Chemical Engineering Science 131 (2015) 91–100
`
`Contents lists available at ScienceDirect
`
`Chemical Engineering Science
`
`journal homepage: www.elsevier.com/locate/ces
`
`Engineering batch and pulse refolding with transition of aggregation
`kinetics: An investigation using green fluorescent protein (GFP)
`Siqi Pan b, Nora Odabas b, Bernhard Sissolak a, Moritz Imendörffer b, Monika Zelger b,
`Alois Jungbauer a,b, Rainer Hahn a,b,n
`a Department of Biotechnology, University of Natural Resources and Life Sciences Vienna, Muthgasse 18, 1190 Vienna, Austria
`b Austrian Centre of Industrial Biotechnology, Muthgasse 18, 1190 Vienna, Austria
`
`H I G H L I G H T S
` Aggregation transits from 2nd to 1st order as intermediate depletes during refolding.
` Better prediction in batch and pulse refolding using proposed transition model.
` Native model protein (sGFPmut3.1) does not aggregate with intermediates.
` Potential engineering tool to optimize in vitro refolding in bioprocess settings.
`
`a r t i c l e i n f o
`
`a b s t r a c t
`
`Article history:
`Received 5 January 2015
`Received in revised form
`25 March 2015
`Accepted 30 March 2015
`Available online 6 April 2015
`
`Keywords:
`Inclusion bodies
`Renaturation
`Folding
`Reaction order
`GFP
`
`1.
`
`Introduction
`
`Pulse refolding is a strategy to overcome concentration dependent aggregation, assuming that
`aggregation is significantly suppressed under diluted conditions. When a typical 2nd or higher order
`aggregation kinetics is assumed, kinetics over predicted yields at low refolding concentrations. Using
`GFP as our model protein, we found a transition in aggregation kinetics from 2nd to 1st order when
`intermediates deplete from 100 to 60 mg/ml. Taking this transition into account, the model can better
`predict refolding yields in batch and pulse refolding strategies. This model is suited for the design of
`refolding processes since this deviation from 2nd or higher order aggregation was also previously
`observed in other proteins.
`& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
`license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
`
`Pulse refolding is a strategy to overcome aggregation by feeding
`denatured proteins in discrete amounts over specific time intervals
`into refolding buffer. Previous work showed improved yields when
`performed in both batch and continuous reactors (Katoh and Katoh,
`2000; Linke et al., 2014; Pan et al., 2014; Schlegl et al., 2005; Winter
`et al., 2002). Other strategies for improving yield and productivity
`include buffer optimization (Berg et al., 2012; Mannall et al., 2009;
`Ordidge et al., 2012), better mixing (Mannall et al., 2006), on-column
`refolding (Li et al., 2004; Schmoeger et al., 2009) and annular
`chromatography (Uretschlager and Jungbauer, 2002)
`However, to accurately quantify process performance and opti-
`mize process of these refolding strategies, a robust kinetic model
`that satisfies different reactor formats is beneficial (Buswell and
`
`n Corresponding author at: Department of Biotechnology, University of Natural
`Resources and Life Sciences Vienna, Muthgasse 18, 1190 Vienna, Austria. Tel.:
`þ43 1 47654 6671; fax: þ43 1 3697615.
`E-mail address: rainer.hahn@boku.ac.at (R. Hahn).
`
`Middelberg, 2003). Moreover, a correct biomolecular reaction
`scheme would help facilitate product quality and acceptable varia-
`bility of process parameters by serving as a mechanistic model
`support tool in Process Analytical Technology (PAT) (Glassey et al.,
`2011) as part of the Quality by Design (QbD) concept (Rathore and
`Winkle, 2009) in biomanufacturing.
`The key criteria in developing rigorous kinetic models of
`biologics require knowledge of
`the simplest correct kinetic
`scheme (Buswell and Middelberg, 2003). For example, a kinetic
`scheme for lysozyme refolding and aggregation that involved a
`sequential polymerization with the folding intermediates and the
`native protein (Buswell and Middelberg, 2003) as well as the
`competition between aggregation and self-assembly during virus-
`like particle processing (Ding et al., 2010).
`Similarly, our objective is to characterize and establish a simple
`but process-suited model to predict in vitro refolding yields. Current
`models proposing a fixed 2nd or higher aggregation order (Hevehan
`and De Bernardez Clark, 1997; Kiefhaber et al., 1991) overestimate
`yields at low protein concentrations. This was seen for lysozyme
`(Buswell and Middelberg, 2003), autoprotease EDDIE-pep6His (Kaar
`
`http://dx.doi.org/10.1016/j.ces.2015.03.054
`0009-2509/& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 1
`
`

`

`92
`
`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`et al., 2009) and insulin growth factors (Milner et al., 1995) where
`yields did not reach 100% at low refolding concentrations. This
`suggests a deviation from 2nd and higher order aggregation at dilute
`conditions.
`Consequently, using the mutant sGFPmut3.1 (Franke et al., 2007)
`recovered from inclusion bodies (IBs) lacking the mature chromo-
`phore as our model protein (Reid and Flynn, 1997), we proposed a
`refolding model where aggregation transits from 2nd to 1st order
`aggregation as intermediates deplete below a critical concentration.
`Importantly, this study accounts for the protein concentration,
`denaturant and reducing agent concentrations during refolding.
`The effect on refolding due to the presence of native sGFPmut3.1
`was also tested. To test the predictability of our model on other
`refolding strategies, pulse refolding experiments were also per-
`formed at different refolding conditions.
`
`2. Materials and methods
`
`2.1. Expression and inclusion body recovery
`
`Unless stated otherwise, all chemicals were obtained from
`Sigma Aldrich (St. Louis, MO, USA). Recombinant protein
`sGFPmut3.1 was overexpressed in fed-batch cultivation was as
`described in Clementschitsch et al. (2005). Isolation of IBs was as
`previously described by Kaar et al. (2009).
`
`2.2. Purifying soluble sGFPmut3.1 expressed in Escherichia coli for
`spiking experiments
`
`Cell broth was centrifuged with the Contifuge Stratos Heraeus
`(Thermo Fisher Scientific, Waltham, MA, USA) to gain the cell pellet.
`This was suspended in a chilled (þ4 1C) solution of 10 mM Tris,
`0.1 M NaCl and 0.1% Tween 20 at pH 7.5. Cell disruption was done
`with the Ariete 2-stage high pressure homogenizer (GEA Niro Soavi,
`Parma, Italy) at 80/800 bar in two passages. Subsequently, homo-
`genate was cleared (Contifuge) and filtrated with Sartopure PP2,
`Sartoguard GF and Sartobran P at 1.2, 0.65 and 0.45þ0.2 mm
`respectively. Buffer exchange to 10 mM Tris at pH 7.5 was done
`with a Sartoflow advanced system with a 10 kDa Hydrosart Sarto-
`coon (Satorious, Göttingen, Germany) in 7 volume changes. After,
`3 chromatography steps were performed with the Äkta Pilot System
`(GE Healthcare, Buckinghamshire, UK). For capturing, CaptoQ ion-
`exchange resin was used. Purification was made with Butyl-Sephar-
`ose, a hydrophobic interactions chromatography. Polishing was
`done by size exclusion, Superdex 75. All resins were obtained from
`GE Healthcare. Quantification of
`impurities were analyzed by
`Superdex 75 column at the Äkta Explorer System and by SDS-PAGE,
`BioRad PowerPack basic (Bio-Rad Laboratories, Hercules, USA).
`
`2.4. Determining rate constants of sGFPmut3.1 at different residual
`urea
`
`Refolding was initiated in 5 ml eppendorfs containing 0.3 M L-
`arginine/HCl (SERVA, Heidelberg, Germany), 1 M Tris, 0.25 M
`sucrose, 2 mM EDTA, 20 mM MTG and pH 7.3 refolding buffer in
`predefined urea concentrations. Solution was vortexed immedi-
`ately and inserted onto laboratory rotator (SB3, Stuart) (10 rpm).
`All refolding in this study took place at 2371 1C. At specified
`times over 7 h, 100 ml samples were drawn and measured for
`fluorescence yield. Refolding concentrations of sGFPmut3.1 were
`25, 38, 56, 114, 158, 190 mg/ml in residual urea concentrations of
`0.24, 0.50, 0.90, 1.12, 1.32, 1.52, 1.80 M. Using Table Curve3D (SPSS,
`Erkrath, Germany), kinetic constants for concentrations 25, 38,
`56 mg/ml at each residual urea were calculated by fitting data sets
`into Eq. (4) while the higher concentrations 56, 114, 158, 190 mg/ml
`were globally fitted with Eq. (3) at each residual urea condition.
`
`2.5. Batch refolding at different reducing agent concentrations
`
`Refolding at 0.2 mg/ml sGFPmut3.1 was initiated as previously
`described at 1:10 ratio dilution where 0.5 ml dissolved IBs were
`added to 5 ml eppendorfs containing 4.5 ml refolding buffer
`previously described but at predefined MTG concentrations. The
`resultant MTG concentration ranges between 10 and 100 mM.
`Samples were drawn and measured for fluorescence yield over
`refolding time.
`
`2.6. Establishing refolding simulation
`
`Using fourth-order Runge–Kutta method, Eq. (1) (2) and (5)
`and rate constants that were experimentally derived with increas-
`ing residual urea, batch and pulse refolding simulations were
`established using Microsofts Office Excel 2013. Simulations were
`verified with analytical solutions of Eqs. (3) and (4) at different
`refolding conditions. Additionally, the total mass balance of inter-
`mediates, native and aggregate species were always 100% over
`refolding time. This simulation was then used to predict the
`refolding experimental results.
`
`2.7. Batch refolding with presence of native GFPmut3.1
`
`Refolding was performed in 50 ml beakers at different refolding
`conditions of 46, 49, 62, 95 mg/ml at a residual urea concentration
`of 0.90, 0.69, 0.90, 0.90 M respectively containing specific amounts
`of pure native sGFPmut3.1. As a control, identical refolding condi-
`tions were also performed without pure native sGFPmut3.1. Yields
`were then calculated after accounting for fluorescence due to
`native pure sGFPmut3.1.
`
`2.8. SEC analysis of refolded sGFPmut3.1
`
`2.3. Denaturing and reducing sGFPmut3.1 IBs
`
`From IBs, sGFPmut3.1 was lyophilized, weighed and suspended
`in 50 mM Tris (pH 7.3) overnight. Suspended IBs were denatured
`and reduced by 1:10 ratio in dissolution buffer containing 10 M
`urea, 50 mM Tris and 100 mM α-monothioglycerol (MTG) at pH
`7.3 for 0.5 h. Protein concentration stock was measured on a Cary
`50 Bio UV–vis Spectrophotometer (Varian, Palo Alto, USA) at a
`theoretical extinction coefficient of 0.813 (mg/ml protein) cm1 at
`280 nm. Stock was further diluted to the desired concentrations
`using buffer containing 9 M urea, 50 mM Tris, 100 mM MTG and
`pH 7.3.
`
`Analytical SEC analysis was performed with Agilent 1290 Infinity
`UHPLC instrument (Agilent, Waldbronn, Germany) together with
`Agilent Bio SEC-5 Column (300 mm 4.6 mm i.d.) with particle size
`5 mm and pore structure of 100 Å. Samples analyzed were purified
`sGFPmut3.1 and refolded sGFPmut3.1 from IBs. Each analysis took
`15 min where running buffer was 1 phosphate saline buffer. Flow
`rate was 0.5 ml/min, column temperature was 25 1C and injection
`volume was 10 ml. Absorbance was measured simultaneously at
`214 nm to detect peptide bonds and 485 nm to detect fluorescence
`chromophore. To determine molecular weight of soluble aggre-
`gates, a high molecular weight kit of 5 proteins (GE Healthcare,
`Buckinghamshire, UK) between 44 and 669 kDa was analyzed.
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 2
`
`

`

`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`93
`
`2.9. Batch refolding of sGFPmut3.1
`
`Refolding was performed in 50 ml beakers at different refolding
`conditions at 46, 49, 51, 62, 66, 95 mg/ml at a residual urea
`concentration of 0.90, 0.69, 0.47, 0.90, 0.62, 0.90 M respectively.
`Refolding solution was stirred using identical magnetic bars at
`500 rpm during initiation and for one minute before it was left to
`stand where samples were drawn and measured for fluorescence
`yield over refolding time.
`
`2.10. Pulse refolding
`
`In four 50 ml beakers, refolding was performed in 1, 2, 3 and
`4 pulses, all reaching a final refolding protein concentration of
`200 mg/ml at 0.90 M urea. Interval time between each subsequent
`pulse was 90 min, protein and denaturant concentrations were
`calculated in each interval for simulation prediction. Between each
`pulse, samples were drawn and measured for fluorescence yield.
`To ensure accurate refolding conditions, pulse volumes were
`corrected based on samples drawn before them.
`
`2.11. Fluorescence measurements
`
`Fluorescence of sGFPmut3.1 was determined by 485 nm excita-
`tion and 535 nm emission with plate reader (TECAN Austria
`GmbH, Salzburg, Austria) in 96 well flat bottom black plate
`(Thermo scientific, Denmark). Overall yield was calculated with
`calibration curve from 4 to 200 mg/ml of 100 ml native sGFPmut3.1
`(in-house standard).
`
`3. Theory
`
`Upon refolding, the formation of the molten globule inter-
`mediate (I) is rapidly formed from its denatured state (U) (Arai
`et al., 2002). After, intermediates undergo kinetic competition
`between refolding and aggregation (Kiefhaber et al., 1991) to form
`native proteins (N) and irreversible aggregates (A) (Baker and
`Agard, 1994) (Fig. 1). Since this is the rate limiting step, the
`conversion of intermediates can be described in a differential
`equation as
`¼ k1IK2In
`
`ð1Þ
`
`dI
`dt
`
`While the formation of native proteins can be described as
`¼ k1I
`
`dN
`dt
`where k1 is the refolding rate constant, K2 is the aggregation rate
`constant, n is the order of aggregation,
`is the intermediate
`I
`concentration and N is the native protein concentration. The rate
`constants are dependent on the refolding environment such as
`temperature, redox potential and denaturant, while the aggrega-
`tion order is the number of intermediates required to collide
`together at enough energy for aggregation to occur.
`
`ð2Þ
`
`Fig. 1. Kinetic scheme of in vitro refolding of proteins. Denatured protein (U) is
`rapidly formed to intermediates (I).
`Intermediates are subsequently depleted
`during the competitive formation between the native (N) and aggregate
`(A) species.
`
`ð1e
`
`ð3Þ
`
`Several proteins such as IGF-1 (Mannall et al., 2009), porcine
`muscle lactic dehydrogenase (Kiefhaber et al., 1991) and 6H-
`EDDIE-sGFPmut3.1 (Achmüller et al., 2007) proposed that inter-
`mediates (I) undergo a competitive 1st order refolding and 2nd
`order aggregation. Using Eqs. (1) and (2), the analytical solution
`
`
`can be described as
`Y tð Þ ¼ k1
`ln 1þI0k2
` k1tÞ
`I0k2
`k1
`where Y is percentage conversion to native proteins, I0 [mol/l] is
`the initial available intermediates, k1 [s1] is the rate constant of
`folding, K2 [M/s] is the apparent rate constant of aggregation, and t
`[s] the time (Kiefhaber et al., 1991) (see Appendix A for derivation).
`Other proteins such as EDDIE-pep6His (Kaar et al., 2009) and
`lysozyme at low concentrations (Buswell and Middelberg, 2003)
`undergo a competitive 1st order refolding and 1st order aggrega-
`tion (misfolding). The analytical solution can be described as
`Y tð Þ ¼ k1
` k1
`ð4Þ
`e
`k1 þk2
`k1 þk2
`where Y is the refolding yield, k1 [s1] the rate constant of folding,
`k2 [s1] the rate constant of misfolding and t [s] the time (Kaar
`et al., 2009) (see Appendix A for derivation).
`We hypothesize there are proteins where aggregation order in
`Eq. (1) changes as a function of intermediate concentration. We
`define a critical intermediate concentration (Icrit), where aggrega-
`tion transits from 2nd or higher to 1st order as intermediate
`depletes. In this case, a single analytical solution for the entire
`refolding course does not exist. For increased flexibility, the
`transition model is solved numerically using Eqs. (1) and (2). This
`we define as
`n ¼ 2 when I 4Icrit
`ð5Þ
`n ¼ 1 when IoIcrit
`where n is the order of aggregation, I [mol/l] the intermediate
`concentration during refolding and Icrit [mol/l] the transition point
`from 2nd to 1st order aggregation during refolding.
`
`ðk1 þ k2Þt
`
`4. Results and discussion
`
`The key objective of this work is to propose that the aggrega-
`tion order of refolding proteins transits from 2nd to 1st order
`aggregation as its intermediate state is depleted overtime, result-
`ing in better prediction in batch and pulse refolding of proteins as
`compared to conventional fixed aggregation order models
`(Hevehan and De Bernardez Clark, 1997; Kiefhaber et al., 1991),
`using sGFPmut3.1 as our model protein.
`Initially, the influence of reducing, protein and urea concentra-
`tions on sGFPmut3.1 was characterized. Additionally, presence of
`back aggregation of native proteins was determined. Subsequently,
`simulations of proposed transition and conventional model was
`established. Lastly, batch and pulse refolding experiments were
`performed. Transition and conventional model prediction from
`simulations was then confirmed by comparing to experiments
`under identical conditions.
`
`4.1. Effect of MTG on refolding yields and kinetics of sGFPmut3.1
`
`Since redox environment may affect refolding kinetics as
`observed in lysozyme (Hevehan and De Bernardez Clark, 1997)
`and α-lactalbumin (Schlegl et al., 2005), refolding was performed
`at different α-monothioglycerol (MTG) concentrations between 10
`and 100 mM, at 200 mg/ml sGFPmut3.1 and 0.9 M residual urea. As
`shown in Fig. 2A, kinetic profile and yields in all cases were very
`similar, where changes in MTG have no influence on rate constants
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 3
`
`

`

`94
`
`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`Fig. 2. (A) Refolding kinetics of 200 mg/ml sGFPmut3.1 and 0.9 M residual urea in reducing MTG concentrations between 10 and 100 mM. (B) Yields as a function of protein
`refolding concentration. Dots represent refolding yields after 7 h refolding of sGFPmut3.1 in 1.3 M residual urea. Solid line represents model (Kiefhaber et al., 1991) describing
`the competition between 1st order refolding and 2nd order aggregation of sGFPmut3.1.
`
`Fig. 3. Refolding kinetics with and without purified native sGFPmut3.1.
`
`and refolding yields. Similarly, Inouye and Tsuji (1994) reported that
`fluorescence is lost upon treatment with strong reducing agents like
`sodium dithionite, but not with weaker reducing agents like
`2-mercaptoethanol, dithiothreitol or reduced glutathione. Structu-
`rally, GFP has two cysteines at position 48 and 70, but no disulfide
`bonds were formed. Yang et al. (1996) proposed reaction of Cys70
`near one end of the cylindrical structure could disrupt the packing of
`the cap on the end, allowing quenching of the chromophore. Since
`
`MTG concentration in all refolding experiments were below the
`maximal tolerable range of 10–100 mM, it should have no influence
`on sGFPmut3.1 refolding kinetics.
`
`4.2. Effect of protein refolding concentration on refolding yields
`
`Batch refolding of sGFPmut3.1 was performed at 25, 38, 56, 114,
`158, 190 mg/ml in residual urea concentrations of 0.24, 0.50, 0.90,
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 4
`
`

`

`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`95
`
`not shown). Likewise, the yields of other proteins such as lyso-
`zyme (Buswell and Middelberg, 2003), the autoprotease EDDIE-
`pep6His (Kaar et al., 2009) and insulin growth factors (Milner
`et al., 1995) also did not reach 100% at very low concentrations.
`Physically, the loss in yield suggests misfolding originating from
`imperfect mixing (Mannall et al., 2006); upon addition of dena-
`tured protein into refolding buffer, a gradient of moderately
`denaturing conditions were formed. This populates a portion of
`intermediates that favor the aggregation pathway (Brems, 1988).
`Once refolding environment is homogenous, collisions between
`proteins were minimized due to low refolding concentrations,
`thus overall reaction was dominated between 1st order refolding
`and 1st order aggregation.
`In lysozyme (Buswell and Middelberg, 2003) and insulin
`growth factors (Milner et al., 1995), this misfolding was presum-
`ably due to incorrect disulfide formation (Roux et al., 1997; Saxena
`and Wetlaufer, 1970). Since refolding were performed under
`reducing conditions and native GFP has no disulfide bonds,
`misfolding by incorrect disulfide formation was ruled out.
`The exact biological reason for misfolding in GFP is unclear.
`Misfolding could occur during the three sequential steps to chro-
`mophore formation of GFP derived from inclusion bodies (Reid and
`Flynn, 1997). Misfolding could also occur during inclusion body
`formation and partially preserved even after inclusion body dissolu-
`tion and reduction, since studies have shown residual secondary
`structures in protein denaturation conditions by urea and GdnHCl
`(Matsuo et al., 2007). Future work is needed to determine whether
`GFP was formed in vivo or/and during the refolding process.
`
`4.3. Refolding kinetics with the presence of pure native sGFPmut3.1
`
`Buswell and Middelberg (2003) reported that the presence of
`native lysozyme significantly decreased the effective refolding yield.
`This was because that native lysozyme was able to polymerize with
`aggregates (Buswell and Middelberg, 2002). We checked this
`possibility by adding pure native sGFPmut3.1 in our refolding buffer
`before refolding.
`In contrast to decrease in yields in the presence of native
`lysozyme (Buswell and Middelberg, 2003), refolding yields remained
`unaffected in the presence of pure native sGFPmut3.1 (Fig. 3). This
`was observed even when native sGFPmut3.1 concentration was more
`than 3 times the amount of refolding concentration (Fig. 3B). If no
`decrease in yields were seen, back aggregation of native sGFPmut3.1
`into a non-fluorescence conformation can be ruled out.
`Separately, there was a higher noise level in kinetic yields
`containing native sGFPmut3.1 (Fig. 3) because kinetic yields were
`obtained by subtracting the total sGFPmut3.1 fluorescence in
`solution (pure nativeþrefolded sGFPmut3.1) with the fluorescence
`contributed by pure native sGFPmut3.1.
`
`4.4. SEC analysis of refolded sGFPmut3.1
`
`To ensure aggregates do not contain the fluorescing native
`sGFPmut3.1, SEC analysis was performed on both refolded
`sGFPmut3.1 and pure native sGFPmut3.1. Soluble aggregates could
`be seen at 214 nm (Fig. 4A), however no signal was detected at
`485 nm at the same retention time (Fig. 4B). Because native
`sGFPmut3.1 chromophore absorbs at 485 nm, this suggests that
`soluble aggregates do not contain elements of native sGFPmut3.1.
`Separately, soluble aggregates found after refolding was mea-
`sured against a standard containing proteins between 44 and
`669 kDa (Fig. 4A). Most of the aggregates peaked around 669 kDa
`against the standard. Additionally a broad distribution of smaller
`aggregates were found at lower molecular weights. Clearly, poly-
`merization of aggregates was observed since aggregates were 25
`folds larger than the 26.7 kDa sGFPmut3.1. Still, whether the
`
`Fig. 4. Representative analytical SEC chromatograms of pure and refolded
`sGFPmut3.1 at (A) 214 nm and (B) 485 nm absorbance.
`
`Fig. 5. Representative refolding kinetics of sGFPmut3.1 at different refolding
`concentration in 1.3 M residual urea. Solid line represent fits of Eq. (3) to 25, 38
`and 56 mg/ml sGFPmut3.1 refolding. Dashed lines represent global fit of Eq. (4) to
`concentrations 56, 114, 158 and 190 mg/ml.
`
`1.12, 1.32, 1.52, 1.80 M. Results were compared against conven-
`tional model proposed by Kiefhaber et al. (1991) using Eq. (3) as
`shown in 1.3 M residual urea (Fig. 2B). Our results showed that
`conventional model
`is generally accurate in the concentration
`dependent region above 56 mg/ml, but fail to predict yields in
`the concentration independent region below 56 mg/ml. This beha-
`vior was consistent for other residual urea concentrations (data
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 5
`
`

`

`96
`
`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`Fig. 6. Linear decrease in fitted rate constants of sGFPmut3.1 with increasing residual urea concentrations. Both (A) and (B) were globally fitted to refolding concentrations
`above 56 m/ml with Eq. (4), while (C) and (D) were fitted to refolding concentrations below 56 mg/ml with Eq. (3).
`
`Fig. 7. Simulation of native, intermediate and aggregation pathways of (A) batch and (B) pulse refolding at 4 pulses of 90 min intervals to a final concentration of 200 mg/ml
`sGFPmut3.1 and 0.9 M residual urea.
`
`aggregation undergo sequential, multimeric (Speed et al., 1997) or
`nucleation polymerization (Jarrett and Lansbury, 1992) would
`require further investigation.
`
`4.5. Simulation of batch and pulse sGFPmut3.1 refolding
`
`Based on initial refolding experiments, evidence show that:
`(1) MTG has little influence on refolding.
`(2) Kinetics was
`
`concentration dependent at high concentration and independent
`at a low concentration. (3) Presence of native sGFPmut3.1 do not
`influence refolding yields and do not participate on sequential
`polymerization with soluble aggregates during the refolding process.
`This supports the transition model where intermediates undergo a
`competing 1st order refolding and 2nd order aggregation above the
`critical intermediate concentration level (Icrit), before transiting to a
`competing 1st order refolding and 1st order aggregation refolding.
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 6
`
`

`

`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`97
`
`Although it is difficult to determine the exact value of Icrit and
`also if a gradient from 2nd to 1st order aggregation exists, based
`on refolding experiments at different protein and residual urea
`concentrations as shown in Fig. 2B and Fig. 5, we assumed a step
`transition at an estimated Icrit value of 56 mg/ml. To determine the
`effect of residual urea on rate constants, kinetics of different
`protein concentrations of the same residual urea above 56 mg/ml
`were globally fitted to Eq. (3), while different protein concentra-
`tion below 56 mg/ml of their respective residual urea were fitted
`with Eq. (4) as shown in Fig. 5. In all cases, fitted rate constants
`showed a linear relationship for both refolding and aggregation
`constants (Fig. 6).
`To test if the transition model is mechanistically accurate, we
`first establish a simulation using fitted rate constants and pro-
`posed model, calculating the conversion of
`intermediates to
`refolded and aggregated species in batch and pulse refolding
`(Fig. 7). As expected with previous empirical studies (Winter
`et al., 2002), pulse refolding attained greater yield than batch
`refolding under same refolding conditions. This simulation was
`later used to predict the kinetics and verified with refolding
`experiments.
`
`4.6. Batch refolding and simulation prediction
`
`To better observe transition effects from 2nd to 1st order
`aggregation as intermediates deplete, batch refolding experiments
`were performed above and below Icrit, between 49 and 95 mg/ml
`(Fig. 8). In all cases, proposed model predicted more accurately
`
`than a pure 2nd order simulation applied in previous studies (Kaar
`et al., 2009; Kiefhaber et al., 1991; Mannall et al., 2009; Schlegl
`et al., 2003). The pure 2nd order model over estimated refolding
`kinetics because it assumes that in low intermediate concentra-
`tions, yields would become higher as aggregation is suppressed. In
`contrast, proposed model accounts for the effects of 1st order
`aggregation (misfolding) once intermediates were depleted to
`lower concentrations at later stages of refolding.
`
`4.7. Pulse refolding and simulation prediction
`
`To ensure that transition model can be applied to other reactor
`formats (Buswell and Middelberg, 2003; Ding et al., 2010), pulse
`refolding was performed experimentally from 1 pulse (batch) to
`4 pulses, reaching a final protein and residual urea concentration of
`200 mg/ml and 0.9 M respectively. Consistent with pulse refolding on
`human proinsulin (Winter et al., 2002) and the autoprotease 6His-
`EDDIE-sGFPmut3.1 (Pan et al., 2014), experimental results showed
`yield improvements from batch to pulse refolding (Fig. 9).
`Overall, pulse refolding for both models over predicted results.
`However over prediction was greater in the pure 2nd order model
`of up to 12% than transition model of up to 5% (Fig. 9). Additionally,
`over estimation increased with increasing number of pulses
`(Fig. 9B–D). From this over estimation, we question if the pure
`native sGFPmut3.1 artificially added into refolding buffer was truly
`representative to sGFPmut3.1 that was just refolded, since the
`presence of pure native sGFPmut3.1 did not affect the yields and
`kinetics of refolding sGFPmut3.1 (Fig. 3). In contrast, yields were
`
`Fig. 8. Predicted simulations and experimental results of batch refolding. Dotted lines represent 1st order refolding and 2nd order aggregation model simulation. Solid lines
`represent transition model.
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 7
`
`

`

`98
`
`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`Fig. 9. Predicted simulations and experimental results of pulse refolding. Dotted lines represent 1st order refolding and 2nd order aggregation model simulation. Solid lines
`represent transition model.
`
`Table 1
`Accuracy of predictions calculated using mean squared error (MSE) between the
`respective simulation and experimental results.
`
`Pulses Urea
`concentration
`(M)
`
`Protein
`concentration
`(mg/ml)
`
`1
`1
`1
`1
`1
`2
`3
`4
`
`0.90
`0.62
`0.90
`0.69
`0.90
`0.90
`0.90
`0.90
`
`95
`66
`62
`49
`200
`200
`200
`200
`
`2nd Order
`aggregation model
`(MSE)
`8.26  104
`2.11  103
`3.02  103
`5.63  103
`1.36  104
`1.44  103
`3.45  103
`4.40  103
`
`Transition
`aggregation
`model (MSE)
`5.04  104
`7.59  104
`5.14  104
`8.26  104
`6.17  104
`4.14  104
`1.21  103
`1.26  103
`
`lower when denatured sGFPmut3.1 were added to solution con-
`taining sGFPmut3.1 that was just refolded, as was the case in pulse
`refolding.
`Considering there was over prediction in the transition model,
`it was possible that freshly refolded sGFPmut3.1 was still not as
`stable as the pure native form, resulting in a small proportion of
`the freshly refolded sGFPmut3.1 reverting to a non-fluorescence
`conformation. With increasing number of pulses, this unstable
`proportion got bigger, as a higher protein concentration should
`result in greater instability, and thus lower fluorescence yields.
`Separately for batch refolding of 1 pulse (Fig. 9A), the pure 2nd
`order model gave a better prediction than transition model. How-
`ever this prediction was reactor specific (Ding et al., 2010) because
`
`using alternative refolding strategies like pulse refolding for Fig. 9B–
`D showed weaker prediction for the pure 2nd order model.
`Specifically, because protein refolding in batch refolding was
`200 mg/ml, the influence of depleted intermediates in the 1st order
`aggregation region (Io56 mg/ml) were smaller compared when
`intermediates were in the 2nd order aggregation region (56 mg/
`mloIo200 mg/ml), therefore the pure 2nd order model was
`sufficient to predict the kinetic pathway.
`Overall, as summarized in Table 1, the mean squared error for
`the transition model is mostly lower than the 2nd order aggrega-
`tion model, suggesting that the accuracy of prediction by the
`former was better. However the transition model showed that
`mean squared values increased with increasing number of pulses.
`This could be due to freshly refolded sGFPmut3.1 reverting to a
`non-fluorescence conformation as previously explained. Still,
`further studies would still be required to confirm this explanation.
`
`5. Further discussion
`
`Theoretically, a higher order of aggregation would mean a higher
`refolding yield (when all other factors are kept equal). This is
`rational because with increasing aggregation order, more intermedi-
`ates are required to collide together at enough energy for aggrega-
`tion formation to occur. Since aggregation becomes less frequent,
`more intermediates follow the competing refolding pathway thus
`higher refolding yield is observed.
`
`APOTEX EX1057
`Apotex Inc., et al. v. Amgen Inc. et al., IPR2016-01542
`Page 8
`
`

`

`S. Pan et al. / Chemical Engineering Science 131 (2015) 91–100
`
`99
`
`In reality, intermediates to aggregate formation is complex
`because multiple aggregation orders could occur simultaneously
`over time. At high intermediate concentrations, aggregate forma-
`tion could occur from a distribution of many different intermedi-
`ates colliding together. As intermediate depletes, higher orders
`become less probable as collisions decreased, while 1st order
`aggregation dominates. Ideally, if all aggregation orders can be
`quantitatively determined, a precise model could be described by a
`model that includes all participating aggregation orders.
`However, we were limited by analytical methods where inter-
`mediates to aggregate formation could not be quantitatively
`me