A regressive approach to the design of continuous capture process with multi-column chromatography for monoclonal antibodies

Although empirical methods have been introduced in the process development of continuous chromatography, the common approach to optimize a multi-column continuous capture chromatography (periodic counter-current chromatography, PCCC) process heavily relies on numerical model simulations and the number of experiments.

Journal of Chromatography A 1658 (2021) 462604

Contents lists available at ScienceDirect

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

Chyi-Shin Chena, b, Fuminori Konoikea, b, Noriko Yoshimotoa, b, c, Shuichi Yamamotoa, b, c, ∗

a Graduate School of Science and Technology for Innovation, Yamaguchi University, Ube,755-8611 Japan

b Manufacturing Technology Association of Biologics, Shin-kawa, Chuo-ku, Tokyo, 104-0033, Japan

c Biomedical Engineering Center (YUBEC), Yamaguchi University, Ube, 755-8611, Japan

a r t i c l e i n f o

Article history:

Received 19 June 2021

Revised 20 September 2021

Accepted 2 October 2021

Available online 8 October 2021

Keywords:

Continuous chromatography

Monoclonal antibody

Periodic counter-current chromatography

Process development

Protein A

a b s t r a c t

Althoughempiricalmethodshavebeenintroducedintheprocessdevelopmentofcontinuous chromatog-raphy,thecommonapproachtooptimizeamulti-columncontinuouscapturechromatography(periodic counter-currentchromatography,PCCC) processheavilyreliesonnumerical modelsimulationsand the numberofexperiments.Inaddition,differentmulti-columnsettingsinPCCCaddmoredesignvariables

inprocess development.Inthisstudy, wehave developedarational methodfor designingPCCC pro-cessesbasedoniterativecalculationsbymechanisticmodel-basedsimulations.Breakthroughcurvesofa monoclonalantibodyweremeasuredatdifferentresidencetimesforthreeproteinAresinsofdifferent particlesizesandcapacitiestoobtaintheparametersneededforthesimulation.Numericalcalculations wereperformedfortheproteinsampleconcentrationintherangeof1.5to4g/L.Regressioncurveswere developedtodescribetherelativeprocessperformancescomparedwithbatchoperation,includingthe resincapacityutilizationandthebufferconsumption.Anotherlinearcorrelationwasestablishedbetween breakthroughcut-off (BT%)andamodifiedgroupcomposedofresidencetime,masstransfercoefficient, and particlesize.By normalizingBT%with bindingcapacityand switching time, thelinearregression curveswereestablishedforthethreeproteinAresins,whichareusefulforthedesignandoptimization

ofPCCCtoreducetheprocessdevelopmenttime

© 2022TheAuthors.PublishedbyElsevierB.V ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/)

1 Introduction

Compared to the traditional batch chromatography in the

downstream process of manufacturing monoclonal antibodies

(mAbs), continuous chromatography for the capture process aims

to reduce the production cost by decreasing the amount of column

medium and buffer required, and shortening process time In a pe-

riodic counter-current chromatography (PCCC), multiple columns

are switched in between loading step, post-load wash (PLW) step,

and turnaround steps including wash, elution, clean-in-place (CIP),

and equilibration Two columns are connected in series for loading

while the other column(s) undergoes turnaround cycles The con-

nected tandem columns can thus capture additional product break-

through compared to a single column In order to accomplish the

continuous sample feed, the time for non-loading operation needs

to be equal or shorter than the loading time To prevent product

loss from the outlet column, the loading needs to be controlled

in a threshold range, which is generally defined as 1 – 3 % break-

∗ Corresponding author

E-mail address: shuichi@yamaguchi-u.ac.jp (S Yamamoto)

through (BT%) from the outlet column or the corresponded % in the dynamic binding capacity (DBC) [ 1, 2] Theoretically, the utilization

of column capacity can be increased by PCCC operation compared

to a batch process at the same productivity, and the material cost can be cut down because of the buffer consumption reduction [3– 5]

Commercial apparatus for PCCC with 2 to 16 columns are al- ready available The control system for column switching in PCCC can be categorized as static control and dynamic control Static control switches columns at a fixed time duration, while dynamic control switches columns when a trigger signal is received from an in-line UV detector As the column performance degradation or the variation of the sample feed concentration with time is not consid- ered in static control, dynamic control can better maintain a sta- ble performance if the monitoring is precise and reliable Although the hardware is well constructed and ready for continuous opera- tion, the process parameters including column switching time and the flow rate corresponding to the feed concentration and desired performances are difficult to obtain without pre-knowledge of the columns, feed materials, adsorption isotherms and mass transfer mechanism

https://doi.org/10.1016/j.chroma.2021.462604

0021-9673/© 2022 The Authors Published by Elsevier B.V This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )

Many studies have been published for the methodology in mod-

eling and simulation of PCCC in different column settings, includ-

ing the schematic design and loading strategy [ 2, 6–8] Generally,

the design of the model-based PCCC approaches includes the ad-

sorption isotherm of mAb for protein A chromatography columns

related to the binding capacity, and the mass transfer kinetics

dominated by the (intraparticle) pore diffusion Although analyti-

cal solutions such as constant pattern approximation (CPA) can be

used in batch loading [ 9, 10], numerical calculation is more widely

adopted as the limitation of continuous flow in PCCC requires more

process parameters to be synchronized The modeling of PCCC re-

quires the knowledge of the adsorption, mass transfer, and the op-

timal loading threshold, which can be determined by iterations of

the numerical calculation However, the numerical solution is nec-

essary every time when there are changes in the column or the

feed The number of columns in PCCC will also affect the switching

condition The cost in development in a continuous multi-column

chromatography process is thus higher than a batch operation [11]

In this study, we aim to develop a rational method to assess the

column switching threshold in PCCC Three commercially available

protein A media were used By simulating the loading profile in

PCCC using mechanistic models, the operation conditions with the

maximum productivity can be obtained Series of experiments in

2 column (2C) PCCC and 4 column (4C) PCCC under different

columns and concentrations were conducted for verification The

process performances including productivity ( P), resin utilization,

and buffer consumption were also examined for each PCCC pro-

cess Based on the simulations, the effects of feed concentration

and resin particle size to the loading cut-off BT% were investigated

The results were used to develop a linear relationship between BT%

and new groups including particle size and mass transfer coeffi-

cients in PCCC processes, which can be used to determine the BT%

threshold

2 Theory

For a 2 column-PCCC (2C-PCCC), two columns of the column

cross sectional area, Ac ( = π/4 • dcol2) and the column bed height,

Z are used (the column diameter dcol, the column bed volume Vt

=AcZ and the column void volume Vo = εVt where ε is the void

fraction) The loading can be kept continuous except at the PLW

step, which flushes the sample in the mobile phase in the first col-

umn to the second column to prevent the loss of the sample To

satisfy the constraint that non-stop loading should be kept while

the other column is at non-loading operations (wash, CIP, and re-

equilibration), the loading flow rates in the tandem-column phase

( Fv1) and in the single column phase ( Fv2) can be different By us-

ing a lower Fv2 for the single column, the DBC ( =C0VB/ Vt, where

C0 is the sample feed concentration and VB is the breakthrough

volume) can be increased [12–14] It can also prevent the possi-

ble leakage from the column when the other column is undergo-

ing non-loading operations As a result, the continuous loading is

possible when its duration is matched to the required non-loading

time

To better illustrate a 2C-PCCC, the operation can be divided to

the following four steps

I Starting step: The loading to the connected columns (tandem

column) starts at Fv1and ends at t=t1

II PLW (post-load wash) step: The wash buffer is fed to the

tandem-column

III Non-loading step (turnaround steps): While the first column is

at the non-loading steps, the sample feed is loaded to the other

column at Fv2

IV Loading step: The sample feed is loaded to the (re-connected) tandem column at Fv1for the duration t2C

The steady state PCCC cycle is composed of step ( Ⅱ) to step ( Ⅳ) The general assumptions for each PCCC calculation in this study include (a) constant amount of loading for every cycle, (b) PLW, (c) identical columns in multi-column settings, (c) 100% recovery during elution, and (e) no degradation or fouling in the column and the sample The schematic diagram of a 2C-PCCC process can

be shown as Fig.1(A)

The design variables in a 2C-PCCC process in this study include the time duration at each step ( tnonload, tPLW, and t2Cload), BT% at the switching point, and flow rates Fv1 and Fv2 The non-loading protocol and initial loading flow rate are pre-defined, and BT% at switching point and Fv2are calculated by satisfying the constraints

in the PCCC setup The calculation results for BT% and Fv2 are aimed to achieve the highest productivity while maintaining less than BT1% ( C/ C0<0.01) from the column outlet

During the step I, BT% 1C,max can be calculated by DBC 1% at Fv1

A BT% smaller than BT% 1C,maxis then assigned, and the total mass during loading ( Mtot) and the accumulated mass ( Maccu) in the sec- ond column can be calculated as

M tot=F v1t 1C 0<DBC1%, Fv1(2V t) (1) where t1is the start-up time

M accu=A c

2Z

Z

[εC +(1−ε)C s]dz =

VB

0

C is the concentration in the mobile phase, Cs is the concentration

in the stationary phase, and is the distance from the column in- let Maccu can be also calculated by integrating the concentration

of the breakthrough curve (BTC) of the single column, C( V) from V

= 0 to VB =Fv1t1 (breakthrough volume at the BT% assigned) During PLW, the washed amount from the first column is as- sumed as one column volume material , which is defined as

MPLW = C0Vt As long as Eq.(3)is satisfied, the total loss is con- trolled below BT1%

The tnonload is the time needed for the non-loading steps, and

is equal to the loading time for the second column at Fv =Fv2 As shown in Fig.1(A), the sum of the loading at Fv2 ( M Fv2) and the accumulation should be below the BT1% point at Fv2, which can be presented as

M Fv2=F v2t nonloadC 0≤ DBC1%, Fv2V t− MPLW− Maccu (4) where 0 < Fv2 ≤ F v1, so the loading can be continuous with leak- age less than BT1% within the period of tnonload It should be noted that the limit of the loading capacity changed to DBC 1%at Fv =Fv2

instead of Fv =Fv1 at this step

Finally, the duration of t2Cload can be calculated as

t 2Cload= M tot− Maccu− MPLW− M Fv2

C 0F v1

(5) The column switching time ( tswitch) is thus given by

t switch= t 2Cload+t PLW+t nonload=t C/2= t C/Ncol (6) where tCis the duration of a complete cycle and Ncolis the number

of the columns

Process performance for a 2C-PCCC is defined as shown in Eqs (7) (9)

P=M 2totV − Maccu

RU =M tot− Maccu

Q(1−ε)V t =M tot− Maccu

(SBC)V t

(8)

2

C.-S Chen, F Konoike, N Yoshimoto et al Journal of Chromatography A 1658 (2021) 462604

Fig 1 Schematic illustration of (A) full process from start to cycling phase in 2C-PCCC

Time for each step is labeled in the left with subscripted labels (including PLW, non-loading, and 2C-connected phase), and the loading flow rate is written as F v1 and F v2

in each step (B) 3C-PCCC from starting phase ( t 1 as duration) followed by one non-loading operation (including PLW and 2C-connected phases) (C) 4C-PCCC Loading to the two connected columns can be completed respectively in t nonload with each loading duration up to t PLW + t 2Cload Color layers represent the loading amount at different stages All loading profiles are theoretical representatives

M tot− Maccu = (V NL/V t)

where P, RU, and BF represent productivity, resin utilization, and

buffer consum ption respectively Q is the saturated (maximum)

protein capacity of resin, SBC is the static binding capacity, and VNL

is the total volume of buffer in the non-loading operation Since

( VNL/ Vt) and (SBC) are fixed for the operation, BF is inversely pro-

portional to RU Namely, BF and RU provide the same information

Since the column(s) is available during PLW step for 3C- and

4C-PCCC settings, loading can be completely continuous as shown

in Fig.1(B) – (C) In a complete cycle, the majority of the time is

used for the loading to the tandem column The loading to the sin-

gle column is carried out only during the PLW step Since the du-

ration of PLW is short compared to the overall cycle time in most

cases, loading flow rate can be kept constant for 3C- and 4C-PCCC

even at the single column loading step as long as Eq.(10)is sat-

isfied Only when a column completes CIP and re-equilibrium be-

fore the loading to the connected columns reaches BT1%, it can be

redirected to the tandem columns and let the primary one to be

disconnected

t load= t PLW+t 2Cload=t switch≥ t nonload

Here, tload is the sum of the loading time for the connected

columns (tandem column) and that for the single column Note

that tC=tswitchNcol

Although both 3C- and 4C-PCCC use the tandem column for the

loading, the difference lies in the number of columns for the non-

loading operation When the same non-loading protocol is em- ployed, one column is used for the non-loading operations in 3C- PCCC whereas two columns are used sequentially in 4C-PCCC As the duration of the non-loading protocol is shared by the two columns, it becomes 2 times shorter compared to 3C-PCCC as illus- trated by Fig.1(B)-(C) As Eq.(10)is the main constraint in PCCC,

a shorter turnaround time can allow a faster flow rate in the 4C- PCCC compared to the 3C-PCCC It can also provide more flexibility for a process having a feed in a higher titer or a lengthy cleaning procedure (long non-loading time, tnonload)

Compared to the 2C-PCCC, the number of design variables is less because of the constant flow rate in loading, which is deter- mined by the DBC 1% at Fv When a desired Fv (or residence time, RT) is selected, BT% for column switching becomes the only factor that needs to be obtained

The processes shown in Fig 1(B)-(C) can be broken down to the starting step, the PLW step, and the non-loading step For the starting step Mtotis given by Eq.(11), which is similar to Eq.(1)

The accumulated amount in the second column Maccu can be calculated by Eq.(2) Considering the single column loading in PLW step, the inequality relation for the PLW step is given by Eq.(12),

which is similar to Eq.(3). Similar to 2C-PCCC, MPLW = C0Vtis cap- tured by the second column

To guarantee that loss will not be over BT1% throughout the process, the loading amount to the single column after PLW,

MSL = Fvt2CloadC0 should be given by Eq (13) By having the same criteria for every column switching, loss below BT1% can be

achieved

PCCC simulations can be carried out by using mechanistic mod-

els described in section2.3 Then, the process performance can be

calculated by Eqs (7) (9) Productivity can also be estimated by

Eq.(14)for 3C- and 4C-PCCC

P = F vC 0

N colV t

(14)

2.3 Mechanistic models

The linear driving force equation for the mass transfer along

with the axial dispersion in the mobile phase was used as a

lumped rate model because of its simplicity for faster calculating

speed in PCCC The equation for the mobile phase is given by

∂C

∂t +H ∂C s

∂t = D L∂2C

∂z 2 − u∂C

where u = Fv/( εAc) is the mobile phase velocity, H = (1 −ε)/ ε is

the column phase ratio, DL is the axial dispersion coefficient and

C sis the average concentration of protein in the stationary phase

For protein A chromatography, the Langmuir isotherms are

widely used, which can be inserted into the linear driving force

approximation expressed by Eq.(16)

d C s

dt =K s(C s − Cs)= K s( Q K LC

where Cs,iis the surface (interface) concentration of protein in the

stationary phase, KL is the Langmuir constant, Q is the maximum

concentration, and Ksis the lumped mass transfer coefficient

The constant pattern approximation (CPA) solution for the

Langmuir isotherm is given by Eqs.(17)-(19)[ 10]

R eqlnX − ln(1− X)−(1− Req)

1− Req = K s H K 0(Z/u )



(tu /Z)− 1

H K 0 − 1



(17)

K s=60εpψpD p

K 0d 2

p

Here Req is the separation factor, X= C/ C0, K0 = Cs0/ C0, Cs0 is

Csat C=C0, εpis the particle porosity, ψpis the correction factor,

Dpis the stationary phase (pore) diffusion coefficient, and dpis the

particle diameter

3 Materials and Methods

Clarified supernatant containing monoclonal IgG (mAb) from

fermentation of Chinese hamster ovary cell culture was used as

feed material Purified mAb from the clarified supernatant of var-

ious concentrations were used for measuring the breakthrough

curves

The following three protein A gels (resins or media) of differ-

ent particle diameter dp were used in this study: MabSelect SuRe

(MSS) ( dp = 85 μm) from Cytiva (Uppsala, Sweden), KanCapA 3G

(KC3) ( dp = 75 μm) from Kaneka (Osaka, Japan) and Amsphere

A3 (AA3) ( dp = 50 μm) from JSR (Tokyo, Japan) Prepacked Hi-

Trap columns (0.7 cm i.d × 2.5 cm) were used for MSS OPUS

MiniChrom columns packed with KC3 (0.8 cm i.d × 2 cm) sup- plied from Repligen (Waltham, MA) and self-packed to empty plas- tic columns (0.7 cm i.d × 2.5 cm) were used for KC3 AA3 was self-packed to Cytiva Tricorn empty columns (0.5 cm i.d × 5 cm) Breakthrough curves (BTCs) were measured by ÄKTA pure, ÄKTA pcc (Cytiva), and Contichrom CUBE HPLC 30 system (YMC, Kyoto, Japan), and were used for inverse fitting of isotherm pa- rameters and lumped mass transfer coefficients by ChromWorks (Ypso Facto, Nancy, France) Experiments were carried out at differ- ent concentrations and flow rates until BT% over 70% was reached Details of the column geometry and experimental condition are shown in Table 1

2C-PCCC experiments were performed on the Contichrom sys- tem with the process parameters provided by the embedded soft- ware The column switching was controlled by fixed time 4C-PCCC experiments were carried out on the ÄKTA pcc system Dynamic control was enabled for column switching

20 mM phosphate with 150 mM NaCl at pH 7.2 was used for equilibration and wash operation 100 mM acetic acid at pH 3.2 was used as elution buffer, and 0.1 M NaOH was used for CIP Non- loading protocols were described in Supplementary Material Part

2

3.2 Simulation platform

ChromWorks served as a tool for fitting the Langmuir isotherm parameters and mass transfer coefficients from experimental BTCs

R (Open source software) was used for the calculation of column switching time and related process performances such as produc- tivity, resin utilization, and buffer consum ption rate in PCCC The details of the models and the calculations are described in Supple- mentary Material Part 1

Model-based approach was adopted for process optimization in

a single column batch chromatography as well as PCCC, which requires the two mass transfer parameters, Ks and DL, and the two Langmuir isotherm parameters, Q and KL Ks values were de- termined by fitting the experimental BTCs with the numerical simulation or with the constant pattern approximation given by Eqs.(17) – (19) [ 10] DL was assumed to be 4 times the product

of dpand u[15–18]

R has several packages that can solve PDEs such as ReacTran, deSolve, and rootSolve [ 20, 21] ReacTran has a function of finite difference approximations of the general diffusive-advective trans- port equation, which is in the form as

∂C

∂t =−u∂C

∂z +D L∂2C

∂z 2 + f(t, z, C) (20) Equation (20) can be used as the model equation by

Eq (15) and (16) provided that the source term f ( t, z, C) can

be regarded as the linear driving force Eq.(16) The appropriate boundary conditions were employed for the numerical calculations

of Eq.(20) The validity of the calculation method was confirmed through good agreements between the calculated BTCs by R and those by ChromWorks

The optimization for PCCC was performed with restrictions of loss below BT1% and Eq (10) using self-programmed code with Reactran package in R [22] Different residence times and concen- trations were examined for the optimal switching BT% by iterations for the highest productivity P and resin utilization RU as given by

arg max

BT % {P (BT%), RU (BT%) } (21) Buffer consumption BF values were calculated with the cor- responded BT% The range of residence time was chosen below the maximum flow rate from the Carman-Kozeny equation, where 0.2 MPa was set for the maximum process pressure Loading flow rate ( Fv) was kept constant in single batch, 3C-PCCC, and 4C-PCCC,

4

C.-S Chen, F Konoike, N Yoshimoto et al Journal of Chromatography A 1658 (2021) 462604

Fig 2 Breakthrough curves from experimental data (symbols) and simulation (lines) from ChromWorks or the constant pattern approximation CPA (only in (B)) of (A) – (B)

MSS, (C) – (D) AA3, (E) – (F) KC3 The details of calculation by ChromWorks are shown in Supplementary Part 4

while it was changed in the tandem (connected) column ( Fv1) and

the single column step ( Fv2) in 2C-PCCC to reach maximum pro-

ductivity PLW is included in the non-loading protocols in 2C-, 3C-,

and 4C-PCCC

4 Results

4.1 Breakthrough curves

Breakthrough curves (BTCs) for MSS were well fitted by the nu-

merical solutions using ChromWorks as shown in Fig 2 (A) The

constant pattern approximation (CPA) was also able to fit the BTCs

as shown in Fig 2(B) with the same Ks value The constant pat-

tern approximation was used for the estimation of Ks at different

C0 in later simulations

BTCs with AA3 and KC3 are shown in Fig.2(C) and (D), respec-

tively As AA3 has smaller dp(50 μm) compared to MSS (85 μm),

the lower mass transfer resistance was reflected on the larger Ks

values (0.0015 – 0.0025 −1) As dp of KC3 (75 μm) is in between

AA3 and MSS, the fitted Ks is among the two ( Ks = 0.001 0.002

s−1)

It should be noted that although DBCs simulated by the simple

Langmuir isotherm model are lower than the values by the exact

simulation at short RT such as RT = 1 min, such low DBC values were not used in our PCCC simulation system since they did not meet the requirement given by Eqs (3), (4), (11) and (12)

4.2 Process simulation

The simulation of batch and PCCC was carried out Validation was performed by comparing the PCCC experiments with the sim- ulation results A wide range of process conditions was examined including the feed concentration (2 – 3.5 g/L of IgG) in either 4C-PCCC or 2C-PCCC settings Details of the experiments and the resulted switching time ( tswitch) along with productivity are pre- sented in Table 1 The non-loading protocol was different from resin to resin in 4C-PCCC A lower feed concentration was applied

in 2C-PCCC

All 4C-PCCC experiments were performed first with BT% fixed

at 25% or 30% By applying the condition to our model with cor- responding non-loading protocol, switching time can be obtained

by the method described in section 2 Dynamic control was em- ployed for 4C-PCCC The difference between estimated switching time and actual switching time by UV is less than 6% as shown in Table1 The experimental P values were lower than the simulation values as the time for start-up and end sequences were not con-

Table 1

PCCC experiments conducted for 4C-PCCC a and 2C-PCCC b at RT 2 min Relative errors in parentheses were calculated by comparing simulation (Sim) with experimental results (Exp)

Ncol

Column (dimension)

C0 (g/L)

tnonload (min) BT%

tswitch (min) Productivity (g/L/hr)

4 MSS (0.7 cm id × 2.5 cm) 3.5 38 25 26 27 ( + 3.8%) 23.8 d 26 ( + 9.2%)

4 AA3 (0.5 cm id × 5 cm) 3.5 49 30 35 33 (-5.7%) 24.1 d 26 ( + 7.9%)

4 AA3 (0.5 cm id × 5 cm) 3.5 11.5 30 34 33 (-2.9%) 25.7 d 26 ( + 1.2%)

4 KC3 (0.7 cm id × 2.5 cm) 3.5 38 25 33 32 (-3.0%) 24.9 d 26 ( + 4.4%)

2 KC3 (0.8 cm id × 2 cm) 2 39 30 76.2 74.8 (-1.9%) 20.3 e 20.3 (0%)

a 10 cycles performed

b 2 cycles performed

c Averaged values from cycles

d Overall productivity (start-up time and end time included)

e Calculated values from Contichrom software

sidered in the simulation Since extra holding time will be needed

from the start to the end with additional line-wash step employed

in the PCCC apparatus, slight deviation exists between the overall

productivity and the estimated productivity from Eq (14) Never-

theless, the results show that our model can simulate the concen-

tration profiles in the columns in PCCC, and can be used to derive

the critical operation parameter, tswitch accurately with errors be-

low 6% The model can be applied to different process conditions

such as Ncoland non-loading protocol

4.3 Process efficiency

Since the feed concentration C0 influences the shape of BTC

heavily, it also affects the process parameters in PCCC The product

titer from upstream is increasing with the improvement in fermen-

tation technology for mAbs [23–25], which makes the capability to

process high C0 essential in the downstream process

The maximum P for 2C-PCCC with MSS, AA3 and KC3 were ob-

tained by the simulation The C0from 1.5 to 4 g/L were examined

as the average titer for mAb products is expected to be above 3 g/L

with recent industrial standard [25] Corresponded process perfor-

mances were calculated based on the loading results derived from

R, and the process parameters were listed in Supplementary Mate-

rial Table S3

To further evaluate the benefit of PCCC compared with the

batch chromatography, the two process evaluation parameters

were introduced; relative buffer consumption ( BF ) and relative

resin utilization ( RU ) They were defined as the ratio of the per-

formance in PCCC over batch at the same productivity by Eq.(22)

Fig.3shows BF and RU as a function of P for 2C-PCCC and a batch

repeated cyclic operation (RCO) By comparing the results between

PCCC and RCO, the relative process performances can be calculated

Lower BF ( < 1.0) and higher RU ( > 1.0) imply the superiority of

PCCC

BF = BF PCCC/BF batch

Fig 4 shows RU and BF vs P curves as a function of C0 for

2C-PCCC Each curve has a parabolic to polynomial profile with

an optimal point, the lowest BF or the highest RU The mini-

mum/maximum values were not clearly shown at high C0for MSS

and KC3 as the loading amount due to low DBC is so small that the

short loading time cannot improve the productivity anymore By

using the same non-loading protocols for all three resins, the high-

est RU or the lowest BF values can be linearly correlated with P

as shown in Fig 4 Note that ( BF × RU ) calculated by the cor-

relation equations are constant, which can be easily understood by

Eqs.(8)and (9) From these correlations, the effect of C0can be un-

derstood For instance, the resin utilization can be improved up to

1.3- to 2.5-fold compared to batch mode for AA3 and MSS respec- tively The buffer volume can be reduced by 20% to 60% for low C0, while the difference between PCCC and batch operation reduces if

C0 goes higher The overall trend of BF increases as C0 increases, while RU decreases over P as shown from the slopes In the case

of KC3 shown in Fig 4 (E) – (F), the reduction of buffer can be over 70% when C0< 1.5 g/L Even when C0 = 4 g/L, 40% reduction

in buffer consum ption can still be achieved The resin utilization is increased by a factor of 2 to 3

Although high concentration leads to higher productivity, the trade-off in RU and BF should be considered by examining the RU

and BF vs P curves When BF < 1 or RU > 1, it is more ben- eficial to use PCCC Instead of conducting numerous experiments

or simulations for several C0 values every time, the RU and BF

vs P may be developed with just the extremum of C0 as high lin- earity exists ( R2 > 0.9) based on the results in Fig.4 The trends shown in Fig 4indicate that at lower C0, PCCC has more advan- tages in buffer consum ption and resin utilization compared to the batch operation It is also shown that PCCC is more beneficial when mass transfer rates are low such as large particles In the ideal case where the diffusion mass transfer is very fast, the performance dif- ference between PCCC and batch operation will disappear

4.4 Determination of breakthrough threshold in PCCC

We have already shown that the DBC normalized by the SBC,

E = DBC/SBC is a linear function of the dimensionless group

F = dp2/[ Ds( Z/ u)] [ 16, 19, 26]

SBC = f (F )= f



d p2

D s(Z/u )



(23) where Dsis the stationary phase (pore) diffusivity The E -F lin- ear relationship was used in batch operation with DBC 10%or DBC 1% [ 26] However, as the switching BT% in PCCC is influenced by other factors such as the number of columns and the non-loading proto- cols, the application of Eq.(23)to PCCC is not straightforward The important parameter for PCCC, BT%, changes with the load- ing flow rate Although automatic iterations can be used to find the optimal BT%, calculation time heavily depends on the compu- tation power and algorithms applied To improve the process de- velopment speed and the accessibility for general users of PCCC,

we examined the simulation data for different protein A media in- cluding MSS, AA3, and KC3 with the same non-loading protocol (Supplementary Material Table S2) at different C0 to find a good correlation for BT%

As the switching BT% in PCCC over different RTs is not a con- stant DBC x%, BT% was normalized as BT% norm with SBC and tswitch After replacing Ds/ dp2 with dp2Ks/60, the following relationship

6

C.-S Chen, F Konoike, N Yoshimoto et al Journal of Chromatography A 1658 (2021) 462604

Fig. 3 The process performances including RU (open symbols) and BF (filled symbols) over P from AA3 at C 0 = 2 g/L of 2C-PCCC (circle) and batch RCO (square) The following non-loading protocol was used 1) equilibration 3CV, RT = 1 min 2) PLW 2CV, RT = 0.5 min, 3) Wash 2CV, RT = 0.5 min, 4) Elution 4CV, RT = 0.5 min, 5) CIP 3CV,

RT = 1min

Fig. 4 RU ∗ and BF ∗vs P (A) – (B) MSS, (C) – (D) AA3 and (E) – (F) KC3 Comparison was made between values from 2C-PCCC and batch processes Regression curves were

developed among the optimal points (filled or bold) with the highest RU ∗ or lowest BF ∗ from each concentration The mass transfer coefficients used are shown in Table S3

Fig 5 Plots of BT% norm (y-axis) over the new functional group F ’ (x-axis) from three protein A media (A), (C), and (E) are for different column PCCC settings and feed

concentration as marked The scatter plots of switching time over RT from the same dataset are illustrated in (B), (D), and (F) with the same order The linear correlation of BT% norm and F ’ are labeled with regression coefficient R 2 The mass transfer coefficients used for the simulation are labeled for each resin

can be derived between the normalized BT% and a new modified

group ( F’)

BT%norm=(SBCBT)t %switch = f( 60

ε(RT)K sd p2

)= f(F ) (24)

In an example using 3.5 g/L of IgG as feed for the three different

resins in a 3C-PCCC setting, results from Eq.(24)is shown in Fig.5

(A) – (B)

A linear relation of BT% norm and F’ was observed for the three

different media with R2 = 0.97 The points that diverged out from

the linear regression are the extreme values at the lowest RT in

3C-PCCC Since higher RT is usually adopted in actual processes,

the outliers in Fig.5(A) could be possibly excluded from the linear

trend The switch time can also be expressed as a linear function

of RT as shown in Fig.5(B) As a result, the BT% at different con-

ditions can be estimated by the relationship in Eq.(24)once the

regression has been obtained The comparison of BT% from the re-

gression (BT% reg) and BT% from numerical simulations (BT% sim) at

RT 2 min are presented in Table2 The differences of BT% between the linear regression and iterative results from numerical solutions are within 10% As the calculation time of iterations usually takes

up from minutes to hours (depending on the scanning area), the linear regression can reduce significant calculation resources and provide the comparable results The schematic illustration for the determination method of BT% is presented in Fig.6 Although the example was for 3C-PCCC settings, 4C-PCCC will have the same trend since tswitch and BT% will not change with the number of columns, only productivity and performance have dependency on column numbers when we compare 3C-PCCC with 4C-PCCC The linear relationship still holds for C0 = 2 g/L with the same resins and the non-loading protocol as shown in Fig 5 (C) with

R2= 0.98 and similar performance between BT% regand BT% sim The slope of BT% normand F’ becomes steeper for higher feed concentra- tion, which matched to the BTCs ( Fig.2) Higher C0leads to steeper BTCs and shorter VB,1%and makes the switch time shorter, which

is reflected in Fig.5(D)

8

C.-S Chen, F Konoike, N Yoshimoto et al Journal of Chromatography A 1658 (2021) 462604

Table 2

Comparison of BT% from linear regression and numerical simulation in 3C-PCC at C 0 = 3.5 g/L under RT = 2 min

Resin SBC (g/L) a tswitch (min) F’ ( μm −2 ) BT% norm (%Lg −1 min −1 ) BT% reg (%) BT% sim (%) Error (%) b

a Calculated from fitted Q, V t , and ɛ

b (BT% reg − BT% sim )/BT% sim

Fig 6 Steps for determining BT% for PCC switching from regressions developed in Fig 5

In 2C-PCCC, the starting RT cannot represent the whole load-

ing since the flow rates may be different during the single column

loading ( Fv2) and the tandem column loading ( Fv1) The tswitch in

Eq.(24)was replaced by t2C, which is the loading time for the tan-

dem column The modified BT% normis re-written to Eq.(25)

BT%norm=BT%/ SBC/ t 2C= f( 60

ε(RT)K sd p2

)= f(F ) (25) With the same non-loading protocol, results for BT% norm in

2C-PCCC are shown in Fig 5 (E) By considering the two flow

rates with t2C, a linear relationship between BT% norm and F’ with

R2 = 0.96 was obtained The slope was steeper than the one in

3C-PCCC with the same C0 By reducing Fv2 for loading, DBC in-

creases and elevates the possible loading amount in the column

Hence, the BT% for column switching can be increased in 2C-PCCC

although t2Cin Fig.5(F) is similar to tswitch in the 3C-PCCC

By expressing a normalized BT% as a function of F’, a general-

ized linear relationship can be acquired across resins in different

particle diameters without dependency on column geometry with

the same non-loading protocol applied As shown in our previous

study [19], by using such resin properties as particle size, SBC, and

mass transfer coefficient (pore diffusion coefficient), PCCC operat-

ing conditions can be easily estimated The linear relationship can

be constructed under different feed concentrations or the number

of columns in PCCC settings including 2C-, 3C-, and 4C-PCCC

5 Discussion

In manufacturing of bio-based drugs, continuous integrated

process has become a production strategy enabling agile produc-

tion of high quality drugs with less regulatory oversight [ 27, ]

PCCC is a well-established method for continuous capture chro-

matography [1–8] However, compared with the standard batch capture chromatography, choosing the proper operating conditions and/or column properties is difficult By replacing experimental runs with in silico simulations based on mechanistic models, pro- cess performances and operation parameters can be determined in

a systematic way

Design strategies for PCCC processes that match to the current available PCCC instruments were presented in the study The meth- ods of modeling and process simulation for repeated cyclic batch operation (RCO) and PCCC were developed

Among the process performances, productivity P describes the amount of product being processed per volume of resin per unit

of time, which can connect to the time needed to process a cer- tain amount of the material As buffer consumption BF is defined

as per liter of buffer required to process per gram of mAb, it can

be related to buffer cost easily Resin utilization RU, on the other hand, expresses the amount of product can be loaded per vol- ume of resin per cycle, which correlates to the cleaning cycles on the resin affecting its lifetime As easily shown by Eq (9), BF is inversely proportional to RU. While protein A resin is expensive, higher resin utilization can lead to lower frequency of the replace- ment of resin hence reducing overall production cost [28–32] Al- though PCCC cannot eliminate the tradeoff between P and other performances, it is possible to achieve better balance in RU or BF

while maintaining the same or higher P compared to batch opera- tion

Previous studies [5–7] have also shown that P of PCCC is not higher than that of batch operation whereas BF of PCCC can be re- duced at the same P In this study effect of feed concentration C0, and mass transfer coefficient (particle diameter, dp) on the relative productivity and buffer consumption values were shown Higher C0

and smaller dp resulted in higher P and lower BF, which are simi-

lar to the repeated batch cyclic operation (RCO) [19] However, the

benefit of PCCC over RCO decreases with decreasing dp In addi-

tion, the column back pressure p increases with 1/ dp2 [19] The

calculations in this study were carried out with the column height

Z = 2 –5 cm For such short columns, p was below 0.2 MPa at

residence time RT = 1 min even for dp = 50 μm Process-scale

chromatography column height is longer than 10–15 cm As p is

over 1 MPa at RT = 1 min for Z = 15 cm with dp = 50 μm, longer

RT (lower flow velocity) may be needed to lower p as the pro-

cess pressure It is also important to keep in mind that higher P

with small dp results in shorter tC, which in turn requires a larger

number of cycles Therefore, understanding the fouling mechanism

to develop a proper cleaning protocol or monitoring the fouling

is essential for a stable PCCC operation [33–35] Although column

fouling is not considered in the present study, it is possible to in-

troduce a safety factor (e.g 0.9) in the normalized BT% or lower

the BT% from 1% to 0.5 – 0.8% depending on the expected resin

life cycle, and develop the corresponded linear correlation next

In our 2C-PCCC experiments (five runs), the column performance

did not change after 72 hour operation according to the yield

( >80%) measured by the UV absorbance at 280 nm and the pu-

rity ( >95%) determined by size exclusion chromatography with a

TSK G30 0 0SW XLHPLC column (data not shown)

Another important point to notice is that the fitting of BTC at

shorter RTs was not good compared with the BTC at longer RTs as

shown in Fig.2 Therefore, simulated results for high P values in

Fig.4are not very precise However, the trends shown in Fig.4do

not change even when somewhat different fitted values from dif-

ferent chromatography models are used It was not our purpose

to show how to find the optimum conditions As is clear from

Figs.3and 4, BF increases sharply near the maximum P, and hence

it is not beneficial to choose the maximum P conditions

Conclusion

We have developed a method for designing the PCCC conditions

based on the linear driving force mechanistic model with the mass

transfer coefficients and the Langmuir isotherm parameters The

performance between PCCC and the batch operation were com-

pared by the relative buffer consumption BF and resin utility RU

values as a function of productivity, P Linear correlations were ob-

served between BF or RU and P

We have also proposed a useful linear correlation between the

normalized BT% and F’, which includes binding capacity, the dura-

tion of non-loading protocols, particle size, and mass transfer co-

efficient By using this correlation, BT% at different conditions in

PCCC can be determined without the numerical calculations

Credit authorship contribution statement

Chyi-Shin Chen: Modelling, Simulation, Writing –original draft,

review and editing Fuminori Konoike: Experiment design and

run, Writing- review Noriko Yoshimoto: Supervision, Writing –

review & editing Shuichi Yamamoto: Framework of the research,

Funding acquisition, Project management, Writing –review & final

editing

Declaration of Competing Interest

The authors declare that they have no known competing finan-

cial interests or personal relationships that could have appeared to

influence the work reported in this paper

Acknowledgment

This research was partially supported by AMED under Grant Number JP20ae0101056, JP20ae0101058, JP21ae0121015, and JP21ae0121016

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.chroma.2021.462604

References

[1] V Warikoo, R Godawat, K Brower, S Jain, D Cummings, E Simons, T Johnson,

J Walther, M Yu, B Wright, J Mclarty, K.P Karey, C Hwang, W Zhou, F Riske,

K Konstantinov, Integrated continuous production of recombinant therapeutic proteins, Biotechnol Bioeng 109 (2012) 3018–3029, doi: 10.1002/bit.24584 [2] M Angarita, T Müller-Späth, D Baur, R Lievrouw, G Lissens, M Morbidelli, Twin-column CaptureSMB: A novel cyclic process for protein A affinity chro- matography, J Chromatogr A 1389 (2015) 85–95, doi: 10.1016/j.chroma.2015 02.046

[3] O Ötes, H Flato, J Winderl, J Hubbuch, F Capito, Feasibility of using con- tinuous chromatography in downstream processing: Comparison of costs and product quality for a hybrid process vs a conventional batch process, J Biotechnol 259 (2017) 213–220, doi: 10.1016/j.jbiotec.2017.07.001

[4] J Pollock, J Coffman, S.V Ho, S.S Farid, Integrated continuous bioprocessing: Economic, operational, and environmental feasibility for clinical and commer- cial antibody manufacture, Biotechnol Prog 33 (2017) 854–866, doi: 10.1002/ btpr.2492

[5] O Kaltenbrunner, L Diaz, X Hu, M Shearer, Continuous bind-and-elute pro- tein A capture chromatography: Optimization under process scale column con- straints and comparison to batch operation, Biotechnol Prog 32 (2016) 938–

948, doi: 10.1002/btpr.2291 [6] D Baur, M Angarita, T Müller-Späth, F Steinebach, M Morbidelli, Comparison

of batch and continuous multi-column protein A capture processes by optimal design, Biotechnol J 11 (2016) 920–931, doi: 10.1002/biot.201500481 [7] D Pfister, L David, M Holzer, R.M Nicoud, Designing affinity chromatographic processes for the capture of antibodies Part I: A simplified approach, J Chro- matogr A (2017), doi: 10.1016/j.chroma.2017.02.070

[8] R Godawat, K Brower, S Jain, K Konstantinov, F Riske, V Warikoo, Peri- odic counter-current chromatography - design and operational considerations for integrated and continuous purification of proteins, Biotechnol J 7 (2012) 1496–1508, doi: 10.10 02/biot.20120 0 068

[9] V Natarajan, A.L Zydney, Protein a chromatography at high titers, Biotechnol Bioeng 110 (2013) 2445–2451, doi: 10.1002/bit.24902

[10] M LeVan , G Carta ,Perry’s Chemical Engineers’ Handbook: Adsorption and Ion Exchange, McGraw-Hill Professional, New York, 2008

[11] H Mahal, C Stamatis, S.S Farid, Integrated continuous bioprocessing: Costs

of goods versus cost of development, Integrated Continuous Biomanufacturing

IV, Brewster (Cape Cod), Massachusetts, USA, 2019 https://dc.engconfintl.org/ biomanufact _ iv/82/

[12] G Carta, Predicting protein dynamic binding capacity from batch adsorption tests, Biotechnol J 7 (2012) 1216–1220, doi: 10.10 02/biot.20120 0136 [13] S Ghose, D Nagrath, B Hubbard, C Brooks, S.M Cramer, Use and Optimiza- tion of a Dual-Flowrate Loading Strategy To Maximize Throughput in Protein-

A Affinity Chromatography, Biotechnol Prog 20 (2004) 830–840, doi: 10.1021/ bp0342654

[14] R.L Fahrner, H.V Iyer, G.S Blank, The optimal flow rate and column length for maximum production rate of protein A affinity chromatography, Bioprocess Eng 21 (1999) 287, doi: 10.10 07/s0 04490 050677

[15] M.R Ladisch , Bioseparations Engineering: Principles, Practice, and Economics, Wiley, 2001

[16] G Carta, A Jungbauer, Protein Chromatography, Wiley, 2010, doi: 10.1002/

9783527630158 [17] G Guiochon , D.G Shirazi , A Felinger , A.M Katti , Fundamentals of Preparative and Nonlinear Chromatography, Academic Press, 2006

[18] N Yoshimoto, Y Sugiyama, S Yamamoto, A simple method for calculating the productivity of polyphenol separations by polymer-based chromatogra- phy, Biosci Biotechnol Biochem 81 (2017) 812–816, doi: 10.1080/09168451 2017.1283210

[19] C.-S Chen, N Yoshimoto, S Yamamoto, Prediction of the performance of capture chromatography processes of proteins and its application to the re- peated cyclic operation optimization, J Chem Eng Jpn 53 (2020) 689–697, doi: 10.1252/jcej.20we116

[20] K Soetaert, J Cash, F Mazzia, Solving Differential Equations in R, Springer, Berlin Heidelberg, 2012, doi: 10.1007/978- 3- 642- 28070- 2

[21] K Soetaert, F Meysman, Solving partial differential equations, using R pack- age ReacTran, 2009 https://pbil.univ-lyon1.fr/CRAN/web/packages/ReacTran/ vignettes/PDE.pdf

[22] K Soetaert, F Meysman, Package ‘ReacTran, ’ R Doc (2012)

[23] J Xu , M.S Rehmann , X Xu , C Huang , J Tian , N.-X Qian , Z.J Li , Improving titer while maintaining quality of final formulated drug substance via optimization

10