. An advantage of their qualitative cost function is the dispensability of

Fig. 4. Comparison of the cost-efciency of all steps within the four example processes by SCI. The dashed lines indicate the manufacturing costs of each process (Table 2): (a) process I, (b) process II, (c) process III, and (d) process IV.

T. Winkelnkemper, G. Schembecker / Separation and Purication Technology 72 (2010) 3439

39

absolute specic cost data, which is fundamental for the SCI. On the other hand the SCI provides absolute values that can be related to the real total operating costs and even the nal product value, if specic costs are chosen properly. The major advantage of SCI is its independence of a downstream process concept. Assessment of effort for the unknown part of a purication process is possible by weighting and normalizing purication. 5. Conclusion During purication process development many decisions of increasing detail must be made, starting from alternative unit operations and ending at operating conditions for all steps. Purity difference and purication factors can only serve for direct comparison of different purication steps starting from the same purity level, and there is no meaningful normalization to process boundaries. In contrast to these measures the newly introduced PPI allows purication rating of steps at any position within a process with respect to the process boundaries. Interlinked with the step yield a new measure for single step cost-efciency, the SCI is presented, which rates costs per purication and hence forms a basis for economy-driven downstream process development from the beginning. An example process from literature was used to introduce the new concepts. Notation

n x Y

number of purication steps in a downstream process purity yield

References
[1] K. Cziner, T. Virkki-Hatakka, M. Hurme, I. Turunen, Chemical Engineering and Technology 28 (2005) 14901499. [2] M. Degerman, N. Jakobsson, B. Nilsson, Chemical Engineering and Technology 31 (2008) 875882. [3] Z.L. Xiu, A.P. Zeng, Applied Microbiology and Biotechnology 78 (2008) 917 926. [4] J.A. DiMasi, R.W. Hansen, H.G. Grabowski, Journal of Health Economics 22 (2003) 151185. [5] D.C. Hempel, Engineering in Life Sciences 6 (2006) 443447. [6] B. Harjo, C. Wibowo, K.M. Ng, Chemical Engineering Research and Design 82 (2004) 10101028. [7] B.K. Nfor, T. Ahamed, G.W.K. van Dedem, L.A.M. van der Wielen, E.J.A.X. van de Sandt, M.H.M. Eppink, M. Ottens, Journal of Chemical Technology and Biotechnology 83 (2008) 124132. [8] K. Bauer, G. Schembecker, Chemie Ingenieur Technik 80 (2008) 185190. [9] M. Jobson, D. Hildebrandt, D. Glasser, Chemical Engineering Science 51 (1996) 47494757. [10] S.B. Gadewar, G. Schembecker, M.F. Doherty, Chemical Engineering Progress 102 (2006) 2232. [11] E.W. Leser, J.A. Asenjo, Journal of Chromatography: Biomedical Applications 584 (1992) 4357. [12] M.E. Lienqueo, J.C. Salgado, J.A. Asenjo, Journal of Chemical Technology and Biotechnology 74 (1999) 293299. [13] J.A. Asenjo, B.A. Andrews, Journal of Molecular Recognition 17 (2004) 236 247. [14] M.A. Steffens, E.S. Fraga, I.D.L. Bogle, Biotechnology and Bioengineering 68 (2000) 218230. [15] E. Simeonidis, J.M. Pinto, M.E. Lienqueo, S. Tsoka, L.G. Papageorgiou, Biotechnology Progress 21 (2005) 875884. [16] M.E. Lienqueo, J.C. Salgado, O. Giaverini, J.A. Asenjo, Separation and Purication Technology 65 (2009) 8694. [17] M.E. Groep, M.E. Gregory, L.S. Kershenbaum, I.D.L. Bogle, Biotechnology and Bioengineering 67 (2000) 300311. [18] E. Huenupi, A. Gomez, B.A. Andrews, J.A. Asenjo, Journal of Chemical Technology and Biotechnology 74 (1999) 256263. [19] C. Shene, A. Lucero, B.A. Andrews, J.A. Asenjo, Biotechnology and Bioengineering 95 (2006) 704713. [20] W. Marquardt, M. Nagl, Computers and Chemical Engineering 29 (2004) 65 82. [21] A. Carvalho, R. Gani, H. Matos, Computer Aided Chemical Engineering 21 (2006) 817822. [22] J.A. Asenjo, in: J.A. Asenjo (Ed.), Separation Processes in Biotechnology, Marcel Dekker, Inc., New York, 1990, pp. 316. [23] J. Hubbuch, M.R. Kula, Journal of Non-Equilibrium Thermodynamics 32 (2007) 99127. [24] D.K. Follman, R.L. Farner, Journal of Chromatography A 1024 (2004) 7985. [25] R.G. Werner, Journal of Biotechnology 113 (2004) 171182. [26] S.S. Farid, Journal of Chromatography B 848 (2007) 818. [27] R. Harrison, P. Todd, S.R. Rudge, D.P. Petrides, in: K.E. Gubbins (Ed.), Bioseparations Science and Engineering, Oxford University Press, New York, 2003. [28] S. Sommerfeld, J. Strube, Chemical Engineering and Processing 44 (2005) 11231137.

Indices 0 initial, before the downstream process A an exemplary initial process step B an exemplary intermediate process step C an exemplary nal process step conversion conversion step in a process f nal, at the end of the complete process in at the inlet of a step i all purication steps from 1 to i j purication step j out at the outlet of a step prod product pur purication step in a process tot total process Variables and parameters specic costs related to the target compound m mass