Optimal loading flow rate trajectory in monoclonal antibody capture chromatography

In this paper, we determined the optimal flow rate trajectory during the loading phase of a mAb capture column. For this purpose, a multi-objective function was used, consisting of productivity and resin utilization.

Contents lists available at ScienceDirect

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

chromatography

Joaquín Gomis-Fonsa, b, 1, ∗, Mikael Yamanee-Nolina, 1, Niklas Anderssona, Bernt Nilssona, b

a Department 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 5 August 2020

Revised 23 October 2020

Accepted 23 November 2020

Available online 26 November 2020

Keywords:

Flow programming

Flow trajectory

Protein A chromatography

Monoclonal antibody

Multi-objective optimization

Chromatography scale-up

a b s t r a c t

Inthispaper, wedeterminedtheoptimal flowrate trajectoryduringthe loadingphaseofamAb cap-turecolumn.Forthispurpose,amulti-objectivefunctionwasused,consistingofproductivityandresin utilization.Severalgeneraltypesoftrajectorieswereconsidered,andtheoptimalParetopointswere ob-tained forallofthem.Inparticular, thepresented trajectoriesincludeaconstant-flowloadingprocess

asanominalapproach,astepwisetrajectory,andalineartrajectory.Selectedtrajectorieswerethen ap-pliedinexperiments withthe state-of-the-artproteinA resinmAb Select PrismATM,running inbatch modeonastandard single-column chromatographysetup,and usingbothapurifiedmAb solution as wellasaclarifiedsupernatant.Theresultsshowthatthissimpleapproach,programmingthevolumetric flowrateaccordingtoeitheroftheexploredstrategies,canimprovetheprocesseconomicsbyincreasing productivitybyupto12%andresinutilizationbyupto9%comparedtoaconstant-flowprocess,while obtainingayieldhigherthan99%.Theproductivityvaluesweresimilartotheonesobtainedina multi-columncontinuousprocess,andrangedfrom0.23to0.35mg/min/mLresin.Additionally,itisshownthat

amodelcalibrationcarriedoutatconstantflowcan beappliedinthesimulationand optimizationof flowtrajectories.Theselectedprocesseswerescaleduptopilotscaleandsimulatedtoprovethateven higherproductivityandresinutilizationcanbeachievedatlargerscales,andthereforeconfirmthatthe trajectoriesaregeneralizableacrossprocessscalesforthisresin

© 2021 The Authors Published by Elsevier B.V ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/)

1 Introduction

Monoclonal antibodies (mAbs) are used to treat a wide range

of different diseases, such as rheumatoid arthritis, Crohn’s dis-

ease and chronic lymphocytic leukemia [1] However, mAb treat-

ments can become very expensive due to high manufacturing costs

[2] and low research and development productivity [3] Further

considering the fact that downstream processing may account for

over 60% of the total manufacturing cost of a mAb product, and

that the capture step is crucial to overall process efficiency, im-

provements of this step will have great impact on process eco-

nomics [2]

∗ Corresponding author at: 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), mikael.yamanee-nolin@chemeng.lth.se (M Yamanee-Nolin),

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

(B Nilsson)

1 Joaquín Gomis-Fons and Mikael Yamanee-Nolin are co-first authors with equal

contribution

Most processes for the purification of mAbs are currently op- erated in batch mode, and these processes are simple, robust, and well-known [ 4, 5]; however, they are also inefficient [6-8] To in- crease efficiency, an alternative is to adopt an integrated and con- tinuous bioprocess (ICB), which could lead to higher productiv- ity, lower cost of goods, and higher resin utilization, as shown

in previous implementations and studies [ 7, 9, 10] Most down- stream steps in previous ICB studies are based on multi-column chromatography processes, which for example include sequen- tial multi-column chromatography (SMCC) [11], capture simulated moving bed (CaptureSMB) [12], multi-column counter-current sol- vent gradient purification (MCSGP) [ 13, 14], and periodic counter- current chromatography (PCC) [ 6, 15] In general, these strategies make use of multiple columns, valves and pumps with a sequential operation; thus, this adds an extra layer of complexity to the pro- cess design and operation In addition, there are technology gaps that need to be address before the implementation of these pro- cesses at commercial scale [16], which is why integrated and con- tinuous biomanufacturing is still not prevalent at commercial scale [7]

https://doi.org/10.1016/j.chroma.2020.461760

0021-9673/© 2021 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/ )

Fig 1 Illustrative comparison of the breakthrough curves (BTC) with constant flow

rate (u 1 ) and variable flow rate (u 2 )

Another approach towards increasing efficiency is to apply a

programmed variable flow rate in the loading of the chromatog-

raphy steps, and the underlying idea is illustrated in Fig 1 The

theoretical background is that higher loading flow rates lead to

higher productivity, but they also result in a flatter breakthrough

curve [ 6, 17] As a result of a flatter breakthrough curve, the loading

time must be shortened in order to keep the yield high, thus lead-

ing to a decreased resin utilization Similarly, when a lower flow

rate is applied, the breakthrough curve is sharper, and the resin

utilization for a specific yield requirement is increased In order

to find the optimal balance between high productivity and high

resin utilization, flow programming can be used to find an optimal

flow trajectory Using this technique, a higher flow rate is applied

at the beginning when all binding sites are available, and a lower

flow rate is used to give the protein more time to diffuse into the

pores As shown in Fig.1, the breakthrough curve of the variable-

flow process appears earlier as a result of a higher flow rate at the

beginning, but as the flow rate diminishes, mass transfer in the

column improves leading to less product loss in the breakthrough

Flow programming has been previously used [18], showing a

productivity increase with a variable flow rate profile obtained

with a design of experiments (DoE) approach Lacki [19]has also

demonstrated the potential of flow programming, but their results

also showed that if the flow rate trajectory is not chosen prop-

erly, the productivity could be even lower than in the correspond-

ing constant-flow operation

In order to avoid sub-optimal flow rate trajectories and their

resulting performance in terms of process economics, model-based

optimization can be a useful tool to optimize key process perfor-

mance indicators such as productivity, resin utilization, and yield

This approach has been explored by Ghose et al [20], who ap-

plied a dual-flow rate loading strategy with a variable switching

time and showed that it outperformed single-loading strategies

without requiring any extra equipment or columns This strategy

was expanded later to include any number of constant flow rates

evenly distributed over the loading phase [21] This in-silico study,

which was not based on mAbs but on a model protein instead, fur-

ther highlighted the potential to increase productivity and resin

utilization by modifying the flow rate during the loading phase

of a capture step, whilst retaining the simplicity of the single-

column setup operated in batch mode Further proof of the po-

tential of model-based optimal trajectories in chromatography has

been shown by Sellberg et al [22], who obtained optimal elution

trajectories with variable modifier concentration in ion exchange

chromatography The mass transfer behind this process is differ-

ent from the one behind the loading of a protein A column, but it

shows the experimental feasibility of applying general trajectories

obtained with a computer-aided optimization

Building upon the findings of Sellberg et al [ 21, 22], the pur- pose of the current work is to optimize different flow trajectories

in a protein A step for the capture of mAb with the novel resin mAb Select PrismA TM, and demonstrate their potential in an ex- perimental validation In comparison to the work by Ghose et al [20], we present a more comprehensive study where different flow programming strategies are explored, and we apply the optimal results to a state-of-the-art protein A resin The loading phase in the protein A capture step is often the rate-limiting step in a mAb downstream process [20] Therefore, flow programming was only considered in the loading step To obtain the optimal flow trajec- tories, a model-based multi-objective optimization approach was first applied, utilizing a General Rate model [23], in order to find and compare optimal trajectories for three approaches:

I A nominal approach, applying a constant flow rate

II A stepwise trajectory approach, applying N u >1 decision hori- zons with stepwise flow rate changes

III A linear trajectory approach, applying a flow rate changing lin- early with time

A Pareto front for each of the three approaches was obtained, and they were then implemented at laboratory scale for proof-of- concept The experimental results in combination with the model- based results highlight the potential to improve efficiency and pro- cess economics of the mAb production process using a simple yet high-value solution, i.e., a single-column, batch-mode capture step with variable flow rate during the loading phase The primary ad- vantage compared to an ICB process is that the batch technology and equipment currently used commercially can still be used ap- plying the proposed flow-programming strategies Scalability of the flow-trajectory processes from laboratory scale to pilot scale is cru- cial to be able to maintain the same process through the develop- ment phases of the biopharmaceutical For that reason, the pro- cesses studied were scaled up and simulated to demonstrate that the trajectories can be applied even at a larger scale and are gen- eral at any scale for the resin mAb Select PrismA This is, to be the best of our knowledge, the first time that process scale-up is addressed in relation to flow programming

The remainder of the paper is structured as follows: Section 2 introduces the model-based approach, with the pro- cess model and the optimization problem This is followed by a description of the experimental setup and procedure, and of the scale-up method The results of the model-based optimization, the laboratory experiments and the process scale-up are then presented and discussed in Section 3 The major conclusions are then presented in the final section

2 Material and methods

2.1 Model-based optimization 2.1.1 Process model

The chromatography column was modeled using the General Rate model featuring a heterogeneous binding mechanism with fast and slow sites [23], to simulate and optimize the loading of the capture step The particular model applied in the current work has been previously implemented in Matlab and calibrated suc- cessfully in our previous study [6], using the Finite Volume Method [24] The model was calibrated for several constant flow rates and mAb concentrations to ensure a good fitting for a broad range of conditions The mobile phase and particle concentrations are de- scribed by Eqs 1 and 2, respectively, with boundary conditions specified by equations1a, 1b, 2a, and 2b, and Eq.3describing the kinetics

∂c

∂t =D ax∂2c

∂z2 − v

c

∂c

∂z −1−c

c

3

r p

k f

c − c p|r=r p



(1)

∂c

∂c p

∂t =D e f f1

r2

∂

∂r



r2∂c p

∂r



p

∂ (q1+q2)

∂c p

∂c p

∂r = k f

D e f f(c − c p)atr=r p (2b)

∂q i

∂t =k i



(q max , i − q i)c p−q i

K



(3)

Here, c is the mobile phase mAb concentration, F is the in-

let mAb concentration, p is the particle mAb concentration, q is

the adsorbed mAb concentration, D ax is the axial dispersion coeffi-

cient, v is the superficial fluid velocity, k f is the particle layer mass

transfer coefficient, D e f f is the effective pore diffusivity, εc is the

column void, εp is the particle porosity, pis the particle radius, L

is the column length, q max is the maximum column capacity, K is

the Langmuir equilibration constant, and k i is the adsorption rate

constant, where i can be either 1 or 2, for fast or slow kinetics,

respectively The axial dispersion coefficient was obtained using a

Peclet number correlation [25], the void and porosity parameters

were obtained from Pabst et al [26], and the mass transfer coeffi-

cient was estimated with an empirical correlation [27]

The choice of the chromatography resin has an impact on the

model as the particle diameter and pore size of the resin affect the

mass transfer significantly A higher particle diameter leads to a

longer average distance between the particle surface and the bind-

ing sites, which results in a slower overall mass transfer inside the

particle; and a small pore diameter hinders mass transfer through

the pores by decreasing the effective pore diffusivity [28] There-

fore, a new model calibration and optimization should be carried

out for a different resin

2.1.2 Optimization problem

The main idea behind the optimization problem was to mod-

ify the volumetric flow rate during the loading phase as well as

the duration of the loading phase (decision variables) in order to

improve the process economics, by maximizing productivity and

resin utilization (objective functions) for a specific yield require-

ment (constraint) Two different types of flow trajectories were

employed: a stepwise trajectory with N u decision horizons cor-

responding to constant flow rate levels distributed evenly across

the full duration of the loading phase was employed in Approach

I (a single decision horizon, thus corresponding to the nominal

constant-flow process) and Approach II ( N u > 1 decision horizons),

whereas a linear trajectory was obtained over time and applied in

Approach III The choice of these two types of trajectories resulted

in two slightly different optimization problems The optimization

problem for the stepwise trajectories employed in Approaches I

and II is presented in Eq.4below:

u0, u1, , u N u , t f

Constraints: Y ≥ Y min , D V i=

D V lb ,i , D V ub ,i Here, F is the objective function vector consisting of the nor-

malized productivity, P, and the normalized resin utilization, U;

DV is the decision variable vector containing N u decision variables

for the stepwise constant flow rates, and the total duration of the loading phase, giving a total of N u+1 decision variables; Y rep- resents yield, and Y min is the minimum required yield, set to 99%; and [ D V lb,i , D V ub,i] are the lower and upper bounds for the decision variables, set to [ 0 .2 , 1 .5 ] mL/min for the flow rates and [ 60 , 300 ] min for the loading time The optimization problem for Approach III is presented in Eq 5:

u0, u t f , t f

Constraints: Y ≥ Y min , D V i=

D V lb ,i , D V ub ,i

In this problem, the objectives and constraints were the same

as the ones used in Approaches I and II, but the decision variables were different The decision variables were in this case only the initial and final flow rates, u0 and u t

f, respectively, and the dura- tion of the loading phase, t f, and the resulting trajectory was lin- ear over time The control action at time t, i.e., the volumetric flow rate, u t, was in this approach calculated according to Eq 6

u t=



u t f − u0



t f

Furthermore, for the three approaches, the three key perfor- mance indicators were defined according to Eqs 7-9:

P n= m a

P= P n − P min

P max − P min

(7b)

U n=q m a

U= U n − U min

U max − U min

(8b)

Y = m a

m in

(9)

where m a is the amount of adsorbed mAb, which is determined

as the difference between the amount of mAb loaded ( m in) and the product loss in the breakthrough, calculated by the area under the breakthrough curve; V c is the column volume; and P min, P max,

U min, and U max are nominal minimum and maximum values of the productivity and resin utilization, based on nominal loading pro- cesses at minimum and maximum volumetric flow rates and load- ing phase durations Eq.7adefines the productivity as the amount

of adsorbed mAb divided by the duration of the loading phase and the resin volume Eq.8adefines the resin utilization as the amount

of adsorbed mAb per volume of resin divided by the stationary phase mAb concentration at equilibrium, whose definition is also based on volume of resin Furthermore, Eqs.7band 8bdefine the productivity and resin utilization normalized to the range 0-1 for all operating conditions, which are used in the objective function

Eq.9defines yield as the amount of adsorbed mAb divided by the amount of mAb loaded It should be noted that the key perfor- mance indicators, as applied in the current work, are based on the loading phase of the capture step only, i.e do not include other phases such as elution and CIP, and ignore any remaining mAb in the mobile phase at the end of the loading phase For this rea- son, the way that productivity is defined in Eq.7aresults in higher values compared to how productivity of capture processes is usu- ally reported [ 6, 12], since, in this case, only the process time for the loading phase is included in the definition of productivity For comparison with other processes, the productivity values should be

3

adjusted to include the process time corresponding to the whole

capture step

The optimization problems were solved using a Matlab variant

of the elitist non-dominated sorting genetic algorithm (NSGA-II),

which is available as part of the built-in gamultiobj function The

constraint tolerance was set to machine epsilon, with the function

tolerance set to 10 −6, and the population size was set 300 Using

this kind of global, multi-objective algorithm, a set of Pareto opti-

mal solutions are offered to the user, who can then make a deci-

sion a posteriori regarding how to weigh the objectives [29]

2.2 Experimental setup

2.2.1 Buffers and sample preparation

Experiments were conducted using two different samples: (i)

a 0.48 mg/mL purified mAb solution for a clear illustration of

the results, and (ii) a 0.48 mg/mL clarified supernatant for proof-

of-concept The mAb concentration of the latter was adjusted to

match the concentration at which the experiments with the puri-

fied mAb were performed, so that a direct comparison of the ex-

periments with the two different samples could be done Accord-

ing to the equilibrium data obtained in our previous study [6], the

adsorbed concentration at equilibrium for mobile phase concentra-

tions above 0.5 mg/mL is nearly constant, and the mass transfer

coefficients in the model remain almost constant for higher con-

centrations provided the viscosity does not increase significantly

Therefore, the relationship between the feed concentration and

the time it takes for the product to break through the column is

nearly linear For that reason, almost the same breakthrough curves

would be obtained for equal protein loads in units of mass of prod-

uct loaded per volume of resin, if the residence time is the same

Consequently, it can be assumed that the optimal flow trajectories

are general for any feed concentration above 0.5 mg/mL as long as

the protein load is maintained by adjusting the loading time

The buffers, column volumes and flow rates (except the loading

flow rates) were the ones recommended by the resin manufacturer

for a protein A capture process [30]

2.2.2 Chromatography station setup

In order to carry out the capture experiments using the optimal

trajectories found via model-based optimization, a single ÄKTA TM

pure 150 unit, provided by Cytiva (Uppsala, Sweden), was used

with its standard setup, and it was equipped with the following

devices: two gradient pumps, inlet valves for buffer selection, col-

umn valve with built-in pressure sensor, a fractionator, an outlet

valve, and a sensor package that included a UV, a conductivity and

a pH sensor The sample was injected onto the column with a 100

mL Superloop TM The column was a 1 mL prepacked HiTrap TMcol-

umn with mAb Select PrismA TM resin, from Cytiva (Uppsala, Swe-

den), and the column length and diameter were 2.5 cm and 0.7

cm, respectively

2.2.3 Process control

The ÄKTA pure system used during experiments was controlled

with the Python-based software Orbit, which has been described

in detail elsewhere [31-33] For the particular control problem in

the current work, a function to modify the flow rate based on the

elapsed time was implemented In Approach II, the total load du-

ration ( t f) was divided by N u to obtain equal time horizons with

stepwise constant flow rates The list of flow rates found through

the optimization was specified manually, and used by Orbit to up-

date the flow rate at the start of each horizon In Approach III, the

linear trajectory was approximated by stepwise constant control

actions updated at a sampling rate, i.e 1 Hz, which is much more

frequent than in Approach II Additionally, no flow rate change was

necessary in Approach I, thus resulting in constant flow rate during the whole loading phase

2.2.4 Analytics

The breakthrough curve was detected online with a UV sensor

at a wavelength 280 nm For the experiments with pure mAb, this signal was used to obtain the breakthrough curve in mg/mL using

an extinction coefficient of 1.4 (mg/mL) −1 cm −1 [34] For the experiments with supernatant sample, the breakthrough baseline was above the linear range of the UV detector, which is

20 0 0 mAU, due to the high concentration of impurities that went through the column For that reason, the outlet stream was col- lected in fractions of 2 mL and analyzed offline For the analysis

of the fractions, an ÄKTA Explorer 100 equipped with an autosam- pler was used The autosampler was set up so that 1 mL of each fraction was taken and loaded onto the column A 1 mL prepacked HiTrap TMcolumn with mAb Select PrismA TMresin was used for the analyses The process conditions regarding buffers and flow rates were the same as in the flow trajectory experiments, described above However, the elution time was longer to be able to see the whole elution peak Knowing the injected volume and the extinc- tion coefficient, the concentration of each fraction was calculated with the area of the eluate peak

2.3 Scale-up method

The processes studied were scaled up to pilot scale with a fac- tor of 10 0 0 and simulated to investigate whether the found trajec- tories were generalizable across process scales for the mAb Select PrismA resin A method to scale up the process is to keep the col- umn length constant and increase the diameter, in a way that both the flow velocity and the residence time are kept constant This scale-up method has been proposed by Heuer et al [35] The flow rate trajectories would be converted to velocity trajectories, by di- viding the flow rates by the column section (in this case 0.38 cm 2), and the same velocity trajectories could be used for any process scale In this work, the column length was 2.5 cm, therefore it was not practical to keep the same length at larger scales For that rea- son, another scale-up method is to change both the column diam- eter and length, so that the residence time is kept constant, even

if the velocity is not This method, proposed by Hansen [36], pro- vides flexibility to choose an appropriate length to fulfill a maxi- mum diameter-to-length ratio constraint He shows that the num- ber of theoretical plates, which is an indication of the column effi- ciency, increases at a higher column length and constant residence time, based on a simplified version of the van Deemter equation:

N= A 1

L +C

τ

(10)

where N is the number of theoretical plates, L is the column length, τ is the residence time, and A and C are constant terms in the van Deemter equation In this work, this scale-up method was applied, and an empirical expression for the pressure drop over a packed bed [37]was used to obtain the column length:

P=αvL= αL2

where P is the pressure drop over the column, v is the superficial velocity, which equals the column length divided by the residence time, and αis an empirical constant, which was determined by fit- ting experimental data of pressure drop against velocity provided

by the resin’s vendor At a higher column diameter-to-length ratio, the bed compression increases for a specific velocity and column length due the loss of wall support [37] In turn, this leads to an increase of the empirical constant α For this reason, the experi- mental data used to obtain αcorresponded to a large-scale column

with a diameter-to-length ratio of 50, which was higher than the

expected ratio The value of αobtained was 1.5 •10 −4 bar h cm −2

By solving Eq.11for L, the maximum column length could be cal-

culated as follows:

L max= P maxτ

where the maximum pressure drop over the column ( P max) was

the one provided by the resin’s vendor minus a safety margin of

20%, resulting in a value of 1.6 bar Once the column length was

determined, the column diameter was obtained to achieve the de-

sired column volume while maintaining the residence time

3 Results and Discussion

3.1 Optimization results

The results from the model-based optimization are compiled as

Pareto fronts and presented in Fig 2, in which the nominal ap-

proach – Approach I (black) – is compared with Approach II (gold)

and Approach III (red) in the upper and lower panel, respectively

It should be noted that the productivity and resin utilization re-

sults are presented as actual (not normalized) values, for an easier

comparison with results from other authors As can be seen when

comparing the optimal solutions in the three approaches, there is

potential for improvements by adopting a variable flow rate in-

stead of a constant flow rate, with strikingly similar improvements

resulting from the two trajectory strategies explored in this work

The two strategies presented in Approach II and Approach III al-

ways outperform Approach I, except for the Pareto area at maxi-

mum productivity, where the Pareto fronts collapse into each other

due to the flow rate being set to the upper bound at all times Fol-

lowing from the Pareto fronts, it is possible to improve the loading

phase in terms of productivity and resin utilization to different de-

grees, depending on the point of current operation

Pareto fronts were obtained for Approach II for several num-

bers of horizons ( N u): 2, 5 and 10 For a number of horizons higher

than 5, the performance indicators (productivity and resin utiliza-

tion) did not improve enough to warrant the increased complexity

of the optimization problem, and thus the increased computation

time and resources to solve the problem At the same time, the

results were slightly better for the 5-horizons approach than the

dual-flow rate approach, thus showing an improvement with re-

spect to the strategy presented by Ghose et al [20] For these rea-

sons, the only Pareto front of Approach II considered for compar-

ison with the other approaches corresponds to a number of hori-

zons equal to 5 Regarding Approach III, the linear trajectory was

compared with a quadratic trajectory, with no significant difference

found Due to a higher simplicity, it was decided to consider only

the linear trajectory in the comparison of the three approaches For

comparative reasons, the results for 2 and 10 horizons as well as

the results for the quadratic trajectory are attached in the Supple-

mentary materials section, presented in Figures S1 and S2, respec-

tively

3.2 Experimental validation

The five points highlighted by circles in Fig 2 were selected

for applying the optimal flow rate trajectories in the laboratory A

point from the Pareto front of Approach I, denoted by Case I, was

selected as the nominal case Two points with approximately the

same productivity as that of the Case I and higher resin utiliza-

tion, were selected from the Pareto fronts of Approaches II and III,

and they were denoted by Cases IIa and IIIa, respectively Similarly,

two more points with approximately the same resin utilization and

higher productivity, denoted by Cases IIb and IIIb, were selected

Fig 2 The Pareto fronts generated in the model-based multi-objective optimiza-

tion The Pareto front for the stepwise trajectory with five horizons is presented in panel A; and the front for the linear trajectory is presented in panel B The Pareto front for one horizon (constant flow rate) is plotted in both panels for comparison The points selected for experimental trials are marked by circles, in total five cases: the constant-flow nominal case (Case I), and two cases for each trajectory approach, one with improved resin utilization but nearly the same productivity as in Case I (Cases IIa and IIIa), and one with improved productivity but nearly the same resin utilization (Cases IIb and IIIb)

from the Pareto fronts Experiments using a purified mAb solution were run for all five points to be able to see the breakthrough curve online without the need of offline analyses For proof-of- concept, one of the points (Case IIb, i.e the high-productivity point for the stepwise trajectory approach) was tested with clarified su- pernatant

The optimal flow trajectories are presented in Fig.3, while the maximized productivity and resin utilization values are shown in Fig.4 The constant-flow Case I has a loading duration of 75 min- utes and a flow rate of 0.77 mL/min, resulting in a productivity

of 0.55 mg/min/mL resin and a resin utilization of 31.4%, with a yield of 99.1% The two stepwise trajectories differ from the nom- inal case in the loading duration, and in the flow rate levels In Case IIa, with an improved resin utilization but constant produc- tivity, the loading duration is roughly 5 minutes longer than Case

I (in total 80 minutes), and the average flow rate is 0.77 mL/min, giving a productivity of 0.55 mg/min/mL resin and a resin utiliza-

5

Fig 3 Optimal loading flow rate trajectories The stepwise trajectories plotted in

panel A correspond to Cases IIa and IIb, and the linear trajectories plotted in panel

B correspond to Cases IIIa and IIIb The constant-flow process (Case I) is shown in

both panels for comparison

tion of 34.2%, which means a relative increase in resin utilization

of 8.9%, while the yield is 99.4% Even with an increase in loading

duration, the productivity is nearly the same, and this is primar-

ily due to the increase of adsorbed antibodies onto the resin as an

effect of the decrease of the flow rate towards the end of the load-

ing phase Similarly, but conversely, the loading duration is shorter

by roughly 8 minutes in Case IIb compared to Case I, i.e., in total

67 minutes, with an average flow rate of 0.86 mL/min, resulting in

a productivity of 0.62 mg/min/mL resin and a resin utilization of

31.5%, which means a relative productivity increase of 11.8%, and a

yield of 99.7%

The linear trajectories of Case IIIa and Case IIIb are similar to

their corresponding cases of Approach II, with loading durations

and average flow rates of 81 min and 0.77 mL/min, and 67 min

and 0.86 mL/min, respectively; the resulting productivity and resin

utilization for the two cases are 0.55 mg/min/mL resin and 34.0%,

and 0.61 mg/min/mL resin and 31.2%, respectively The yield is in

both cases above 99%

Given that the performance of the selected points for Ap-

proaches II and III were highly similar, it can be expected that the

trajectories are similar as well However, it seems that the step-

Fig 4 Optimal productivity and resin utilization values for Case I (black), Cases

IIa and IIb (gold), and Cases IIIa and IIIb (red) The laboratory-scale values obtained from the simulation are compared with the ones obtained experimentally with pure monoclonal antibody and with raw supernatant (the latter only for Case IIb) The five cases were also simulated at pilot scale with a scale factor of 10 0 0

wise trajectory does not approximate a linear trajectory In addi- tion, as mentioned, a quadratic trajectory or stepwise trajectories with higher number of horizons did not lead to any significant dif- ference respect to the linear trajectory This may lead to the con- clusion that complex trajectory shapes may not be necessary to achieve a more efficient process, but rather simple yet optimized trajectories are enough to accomplish this goal

The differences between the simulated results and the experi- mental ones were ultimately insignificant, as shown in Fig.4 In the case with highest deviations (Case IIb), the simulated yield, productivity and resin utilization were 99.4%, 0.61 mg/min/mL and 31.5%, respectively, while the experimental data were 99.7%, 0.62 mg/min/mL and 31.4% This shows that a model calibrated using constant flow rate can be successfully used to optimize a trajec- tory with variable flow rate

3.3 Pilot-scale flow rate trajectories

The selected cases were scaled up to pilot scale as described

in Section3.4 As shown in Table1, both the column volume and the flow rate were increased 10 0 0 times compared to laboratory scale The column length was approximately 6 times higher than

at laboratory scale for Case I, while it was 4.5-5 times higher for the other cases The reason for this difference is that the maximum flow rate was lower in Case I than in the other cases, which means that the residence time was higher, and consequently, by Eq.12, the resulting column length was also higher The column diameter was around 10 cm for all cases, leading to a diameter-to-length ra- tio between 0.6 and 0.9, which is much lower than the value of 50 that was considered as a worst-case scenario for the prediction of the pressure drop over the column This means that the predicted pressure drop is overestimated, thus giving an extra safety margin Regarding the superficial velocities, they were higher than at lab- oratory scale, as expected, since the column length was increased and the residence time was maintained

Table 1

Column design results for five selected process cases at pilot scale

Process cases a)

Column volume Column length Column diameter Max flow rate Max velocity

a) Case I: Constant-flow loading; Case II: Stepwise flow rate trajectories; Case III: Linear flow rate trajec- tories; Cases IIa and IIIa are processes with similar productivity as the one of Case I; Cases IIb and IIIb are processes with similar resin utilization as the one of Case I

The pilot-scale cases were simulated with the column dimen-

sions and flow rates obtained, and the productivity, resin utiliza-

tion and yield were calculated The productivity and resin uti-

lization values for the simulated pilot-scale process are shown in

Fig.4for all process cases In agreement with Hansen’s statement

[36], the column efficiency is higher if the column length is in-

creased and the residence time remains constant This leads to a

sharper breakthrough curve, which in turn results in a higher yield

(data not shown) A lower amount of product loss in the break-

through leads to a slightly higher productivity and resin utiliza-

tion, as shown in Fig 4 Another aspect revealed is that having a

variable loading flow rate does not make a significant difference

in terms of process scale-up regardless the type of flow trajectory

applied, as the differences in productivity and resin utilization be-

tween the pilot-scale and the laboratory-scale processes are similar

for all the cases studied, as can be seen in Fig.4

Another aspect about the scale-up is the wall effects To avoid

wall effects the recommended minimum number of resin particles

per column section is 200 [38], and with the 1 mL HiTrap TM col-

umn used, this number is at highest 117 (obtained by dividing the

column diameter, 0.7 cm, by the particle diameter, 60 μm) This in-

dicates that wall effects were present at laboratory scale, but they

should not be present at a larger scale with a broader column,

which means that the separation would be at least the same and

probably better at a larger scale However, further experimental re-

search at pilot scale is required to validate this statement, as well

as to find out the aforementioned effect of the loss of wall support

at a larger scale

3.4 Comparison with multi-column continuous capture

A variable-flow process can be an alternative to multi-column

continuous processes for the increase of the efficiency in the cap-

ture of mAbs In a comparison between the variable-flow processes

presented here and a PCC process presented in our previous study

[6], where the protein A resin and the protein concentration were

the same as in this work, it can be said that productivity values

are similar In order to compare both processes, it is fairer to use

the total capture time in the definition of productivity instead of

only the loading time, because in the results from the PCC pro-

cess, the total capture time is considered The total capture time

is the loading time plus 60 min, which corresponds to the wash,

elution, regeneration and equilibration phases Basing the compar-

ison on the total capture time, the productivity values obtained in

the Pareto fronts vary from 0.23 to 0.35 mg/min/mL resin, which

are not far from the productivity values obtained in a PCC pro-

cess (0.10-0.38 mg/min/mL) Yet, the resin utilization in the PCC

process (ranging from 60 to 99%) is significantly higher than the

values obtained with the flow-programming strategies (from 13 to

50%) This is because in PCC two columns are interconnected dur-

ing the loading phase, which makes it possible to utilize a higher

amount of available binding sites without compromising yield, as

explained in our previous study [6] However, in order to imple- ment a PCC process, a more complex setup is needed, with a min- imum of two pumps and numerous valves to determine the path- ways This could be limiting in cases where there is shortage of resources like chromatography systems and pumps In addition, the benefit of a higher resin utilization in a multi-column process could not pay off the cost of adapting an already-existing batch process to the multi-column setup in some cases Therefore, the potential improvements in productivity and resin utilization with

a flow-programming strategy compared to a conventional capture process, combined with the lower complexity and cost of adapting the process setup, may warrant consideration as an alternative to multi-column continuous chromatography processes

4 Conclusions

Optimal flow trajectories for the loading phase of the capture

of monoclonal antibodies were obtained for the novel protein A resin mAb Select PrismA The two flow-programming approaches presented in this paper are better in terms of productivity and resin utilization than the constant-flow approach, as shown in the optimal Pareto fronts obtained The productivity can be increased

by up to 12%, and up to a 9% increase in resin utilization can

be achieved, while keeping yield above 99% In this work, several types of flow trajectories were studied and compared with each other with a model-based multi-objective optimization method, leading to the conclusion that simple but optimized trajectories are sufficient to achieve a more efficient process compared to a constant-flow approach Experimental validation was carried out for selected trajectories, both with purified mAb and with clari- fied supernatant, and results indicate that the predicted increase

in the two performance indicators can also be achieved experimen- tally, which shows that a model calibrated with constant-flow ex- periments can successfully be used in variable-flow applications In addition, the optimal processes selected were scaled up and simu- lated to show that the productivity, resin utilization and yield are slightly increased at a larger scale, thus showing that the optimal flow trajectories obtained are generalizable and applicable across scales for this specific protein A resin

The productivity obtained in the variable-flow processes imple- mented in this work are in the same range as the one obtained

in a multi-column continuous PCC process [6] Although the resin utilization is significantly lower than in the PCC process, flow- programming approaches can be an alternative to complex multi- column continuous capture processes due to their simplicity and ease of implementation The combination of the practical simplic- ity of the flow-programming approaches, which requires only a single column operated in batch mode with a variable volumetric flow rate, and the potential improvements in process performance indicators, makes this an effective approach towards reducing the cost of the purification of monoclonal antibodies In turn, such im- provements can potentially help reducing treatments costs, and by

7

extension, contribute to a greater availability of life-saving phar-

maceuticals

Declaration of Competing Interest

The authors declare that they have no conflict of interest

Acknowledgements

The authors acknowledge that this research is a collaboration

between the Competence Centre for Advanced BioProduction by

Continuous Processing (AdBIOPRO) [grant number 2016-05181] and

the AutoPilot project [grant number 2019-05314], both funded by

VINNOVA, the Swedish Agency for Innovation

Supplementary materials

Supplementary material associated with this article can be

found, in the online version, at doi: 10.1016/j.chroma.2020.461760

References

[1] H.L Levine , B.R Cooney , The Development of Therapeutic Monoclonal An-

tibody Products, BioProcess Technology Consultants, Inc., Woburn, MA , USA ,

2017

[2] M Dileo, A Ley, A.E Nixon, J Chen, Choices of capture chromatography

technology in antibody manufacturing processes, J Chromatogr B 1068-1069

(2017) 136–148 https://doi.org/10.1016/j.jchromb.2017.09.050

[3] 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, Biotech Prog 33 (2017) 854–866 https://doi.org/10

1002/btpr.2492

[4] A.L Zydney, Perspectives on integrated continuous bioprocessing—

opportunities and challenges, Curr Opin Chem Eng 10 (2015) 8–13

https://doi.org/10.1016/j.coche.2015.07.005

[5] A A Shukla, B Hubbard, T Tressel, S Guhan, D Low, Downstream processing

of monoclonal antibodies—Application of platform approaches, J Chromatogr

B 848 (2007) 28–39 https://doi.org/10.1016/j.jchromb.2006.09.026

[6] J Gomis-Fons, N Andersson, B Nilsson, Optimization study on periodic

counter-current chromatography integrated in a monoclonal antibody down-

stream process, J Chromatogr A 1621 (2020) 461055 https://doi.org/10.1016/j

chroma.2020.461055

[7] L Arnold, K Lee, J Rucker-Pezzini, J.H Lee, Implementation of Fully Integrated

Continuous Antibody Processing: Effects on Productivity and COGm, Biotech-

nol J 14 (2019) 180 0 061 https://doi.org/10.10 02/biot.20180 0 061

[8] A.S Rathore, Harshit Agarwal, Abhisek Kumar Sharma, Mili Pathak,

S Muthukumar, Continuous Processing for Production of Biopharmaceu-

ticals, Prep Biochem Biotech 45 (2015) 836–849 https://doi.org/10.1080/

10826068.2014.985834

[9] J Gomis-Fons, H Schwarz, L Zhang, N Andersson, B Nilsson, A Castan,

A Solbrand, J Stevenson, V Chotteau, Model-based design and control of a

small-scale integrated continuous end-to-end mAb platform, Biotechnol Progr

(2020) e2995 https://doi.org/10.1002/btpr.2995

[10] R Godawat, K Konstantinov, M Rohani, V Warikoo, End-to-end integrated

fully continuous production of recombinant monoclonal antibodies, J Biotech-

nol 213 (2015) 13–19 https://doi.org/10.1016/j.jbiotec.2015.06.393

[11] C.K.S Ng, F Rousset, E Valery, D.G Bracewell, E Sorensen, Design of high pro-

ductivity sequential multi-column chromatography for antibody capture, Food

Bioprod Process 92 (2014) 233–241 https://doi.org/10.1016/j.fbp.2013.10.003

[12] 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 https://doi.org/10.1016/j

chroma.2015.02.046

[13] L Aumann, M Morbidelli, A continuous multicolumn countercurrent solvent

gradient purification (MCSGP) process, Biotechnol Bioeng 98 (2007) 1043–

1055 https://doi.org/10.1002/bit.21527

[14] N Andersson, H.-K Knutson, M Max-Hansen, N Borg, B Nilsson, Model-Based

Comparison of Batch and Continuous Preparative Chromatography in the Sep-

aration of Rare Earth Elements, Ind Eng Chem Res 53 (2014) 16485–16493

https://doi.org/10.1021/ie5023223

[15] 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 https://doi.org/10.10 02/biot.20120 0 068

[16] K.B Konstantinov, C.L Cooney, White Paper on Continuous Bioprocessing May 20–21 2014 Continuous Manufacturing Symposium, J Pharm Sci 104 (2015) 813–820 https://doi.org/10.1002/jps.24268

[17] R Hahn, R Schlegel, A Jungbauer, Comparison of protein A affinity sorbents,

J Chromatogr B 790 (2003) 35–51 https://doi.org/10.1016/S1570-0232(03)

0 0 092-8 [18] GE Healthcare Bio-Sciences AB, Optimizing productivity on high capacity protein A affinity mediumGE Healthcare Bio-Sciences AB, Application note

29190587 AA, Uppsala, Sweden, 2016 [19] K Lacki, Introduction to Preparative Protein Chromatography, in: G Jagschies,

E Lindskog, K Lacki, P.M Galliher (Eds.), Biopharmaceutical Processing, El- sevier, Amsterdam, Netherlands, 2017, pp 319–366 https://doi.org/10.1016/ B978- 0- 08- 100623- 8.00016- 5

[20] 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 Progr 20 (2004) 830–840 https://doi.org/ 10.1021/bp0342654

[21] A Sellberg, M Nolin, A Löfgren, N Andersson, B Nilsson, Multi-flowrate Op- timization of the Loading Phase of a Preparative Chromatographic Separation, in: A Friedl, J.J Klemeš, S Radl, P.S Varbanov, T Wallek (Eds.), Computer Aided Chemical Engineering, Elsevier, Amsterdam, Netherlands, 2018, pp 1619–1624 https://doi.org/10.1016/B978- 0- 4 4 4- 64235- 6.50282- 5

[22] A Sellberg, A Holmqvist, F Magnusson, C Andersson, B Nilsson, Discretized multi-level elution trajectory: A proof-of-concept demonstration, J Chro- matogr A 1481 (2017) 73–81 https://doi.org/10.1016/j.chroma.2016.12.038 [23] E.X Perez-Almodovar, G Carta, IgG adsorption on a new protein A adsor- bent based on macroporous hydrophilic polymers I Adsorption equilibrium and kinetics, J Chromatogr A 1216 (2009) 8339–8347 https://doi.org/10.1016/j chroma.2009.09.017

[24] B Nilsson, N Andersson, Simulation of Process Chromatography, in: A Staby, A.S Rathore, S Ahuja (Eds.), Preparative Chromatography for Separation of Proteins, John Wiley & Sons, Inc., Hoboken, NJ, USA, 2017, pp 81–110 https: //doi.org/10.1002/9781119031116.ch3

[25] S.O Rastegar, T Gu, Empirical correlations for axial dispersion coefficient and Peclet number in fixed-bed columns, J Chromatogr A 1490 (2017) 133–137 https://doi.org/10.1016/j.chroma.2017.02.026

[26] T.M Pabst, J Thai, A.K Hunter, Evaluation of recent Protein A stationary phase innovations for capture of biotherapeutics, J Chromatogr A 1554 (2018) 45–60 https://doi.org/10.1016/j.chroma.2018.03.060

[27] M.C Annesis, L Merrelli, V Piemonte, L Turchetti, Mass Transfer Coefficients, in: M.C Annesis, L Merrelli, V Piemonte, L Turchetti (Eds.), Artificial Organ Engineering, Springer-Verlag, London, UK, 2017, pp 26–27 https://doi.org/10 1007/978- 1- 4471- 6443- 2

[28] H Schmidt-Traub, M Schulte, A Seidel-Morgenstern, Preparative chromatogra- phy, Wiley Online Library, 2012 https://doi.org/10.1002/9783527649280 [29] K Deb, Multi-objective optimization using evolutionary algorithms, John Wi- ley & Sons, Inc., Hoboken, NJ, USA, 2001 https://dl.acm.org/doi/book/10.5555/

559152 [30] GE Healthcare Bio-Sciences AB, Evaluation of Protein A resin lifetime during extensive use (overloading) in continuous chromatography modeGE Healthcare Bio-Sciences AB, Application note 29260553 AA, Uppsala, Sweden, 2017 [31] J Gomis-Fons, A Löfgren, N Andersson, B Nilsson, L Berghard, S Wood, Integration of a complete downstream process for the automated lab-scale production of a recombinant protein, J Biotechnol 301 (2019) 45–51 https: //doi.org/10.1016/j.jbiotec.2019.05.013

[32] N Andersson, A Lofgren, M Olofsson, A Sellberg, B Nilsson, P Tiainen, Design and control of integrated chromatography column sequences, Biotechnol Prog

33 (2017) 923–930 https://doi.org/10.1002/btpr.2434 [33] A Löfgren, N Andersson, A Sellberg, B Nilsson, M Löfgren, S Wood, De- signing an Autonomous Integrated Downstream Sequence From a Batch Sep- aration Process − An Industrial Case Study, Biotechnol J 13 (2018) 1700691 https://doi.org/10.10 02/biot.20170 0691

[34] H Maity, A Wei, E Chen, J.N Haidar, A Srivastava, J Goldstein, Comparison of predicted extinction coefficients of monoclonal antibodies with experimental values as measured by the Edelhoch method, Int J Biol Macromol 77 (2015) 260–265 https://doi.org/10.1016/j.ijbiomac.2015.03.027

[35] C Heuer, P Hugo, G Mann, A Seidel-Morgenstern, Scale up in preparative chromatography, J Chromatogr A 752 (1996) 19–29 https://doi.org/10.1016/ S0 021-9673(96)0 0496-7

[36] E.B Hansen, Chromatographic Scale-Up on a Volume Basis, in: A Staby, A.S Rathore, S Ahuja (Eds.), Preparative Chromatography for Separation of Pro- teins, John Wiley & Sons, Inc., Hoboken, NJ, USA, 2017, pp 227–246 https: //doi.org/10.1002/9781119031116.ch7

[37] J.J Stickel , A Fotopoulos , Pressure-Flow Relationships for Packed Beds of Com- pressible Chromatography Media at Laboratory and Production Scale, Biotech- nol Prog 17 (2001) 744–751

[38] S Kidal , O.E Jensen , Using volumetric flow to scaleup chromatographic pro- cesses, Biopharm Int 19 (2006) 34–44