Optimization study on periodic counter-current chromatography integrated in a monoclonal antibody downstream process

An optimization study of an integrated periodic counter-current chromatography (PCC) process in a monoclonal antibody (mAb) downstream process at lab scale, is presented in this paper. The optimization was based on a mechanistic model of the breakthrough curve in the protein-A capture step.

Trang 1

Contents lists available at ScienceDirect

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

Joaquín Gomis-Fonsa, b, ∗, Niklas Anderssona, Bernt Nilssona, b

a Dept of Chemical Engineering, Lund University, Lund, Sweden

b Competence Centre for Advanced BioProduction by Continuous Processing, Royal Institute of Technology, Stockholm, Sweden

a r t i c l e i n f o

Article history:

Received 13 January 2020

Revised 3 March 2020

Accepted 17 March 2020

Available online 19 March 2020

Keywords:

Periodic counter-current chromatography

Process optimization

Process integration

Downstream processing

Monoclonal antibody puriﬁcation

a b s t r a c t

Anoptimizationstudyofanintegratedperiodiccounter-currentchromatography(PCC)processina mon-oclonalantibody(mAb) downstreamprocess atlabscale,is presentedinthispaper Theoptimization was basedonamechanistic model ofthe breakthroughcurve in theprotein-A capture step Produc-tivityand resinutilization werethe objectivefunctions,whileyield duringtheloadingofthe capture columnwassetasaconstraint.Differentintegrationapproacheswereconsidered,andtheeffectofthe feedconcentration,yieldandtheprotein-Aresinwasstudied.Thebreakthroughcurveandthelengthof theproductrecovery,whichdependedontheintegrationapproach,determinedtheprocessscheduling SeveraloptimalParetosolutionswereobtained.At0.5mgmL−1and99%yield,amaximumproductivity

of0.38mgmL−1 min−1 witharesinutilizationof60%wasobtained.Ontheotherhand,themaximum resinutilizationwas95%withaproductivityof0.10mgmL−1min−1.Duetotheconstraintoftheprocess scheduling,alowerproductivitycouldbeachievedintheintegrationapproacheswithhigherrecovery time,whichwasmoreremarkableathigherconcentrations.Therefore,itwasshownthataholistic ap-proach,whereallthepuriﬁcationstepsareconsideredintheprocessoptimization,isneededtodesigna PCCinadownstreamprocess

1 Introduction

The biopharmaceutical market demand is constantly changing

and there is an increasing pressure on a price reduction for a

global access to biological drugs [1, 2] Continuous bioprocessing is

a way to reduce biologics price by increasing the productivity and

diminishing the manufacturing costs [2, 3] A signiﬁcant improve-

ment has been carried out in upstream by progressively shifting

from fed-batch to perfusion bioreactors However, the productiv-

ity in downstream has not increased accordingly and nowadays

a big proportion of the manufacturing cost are due to the prod-

uct puriﬁcation [4, 5] Continuous downstream processes, like pe-

riodic counter-current chromatography (PCC) [6, 7], Capture SMB

[7, 8]or multi-column counter-current solvent gradient puriﬁcation

(MCSGP) [9, 10], have gained interest in the last years These pro-

cesses offer a higher productivity and resin utilization, while keep-

∗ Corresponding author: Dept of Chemical Engineering, Lund University, P.O Box

124, SE-21100 Lund, Sweden

E-mail addresses: joaquin.gomis_fons@chemeng.lth.se (J Gomis-Fons),

niklas.andersson@chemeng.lth.se (N Andersson), bernt.nilsson@chemeng.lth.se

(B Nilsson)

ing a similar yield than the one obtained in a batch process [7, 10] They all are based on multiple columns, in a way that a column

is loaded with the outlet of another column In MCSGP, the eluted impurities containing product is loaded onto another column, and

it is usually applied for polishing steps [10] For the capture step, the Capture SMB (2-column PCC) and the 3-column or 4-column PCC are common alternatives [7]

In a PCC operation, two columns are interconnected and loaded while the product is recovered in one or two more columns (depending if it is a 3-column or 4-column PCC) [6] To be able to run

a PCC process, a feed continuity constraint must be fulﬁlled so that the harvest can be continuously loaded onto the capture columns [6] Additional scheduling constraints can also be applied to avoid product loss during the loading Furthermore, to make the most

of the potential of a PCC process, resin utilization and productivity should be maximized For those reasons, the PCC is a process that must be carefully designed Several authors have used empirical models of the breakthrough curve to design a PCC [6, 11] While these models are useful to get the process conditions that make the PCC operate, they fail in obtaining an optimal process, since empirical models are only valid for the residence time and feed concentration at which the experiments are run On the contrary, https://doi.org/10.1016/j.chroma.2020.461055

Trang 2

mechanistic models take into account the adsorption equilibrium

and kinetics, and the mass transfer limitations [12] Therefore, once

the model is calibrated, an optimization over the whole range of

residence times and concentrations, and other process parameters

like loading time, can be performed to obtain one or several op-

timal solutions With the help of the model, investigations around

the PCC process, like the effect of the feed concentration or the

particle diameter can be carried out [7, 13]

Model-based optimization of a PCC process has been done pre-

viously [7, 13] However, in these cases only the capture step is

considered when optimizing the PCC process But in a puriﬁcation

train, the capture step is usually followed by a virus inactivation

process and several polishing steps The integration of the capture

step with the rest of the puriﬁcation steps affects the PCC opti-

mization strongly, and depending on the integration approach, the

effect is more or less signiﬁcant This is because the time that takes

to recover the product (recovery time), which is affected by the in-

tegration approach, has to be lower than the cycle time to meet

the feed continuity constraint in PCC [6] Therefore, if the whole

puriﬁcation process is considered instead of only the capture step,

the recovery time is longer, and the process scheduling is affected

In this work, we performed an optimization study of a PCC

integrated in a monoclonal antibody (mAb) downstream process,

for which a mechanistic model of the breakthrough curve in the

protein-A capture step was used Three integration approaches

were evaluated and their inﬂuence on the Pareto front of optimal

solutions is presented The effect of the chromatography resin, the

feed concentration and the yield were also studied, and the pro-

cess was compared to a 1-column batch capture step The inte-

grated PCC process was veriﬁed experimentally at lab scale with

two different Protein-A resins

2 Process integration alternatives

The way the PCC process in the capture step is integrated with

the rest of the puriﬁcation train is not an obvious choice In this

work, several alternatives are considered depending on the num-

ber of systems or pumps that are used, and whether a surge ves-

sel after the PCC is used or the eluate is directly sent to the next

step (see Fig 1) In this case, a 3-column PCC is chosen because

it provides a higher productivity and resin utilization, compared to

Capture SMB or a 4-column PCC [7] However, the same reasoning

around the PCC and conclusions extracted from this work could be

applied for a four-column PCC or Capture SMB

In all cases, the process consists of the same process steps cor-

responding to a typical mAb puriﬁcation process: 1) a capture

step with a protein A resin, 2) a virus inactivation (VI) at pH

3.5 for 60 min, 3) a cation exchange chromatography (CEX) step

in bind-and-elute mode with a gradient elution, 4) an anion ex-

change chromatography (AEX) step in ﬂow-through mode A ﬁnal

ultraﬁltration-diaﬁltration step for product formulation is not in-

cluded in this process because the volume of the ﬁnal product

from the chromatography steps at lab scale is much lower than

the minimum starting volume in the ultraﬁltration step, as shown

in [14], which would require to use much higher column volumes

in the process validation Anyway, at larger scales, the integration

of the ultraﬁltration step with the rest of the downstream process

including PCC, could easily be carried out the same way it is pre-

sented in [14] This could affect the recovery times, but the deduc-

tions and conclusions from this work would still apply

2.1 Integrated PCC process in one system

The ﬁrst alternative to run the PCC process with the virus inac-

tivation and the polishing steps (CEX and AEX steps) is to integrate

all steps in one chromatography system without any hold-up volume between the steps ( Fig 1A) In order to avoid hold-up volumes, the methodology followed in [14–16], called Integrated Column Sequence (ICS), was used The basic principle is that the eluate of one column is loaded directly onto the next column That requires synchronization between the elution of a column and the loading of the next one In addition, since all steps are performed

in a single chromatography system, they have to be carried out in series Consequently, in this case the recovery time corresponds

to the time it takes for the capture step and the polishing steps

to recover the product The incubation time of the virus inactivation is not included because during this time, regeneration and equilibration steps of the other columns are carried out This is a constraint in the minimum cycle time that limits the productivity

On the other hand, an integrated and continuous mAb puriﬁcation process in only a chromatography system is a compact and cheap alternative that can be specially interesting for lab-scale PCC runs

2.2 Integrated PCC process in two systems

In this alternative, the process is integrated the same way as

in the previous one, but two chromatography systems are used instead ( Fig.1B) A system is used for the PCC process and the virus inactivation, and another system is used for the polishing steps The main difference compared to the one-system alternative is that the capture step and the polishing steps are run in parallel That means that the minimum cycle time is not the sum of the process times of the capture and polishing steps, but the highest of them instead, that is, when one system is done, it waits for the slower one That leads to an increase of productivity due to a reduction of the process time However, the investment cost and the space occupied in the lab is twice as high as in the ﬁrst process alternative

2.3 PCC process with a surge vessel in two systems

The last process choice considered in this work is the use of

a surge vessel between the capture step and the virus inactivation step ( Fig.1C) The vessel allows for a desynchronization of the capture step respect to the polishing steps, that is, the minimum cycle time would be only the recovery time in the capture step There- fore, in terms of the PCC design, this alternative is equivalent to the PCC with the capture alone, without the rest of steps That is

an improvement respect to the previous alternative, since the capture step is usually shorter than the polishing steps, based on a typical mAb puriﬁcation process [17], thus leading to a lower minimum PCC cycle time, which means that higher productivity can

be achieved However, this option has some drawbacks Firstly, the complexity of the process is signiﬁcantly increased Since the ﬁll- ing and the emptying of the surge vessel, both discrete operations, are not synchronized (as it can be seen in Fig.4), the liquid level before emptying the vessel is different from cycle to cycle This means that a level sensor and a controller are needed to keep the liquid level in the vessel the same after each cycle In addition, this desynchronization involves that, for certain PCC cycles the volume

in the vessel is higher than for other cycles Therefore, the column volumes of the polishing steps are over-designed for those cycles with lower volume in the surge vessel Furthermore, the presence

of a surge vessel increases the risk of product degradation and the residence time of the product, increases the capital cost, and con- tributes to slowing down the scaling up time [2] Another pitfall

of this process is the slower start-up due to the need of ﬁlling the vessel up to a minimum level before starting the polishing steps Longer start-ups lead to product loss and higher cost [2]

Trang 3

Fig 1 Process alternatives for the integration of PCC with virus inactivation and polishing steps: (A) Process alternative 1: All steps in one chromatography system, (B)

Process alternative 2: the capture and the virus inactivation in a system (on the left), and the polishing steps in another one (on the right), (C) Process alternative 3: a surge vessel is used between the capture and the virus inactivation, and two systems are used

3 Materials and methods

3.1 Materials

Two ÄKTA TM pure 150 units were used to perform all the

calibration and validation experiments Each of the chromatogra-

phy systems is equipped with the following devices: three pumps

(pumps A and B, and sample pump) with inlet valves for each of

them to be able to select different buffers, a column valve with in-

built pressure sensors, a loop valve, an outlet valve, several versa-

tile valves, with which different ﬂow paths can be applied, a con-

ductivity sensor, a pH sensor, and two UV monitors

For the capture step, two protein A resins with different parti-

cle size were evaluated One is mAb Select SuRe TM, with 85 μm in

particle diameter, and the other one is mAb Select PrismA TM, with

a particle diameter of 60 μm The buffers and ﬂow rates for the capture step, except for the loading ﬂow rate, were based on [18] The VI was done in a 50 mL Superloop TM provided by GE Health- care Life Sciences (Uppsala, Sweden) The CEX resin was a Capto TM

S Impact, and the AEX resin was a Capto TM Adhere The process information regarding buffers and ﬂow rates for these two steps, was taken from [17] HiTrap TM prepacked columns with a volume

of 1 mL were used for all chromatography steps All columns and resins, along with the chromatography systems, were provided by

GE Healthcare Life Sciences (Uppsala, Sweden) An in-line conditioning between the steps was performed by dilution Regarding the conditioning buffers, 100 mM acetic acid with a dilution factor

of 0.5 respect to the eluted volume, was used to set the pH at 3.5

Trang 4

in the VI step, whereas a buffer with 50 mM sodium acetate and

100 mM sodium hydroxide with a dilution factor of 0.3 was used

to increase the pH after the VI for the loading of the CEX column

The eluate of the CEX column was diluted with a factor of 1, with

a 50 mM sodium phosphate solution at a pH of 6.8 A Cleaning-In-

Place (CIP) was performed after the elution of the columns, with

0.1 M NaOH for the protein-A resins and 1 M NaOH for the rest of

steps

3.2 Methods

3.2.1 Process modelling

The breakthrough curve of the capture step was modelled in

order to simulate and optimize the PCC process, based on pre-

vious implementations of the general rate model in the research

group [19–21] For this particular application, a modiﬁcation was

introduced based on the model from Perez-Almodovar and Carta,

2009 [12] This model assumes heterogeneous binding mechanism

with fast and slow binding sites The concentration in the mobile

phase and in the particle are described by Eqs.(1) and (2), with

the boundary conditions in Eqs.(1a), ( 1b), ( 2a) and ( 2b), respec-

tively Eq.(3)is the description of the kinetics:

∂c

∂t =D ax∂2c

∂z2−ε v c

∂c

∂z−1−εc

εc

3

r p

k f

c − c p|r=r p

(1)

∂c

∂c p

r2

∂

∂r

r2∂c p

∂r

− 1

εp

∂ (q1+q2)

∂c p

∂r =D k f

∂q i

Where is the mobile phase mAb concentration, F is the inlet

mAb concentration, p is the mAb concentration inside the parti-

cle, q is the adsorbed mAb concentration D axis the axial dispersion

coefficient, v is the superficial fluid velocity, k fis the mass transfer

coeﬃcient in the particle layer, D effis the effective pore diffusivity,

ɛc is the column void, ɛp is the particle porosity, p is the parti-

cle radius, L is the column length, q max is the maximum column

capacity, K is the Langmuir equilibrium constant, and k is the ad-

sorption rate constant, where i can be 1 (fast kinetics) or 2 (slow

kinetics)

3.2.2 Multi-response experiments

The heterogeneous model contains several parameters that

were obtained in different ways The column void ( εc) and particle

porosity ( εp) were obtained from [22]and were based on isocratic

elution experiments with dextran with molecular weights from 10

to 670 kDa The axial dispersion coeﬃcient D ax using the Peclet

number ( Pe ) correlation [23] The mass transfer coeﬃcient is also

obtained through a correlation based on the Sherwood, Reynolds

and Schmidt numbers [24], where the density and the viscosity are

assumed the same as for water at 20 °C

The rest of the parameters were obtained from frontal analy-

sis of the breakthrough experiments at different mAb concentra-

tions and ﬂow rates The column volume ( V ) was 1 mL both for

mAb Select SuRe and PrismA Flow rates ( F F) of 0.2, 0.5, 1 and 1.5 mL min −1 (30, 75, 150 and 225 cm h −1) were applied at a constant mAb concentration of 0.5 mg mL −1 Several feed concentration values were also tested (0.25, 0.5, 1.7 and 7 mg mL −1) at

a constant ﬂow rate of 0.5 mL min −1 (75 cm h −1) The columns were loaded until the outlet concentration was almost as high as the feed concentration ( at t = t f) The last part of the breakthrough curve until reaching the feed concentration was extrapolated to calculate the total amount of adsorbed protein per resin volume ( κ) as follows:

V c(1−εc)

c F t f−t∫f 0

c|z=L dt

(4)

With the adsorbed concentration for every corresponding mobile phase concentration, the isotherm parameters (the equilibrium constant K and the total column capacity q max in Eq (5) were obtained by ﬁtting the data to a Langmuir adsorption isotherm with the least-square method (Fig S1, in Supplementary Material), where q max is expressed as adsorbed product per volume of resin

κ=q max K c F

To obtain the maximum capacity for the fast and the slow kinetics ( q max,1 and q max,2), a parameter between 0 and 1 was introduced ( w ), where 0 meant that all sites were adsorbed with slow kinetics, and 1 meant the opposite: q max,1 = w max, q max, 2 = (1-w) The effective diffusivity ( D eff), the kinetics constants ( k 1and k 2), and the parameter w were obtained by running a calibration using the MATLAB nonlinear least square curve-ﬁtting solver lsqcurve-

ﬁt Since the sampling frequency of the UV sensor was constant, the amount of experimental points was larger in the longer experiments In order to have a balanced calibration with equal impor- tance for all the experiments, the number of points was adjusted

to 200 for all curves by interpolating the raw data to obtain a new point for each time, which resulted in identical curves as the raw ones but with the same number of points Before this reduction of points, the curves were also smoothed to avoid the noise The spa- tial derivatives were discretized using the Finite Volume Method,

as shown in [25], with 10 particle grid point and 30 axial grid points This is a considerably low number, which leads to an increased numerical dispersion, but it was shown to be enough for a good ﬁtting of the experimental data, and increasing this number would have involved much longer calculation times The resulting ordinary differential equations were solved with MATLAB’s ode15s The value of all the parameters and properties used in the model are presented in Table S1, in Supplementary Material

3.2.3 PCC optimization

The Periodic Counter-current Chromatography (PPC) operation requires synchronization between the columns In a three-column PCC in particular, two columns are loaded while the product is being recovered in the third one [6] That means that the product recovery is at least so long as the loading of the columns In other words, the cycle time ( t cycle), which is the time that takes to com- pletely load a capture column, must be equal or higher than the recovery time, deﬁned as the time to recover the product in the capture step plus any necessary waiting time, which depends on the process alternative Fulﬁlling this constraint would be enough

to make the process work [6] However, to have an optimal process, it is desired to: maximize productivity ( P ), deﬁned as adsorbed product per column volume and time ( Eq (6)), where τ

is the residence time during the loading; and maximize resin utilization ( U ), deﬁned as adsorbed product divided by the maximum adsorbed amount of product at that feed concentration, which de-

Trang 5

pends on the Langmuir adsorption isotherm ( Eq.(7))

P=t cycle c F− ∫t cycle

0 c|z=L dt

U=t cycle c F− ∫t cycle

0 c|z=L dt

τ (1−εc)q max1+K c K F c F

(7)

Apart from the time constraint, the yield ( Y ), deﬁned as the

amount of the adsorbed product divided by the loaded product

( Eq.(8)), is also set as a constraint, in order to avoid loss of prod-

uct breaking through the columns The yield constraint was set to

99% In a second step of the process, two columns are intercon-

nected during the wash so that the non-adsorbed product of the

ﬁrst column gets captured in the second column, and meanwhile

the third column gets loaded Setting a yield constraint also avoids

product loss in this step Both the objective functions and the yield

constraint were calculated at steady state, when the breakthrough

curves from the three columns were constant and equal to each

other

Y=1−∫

t cycle

0 c|z=L dt

The decision variables were the residence time during the load-

ing ( τ), which relates to the loading ﬂow rate, and the fraction

of the breakthrough curve height respect to the maximum level

during the interconnected step ( x f), i.e the step where the two

columns being loaded are interconnected The lower and upper

bounds for these two variables were 0.25 and 5 min for the res-

idence time, and 0.20 and 0.95 for the x f The pressure drop was

also considered, but for the lower limit of the residence time (cor-

responding to the highest ﬂow rate), the pressure drop was less

than the maximum, so no explicit pressure constraint was then

necessary in the optimization

To sum up, the optimization problem consists of the following

elements:

minimize f(x)=−[P, U]∈R2

w.r.t x=

τ, x f

∈R2

s.t. 0.25<τ<5 min

0.20< x f <0.95

Y >0.99

t cycle > t rec

An optimization was run for each process alternative, where the

only difference was the last constraint ( t cycle > t rec), with t rec being

the recovery time The optimization solver was gamultiobj, a func-

tion of the Global Optimization toolbox in MATLAB This method

is based on genetics algorithms, and it allows to run constrained

multi-objective optimization problems to obtain a set of optimal

solutions, the Pareto front The population size was 150, with a

stop criterion based on a normalized function tolerance of 10 −6, a

Pareto fraction of 0.35, and a migration faction of 0.2, which takes

place every 20th generation The ﬁtness and the constraint func-

tions were computed in parallel, thus allowing to reduce the cal-

culation time, which was around 42 h per optimization

3.2.4 Experimental set-up

For the process validation, a chromatography system is used for the capture step, with the three-column PCC, and the virus inactivation, and another system is used for the polishing steps, i.e the CEX and the AEX steps The implementation of a PCC process in an ÄKTA pure system requires the use of versatile valves, which en- able the different flow paths present in this process In Fig.2, the loop valve (LV) determines which column is loaded first (red solid line), and which column goes through the recovery step (blue dotted line) Three versatile valves are used to lead the flow to the next column, to waste or to the virus inactivation loop, which is placed in another versatile valve Pump A is used for wash, equilibration, elution and regeneration buffers, Sample pump is used for the feed, and Pump B is used to dilute the eluate, being the dilution point just before the VI loop

The polishing steps are implemented in another ÄKTA pure system following the same ﬂow path and process concept as in [14– 16] The only difference in the set-up is the use of the column valve In this case, this valve is right after the VI loop (see Fig.2)

By using three different positions, this valve enables the simulta- neous regeneration and equilibration of the CEX and AEX column

by using the pumps A and B for the CEX column, and the Sample pump for the AEX column This is a way of decreasing the process time and eventually increase the productivity In addition, the column valve allows to empty the VI loop onto the CEX column by in- terconnecting the two chromatography systems The Sample Pump

is used to increase the pH after virus inactivation and to condition the eluate from the CEX column before being loaded onto the AEX column In Figure S2, in Supplementary Material, several possible ﬂow paths are shown for a better understanding of the process set-up

3.2.5 Process control

Both chromatography systems are controlled by the research software Orbit [26] Details about how Orbit is applied to an in- dustrial purification case can be found in [14, 15] or [16] In this work, the operation of Orbit is similar, but an additional feature is included so that the two systems can communicate to each other and synchronize Two Orbit programs are created, one for each system, and the synchronization between them is based on flags in form of binary communication When one of the systems is finished with a process, the corresponding Orbit program sends a flag to the other Orbit controlling the other system, and it remains waiting for another flag from the other Orbit When the other system is also finished with its task, its corresponding Orbit sends a flag too Once both Orbit programs have sent a flag to each other and interpreted the other system’s flag, they are synchronized, and the overall process can continue This process is repeated every time there are parallel processes, and both systems must synchronize with each other to continue to the next step

3.2.6 Analytics

Each ÄKTA system includes one conductivity, one pH and two

UV sensors at 280 nm that measure continuously inline In the PCC process, a UV sensor is used after the ﬁrst column that is loaded, and the other UV sensor is used for the elution In the polishing steps, a UV sensor is used after each column, and the outlet of the

VI loop is also detected by one of the UV sensors in the polishing ÄKTA system

Additionally, the pool from the AEX column is collected every cycle, and for the last cycle the pool from the capture step is also collected to check the concentration, and then be able to calculate the experimental resin utilization, yield and productivity of the capture step, and compare it with the model to validate the obtained optimal solutions All samples taken were measured on

an ÄKTA pure 150 system by injecting a known volume onto a

Trang 6

Fig 2 Process diagram of the PCC process integrated in a downstream process with two systems: one for PCC and virus inactivation (on the right area), and another one

for the polishing steps (on the left area) The red solid line represents the raw material being loaded onto capture columns 1 and 2 (C1 and C2) Capture column 3 (C3)

is washed, eluted and regenerated (blue dotted line) On the left, the CEX column is eluted and the product is directly loaded onto the AEX column (blue dashed line) and collected, and the sample pump is used to dilute the stream between the two columns (green dashed line) Versatile valves (VV), a loop valve (LV), a column valve (CV) and an outlet valve (OutV) are used to define the flow paths Grey lines represent inactive flow paths On the right, a simplified block diagram is shown for an easier understanding of the flow paths (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

1 mL mAb Select PrismA column and measuring the elution chro-

matogram with a UV sensor at a wavelength of 280 nm The sam-

ple concentration was calculated as the area under the eluted peak

divided by the injected volume The extinction coeﬃcient used was

1.4 (mg/mL) −1 cm −1and it was taken from Maity et al., 2015 [27]

4 Results and discussion

4.1 Model calibration

The aim of using a mechanistic model was to get a good estima-

tion of the breakthrough curves under a variety of conditions with

different mAb concentration and residence times In Fig.3, it can

be seen that the breakthrough curves are well ﬁtted for both resins

except for the lowest ﬂow rate At higher residence times (low

ﬂow rates), the General Rate model deviates more from the exper-

imental breakthrough curve, something that was already shown by

Hahn et al., 2005 [28] and Perez-Almodovar and Carta, 2009 [12],

which the model used in this work is based on

Fig.3also shows a good breakthrough curve ﬁtting for all con-

centrations, except for the lowest concentration Similarly to the

deviation shown with high residence times, this discrepancy be-

tween the model and the experiment corresponding to the lowest

mAb concentration has also been observed in the model calibra-

tion performed by Perez-Almodovar and Carta, 2009 [12], where

they obtained a good ﬁtting for the early rising part of the curve,

but a big deviation for the rest of the curve Other authors [28, 29]

have obtained similar results with either higher effective diffusiv-

ity or lower binding capacities than expected for the low protein

concentration case

Despite the deviations in the model for the lowest ﬂow rate and

mAb concentration, it was not expected to affect the optimization

of the PCC process On the one hand, although there are deviations at the lowest feed concentration (0.25 mg mL −1), the area under the curve, which is what is used to calculate the objective functions and the yield constraint in the optimization, is approximately the same for the simulated and the experimental data On the other hand, although for the lowest ﬂow rate (0.25 mL min −1) there is a larger deviation, the simulation provides a more conser- vative solution than the reality, since the breakthrough curve ap- pears earlier than in the experimental results Regarding the suit- ability of the model for the simulation of a multi-column process like PCC, it has been shown that model calibration from batch experiments can be used to predict the performance in a continuous multi-column process [7, 8]

4.2 PCC scheduling

The choice of the integration approach affects the product recovery time, as seen in Fig.4, where Gantt diagrams are shown for the three process alternatives Process alternative 1 has the longest recovery time because the whole downstream process is run in one system Therefore, due to the lack of enough pumps, all steps must be run in series, thus making the total recovery time longer

In process alternative 2, where two systems are used, the capture step and the rest of the steps are run in parallel, thus reducing the recovery time Finally, process alternative 3 has the lowest recovery time, since the PCC and the rest of steps are de-coupled or de-synchronized due to the presence of the surge vessel, i.e., when the product recovery is done in the capture step, the eluate from the protein-A resin will be hold in the surge vessel (after a pH ad- justment to avoid mAb aggregation during the hold-up time), and the next PCC cycle can then be run directly, without the need to wait for the polishing steps to ﬁnish When the virus inactivation

Trang 7

Fig 3 Calibration of the general rate model: (A) Breakthrough curve fitting at different flow rates and at a constant concentration of 0.5 mg mL −1 for mAb Select PrismA (black curves), and (B) mAb Select SuRe (grey curves) (C) Breakthrough curve fitting for different mAb concentrations and at a constant flow rate of 0.5 mL min −1 for mAb Select PrismA, and (D) mAb Select SuRe The dots are experimental points and the solid lines are fitted curves corresponding to the general rate model

loop is ready to be ﬁlled, which is determined by the scheduling

of the polishing steps, as shown in Fig.4C, the surge vessel will be

emptied Therefore, the ﬁlling frequency of the surge vessel is de-

termined by the PCC cycle time, and its emptying depends on the

scheduling of the virus inactivation and the polishing steps

As mentioned before, the cycle time must be higher than the

recovery time In other words, process solutions with a PCC cycle

time lower than the recovery times stated in Fig.4for each pro-

cess alternative, are unfeasible That means that the PCC cycle time

must be chosen so that it is at least as high as the recovery time,

and the loading ﬂow rate must then be adapted so that the yield

is kept within the constraint for that particular cycle time

A possibility to reduce the recovery time could be to duplicate

the virus inactivation loop and run a semi-continuous virus inacti-

vation where a loop is ﬁlled while the other one is emptied, like in

Pall’s Cadence TMVirus Inactivation System [30] However, in Fig.4,

it can be seen that the time for the polishing steps is higher than

the one for the virus inactivation Therefore, the duplication of the

VI loop would not involve any improvement in the recovery time

in the long run, unless the polishing columns were also duplicated

Other options include the implementation of a fully continuous

virus inactivation process with, for example a packed bed reactor

[31] or a Jig in a Box (JIB) approach [32] The impact of this im-

plementation on the recovery time would depend on the way of

integrating the PCC with the continuous virus inactivation, but the

length of the polishing steps would still limit the minimum overall

recovery time

4.3 PCC optimization

The result of the two-objective optimization is a set of optimal solutions with different values of productivity and resin utilization, the so-called Pareto front Solutions with higher productivity have higher loading flow rate in order to treat more material in less time, but this makes the breakthrough curve flatter [12, 13, 22] and it forces the cycle to be shorter and the resin utilization to be lower to keep a high yield On the contrary, solutions with high resin utilization have higher cycle times, and lower loading flow rate and productivity ( Fig 5) The optimization was executed for different process conditions to evaluate the effect of several parameters of the process

4.3.1 Effect of the integration approach

As explained before, the integration approach inﬂuences the minimum cycle time This means that the solutions with a lower cycle time than the recovery time for a particular integration alternative, are unfeasible That is what is shown in Fig.5A Process alternative 1 has the highest recovery time (184.1 min) This implies that the points with higher productivity (with lower cycle time) are unfeasible for this process alternative, and only the solutions with high resin utilization are viable in this case Process alternatives 2 and 3 have lower recovery time (118.4 and 60.1 min, respectively), thus the range of viable solutions is broader It is therefore shown that the higher the recovery time, the lower the process ﬂexibility and freedom to choose between high resin utilization

Trang 8

Fig 4 Gantt diagrams for the three integration alternatives: (A) Alternative 1, integrated PCC process in one system, (B) Alternative 2, integrated PCC process in two systems,

and (C) Alternative 3, PCC process with a surge vessel in two systems (For interpretation of the references to color in the ﬁgure legend, the reader is referred to the web version of this article.)

and high productivity solutions The optimal Pareto solutions for

process alternative 3 are equivalent to the ones that would be ob-

tained for an optimized capture PCC alone without the rest of the

downstream process Therefore, as evidenced in Fig.5A, the opti-

mal design of a PCC process must be done in a holistic approach,

i.e., taking into account the process integration with the polishing

steps already in the optimization problem Not doing so may lead

to optimal solutions that are feasible when running only the cap-

ture step but are unfeasible when the whole downstream process

is implemented, if no surge vessel is used

4.3.2 Effect of the feed concentration

The inlet loading concentration affects in different ways Firstly,

the equilibrium adsorbed product concentration ( κ) is affected

by the mobile phase concentration according to the Langmuir

isotherm ( Eq.(5)) This means that at a higher concentration, the

adsorbed concentration at equilibrium is higher and therefore the

amount of product that can potentially be loaded in each cycle is

generally higher Secondly, a higher inlet concentration increases

the productivity, because a higher amount of product is being

loaded by amount of time and volume In addition, it can also

improve resin utilization, because a higher concentration means

that a lower ﬂow rate can be applied to load the same amount of

product as in a low concentration process, thus making the break-

through curve sharper, which leads to a higher resin utilization In

Figure5, several Pareto fronts for concentrations ranging from 0.25

to 2 mg mL −1 are shown It is interesting to notice the decrease

in the slope of the fronts for an increasing concentration For the

high concentration cases, a small increase in loading ﬂow rate im-

plies a bigger increase in productivity due to a higher amount of

product per volume being loaded, but the reduction of the resin

utilization due to this ﬂow rate increase is small On the contrary,

for the low concentration solutions, in order to achieve a signiﬁ- cant rise of the productivity, a higher increase in ﬂow rate must

be applied, with the consequent large sacriﬁce in the resin utilization This behavior justiﬁes the decrease of the slope at higher inlet concentrations

Another interesting fact is that the Pareto curves get increas- ingly flatter when approaching a very high resin utilization In these operating points, the flow rate and the percentage of the unutilized resin are very low That implies that a greater decrease of flow rate (with the corresponding drop of productivity) is needed to get a slightly higher resin utilization This is more pronounced at higher concentrations because this flow rate decrease affects more the productivity than if the concentration was lower The lower flow rate that it is needed to apply in the process

as a result of a higher concentration, affects the choice of integration approach As shown in Fig 5A, at 1 mg mL −1, only a few operating points are feasible for process alternative 1, and at

2 mg mL −1 only process alternative 3 is feasible for the solutions

of the Pareto front The reason the optimization method cannot find viable points for alternatives 1 and 2 at 2 mg mL −1 is because it would require a very low loading flow rate to avoid product breakthrough with a cycle time higher than the recovery time for these two process alternatives But there is a low limit in the loading flow rate in the simulation of the process set in 30 cm h −1, because the model was not calibrated for lower flow rates Simu- lations of the process at lower flow rates than the model was calibrated for, would have provided unreliable solutions It should be noticed that, despite this fact, it is possible to run process alternatives 1 and 2 at high concentrations and very low flow rates experimentally, but the simulation and prediction of these processes would be unreliable and further model calibration for lower flow rates would be needed

Trang 9

Fig 5 Pareto fronts with optimal solutions (A) Three different integration alter-

natives (ﬁlled, shaded and crossed points) at four load concentrations (0.25, 0.5, 1

and 2 mg mL −1 ) for mAb Select PrismA (circles) and mAb Select SuRe (squares) (B)

Three yields: 95%, 98% and 99%, for mAb Select PrismA at a load concentration of

0.5 mg mL −1 (C) Four operation modes: 3-column PCC, 1-column batch, 2-column

sequential batch and 3-column sequential batch, for mAb Select PrismA at a load

concentration of 0.5 mg mL −1 The legend of the process alternatives is the same

for all the panels

4.3.3 Effect of the chromatography resin

The optimization was run for two different resins: mAb Select PrismA and mAb Select SuRe, with particle diameters of 60 μm and 85 μm, respectively A lower particle diameter allows for a better mass transfer due to a shorter way from the particle surface

to the adsorption sites [28] This is translated in a sharper breakthrough, which in turn leads to a higher resin utilization for equal flow rate or the possibility to run at higher flow rates, thus increasing the productivity, without sacrificing the resin utilization too much In addition, mAb Select PrismA has a higher capacity, as shown in Table S1 Therefore, it is expected that this resin performs better than mAb Select SuRe In Fig.5A, it is confirmed that mAb Select PrismA has a better compromise of productivity-resin utilization for all the concentrations It is remarkable that the difference between the two resins is bigger at higher productivities This

is due to a lower slope of the curve corresponding to mAb Select PrismA, which is explained by the faster mass transfer in this resin For a certain desired increase of productivity, which is carried out

by a corresponding ﬂow rate increase, the sacriﬁce in resin utilization for mAb Select PrismA is lower than for mAb Select SuRe That

is the reason of the different slopes of the two curves, and, in turn,

of the larger difference between the two resins at operating points with higher productivity

Considering two operating points with the same resin utilization for both resins, the cycle time is higher for mAb Select SuRe than for mAb Select PrismA That means that the number of feasible solutions for the process alternatives 1 and 2 is slightly higher with mAb Select SuRe, because these alternatives, which have higher recovery time, beneﬁt from an increase of the cycle time For example, as it can be seen in Fig.5A, at 0.25 mg mL −1 and at a resin utilization of around 63%, an optimal solution can be operated with process alternative 1 in the case of mAb Select SuRe, but no solution would be feasible with the process alternative 1 at that resin utilization with mAb Select PrismA

4.3.4 Effect of the yield

The optimization was run for three yields: 95%, 98% and 99%, and the three Pareto fronts corresponding to the resin mAb Select PrismA are shown in Fig 5B As expected, if the yield is lower, the productivity and the resin utilization are higher A process with very high yield implies that the losses due to product breakthrough must be very low, which means that the process must be run at a lower velocity to get a sharper breakthrough curve, thus reducing the productivity, or ﬁnish the cycle earlier, with a corresponding drop of the resin utilization Besides, it is shown that the yield does not signiﬁcantly affect the feasible range of optimal points for each integration alternative, since for the three yields there are optimal solutions that are feasible for the three process alternatives

Remarkably, the difference between the curves is smaller in the solutions with higher resin utilization, whereas this difference is more pronounced in the solutions with higher productivity This is due to the different slopes of the breakthrough curve in both cases

At lower yields, a longer loading can be applied because the al- lowed amount of product loss is higher, and therefore higher resin utilization can be achieved But if the breakthrough curve is very sharp, the loss of product will be too high at a slight increase of the loading time Therefore, the beneﬁt of reducing the yield constraint does not lead to a signiﬁcantly higher resin utilization in that case For that reason, the solutions with higher resin utilization, which have a sharper breakthrough curve, do not differ much

at different yields, while in the high productivity solutions, the difference in yield is more signiﬁcant

Trang 10

Fig 6 Chromatogram of the capture step (A) and the polishing steps (B) during a PCC run with mAb Select PrismA The shaded areas represent the PCC cycles

4.3.5 Comparison of PCC and batch chromatography

Periodic counter-current chromatography enables to treat a

continuous stream, but it also provides higher productivity and

resin utilization for the same yield, in comparison to batch chro-

matography [7, 8] For this reason, it is interesting to see the differ-

ences between the studied 3-column PCC and a traditional batch

chromatography process ( Fig.5C) In addition, two sequential batch

processes with 2 and 3 columns, respectively, are also considered,

to compare PCC with other simpler periodic processes In a se-

quential batch process, one column is always being loaded, and

the product recovery is carried out in the other columns, just as

in PCC, with the difference that there is no interconnection be-

tween the columns A multi-objective optimization was solved for

the four cases, considering there is no limitation due to the inte-

gration with the rest of the downstream processing steps, since the

effect of this limitation is already shown in Fig.5A and discussed

in Section 4.3.1 The volume for each column was assumed to be

the same, therefore the total resin volume depended on the num-

ber of columns

As expected, PCC provides the highest productivity and resin

utilization, since the PCC process can be run at higher ﬂow rates,

compared to the batch processes, without compromising the yield

or the resin utilization This is due to the interconnection of the

two columns, which avoids losing the product that is not adsorbed

in the ﬁrst column On the other hand, the 1-column batch pro-

cess performs better than the sequential batch processes, as shown

in Fig 5C This is because the same amount of product can be

loaded in the 1-column batch process and in the sequential pro-

cesses, but the number of columns is different Since the produc-

tivity is deﬁned by the total resin volume, and the sequential pro-

cesses have more columns, the productivity is, in general, higher

for the 1-column batch process However, it is noteworthy that the sequential batch processes can achieve a higher productivity than the 1-column batch process, although at a lower resin utilization,

as seen in Fig.5C The reason is that in the sequential processes, the wash, elution and CIP steps are performed at the same time

as the loading, whereas in the 1-column batch process, these steps are run after the loading Therefore, in some of the solutions, the processing time in the sequential processes is signiﬁcantly shorter than in the 1-column batch process, thus compensating the fact that more columns are used

Fig.5C clearly shows that, in the conversion from batch to continuous capture, the choice of the process is very important While

a PCC process may be a more complex alternative to implement than a sequential batch process, it provides almost 5 times more productivity at a constant resin utilization (in Fig 5C, at a resin utilization of approximately 60%, the productivities are circa 0.38 and 0.08 mg mL -1 min -1, respectively for the 3-column PCC and the 3-column sequential batch process)

4.4 Experimental validation

The process was implemented with the process set-up shown in Fig.2 The process was run for the two resins, and a solution with similar resin utilization was chosen from the Pareto front of each resin Both processes were run at 0.5 mg mL −1 The column volume of the protein-A resin was 1 mL, and the rest of columns were sized according to the same procedures used in [14] Regarding the process integration, alternative 2 was chosen Compared to alternative 1, where the whole downstream process is run in one system,

it was simpler and more ﬂexible to carry out all the steps in two systems than trying to ﬁt every step in one system, since the num-

Tiêu đề	Optimization Study on Periodic Counter-Current Chromatography Integrated in a Monoclonal Antibody Downstream Process
Tác giả	Joaquín Gomis-Fons, Niklas Andersson, Bernt Nilsson
Trường học	Lund University
Chuyên ngành	Biochemical Engineering
Thể loại	Research Article
Năm xuất bản	2020
Thành phố	Lund

Định dạng
Số trang	13
Dung lượng	2,59 MB