B

KC

a

ARRAA

KIVMMLP

1

wta[tpbcp

utHoptfabt

A

0d

Vaccine 29 (2011) 3320–3328

Contents lists available at ScienceDirect

Vaccine

journa l homepage: www.e lsev ier .com/ locate /vacc ine

ioprocess optimization for cell culture based influenza vaccine production

unal Aggarwal1, Frank Jing, Luis Maranga1, Jonathan Liu ∗

ell Culture Development, Vaccine Development, MedImmune, 3055 Patrick Henry Dr., Santa Clara, CA 95054, USA

r t i c l e i n f o

rticle history:eceived 16 October 2010eceived in revised form 24 January 2011ccepted 25 January 2011vailable online 16 February 2011

a b s t r a c t

Uncertainties and shortcomings associated with the current influenza vaccine production processesdemand attention and exploration of new vaccine manufacture technologies. Based on a newly devel-oped mammalian cell culture-based production process we investigated selected process parameters anddescribe three factors that are shown to impact productivity, process robustness and development time.They are time of infection, harvest time and virus input, or multiplicity of infection (MOI). By defining thetime of infection as 4–5 days post cell seeding and harvest time as 2–3 days post-infection and comparing

eywords:nfluenza virusaccine productionultiplicity of infectionDCK cells

AIV

their effect on virus production, MOI is subsequently identified as the most impactful process parameterfor live attenuated influenza vaccine (LAIV) manufacture. Infection at very low MOI (between 10−4 and10−6 FFU/cell) resulted in high titer virus production (up to 30-fold productivity improvement) comparedto higher MOI infections (10−3 to 10−2 FFU/cell). Application of these findings has allowed us to developa platform process that can reduce the development time to approximately three weeks for an influenza

cess f

rocess optimization vaccine manufacture pro

. Introduction

Influenza or flu is a major cause of acute respiratory illnessorldwide, particularly among children and the elderly popula-

ion. Each year infection by flu viruses results in 36,000 deathsnd more than 200,000 hospitalizations in the United States alone1]. Although the traditional influenza vaccines have been usedo provide protection against seasonal flu epidemics, alternativeroduction strategies and different types of vaccines are activelyeing pursued to overcome the limitations inherent to these vac-ines, particularly those that may become more evident during aandemic flu outbreak [2–5].

The seasonal flu vaccines are used to vaccinate the general pop-lation and consist of three different virus strains (type A/H1N1,ype A/H3N2 and type B). These strains are selected by the Worldealth Organization and US Food and Drug Administration basedn surveillance data of flu virus strains circulating in the humanopulation prior to the flu season and recommendations are madeo the flu vaccine manufacturers early in the year, typically February

or the Northern Hemisphere. The virus strains are used to gener-te a vaccine that is typically targeted for pre-season distributioneginning in July to September and initial immunization shortlyhereafter. Thus, the vaccine manufacturers are challenged with a

∗ Corresponding author. Tel.: +1 650 603 2576; fax: +1 650 603 3576.E-mail address: liuj@medimmune.com (J. Liu).

1 Current address: Novartis Vaccines & Diagnostics, Inc., 350 Massachusettsvenue, M/S 4103T/45SS, Cambridge, MA 02139, USA.

264-410X/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.oi:10.1016/j.vaccine.2011.01.081

or new strains.© 2011 Elsevier Ltd. All rights reserved.

tight production schedule and have very limited time to develop oroptimize the production conditions to improve process yields [6].

Flu vaccines have traditionally been manufactured using embry-onated hens’ eggs. In recent years cell culture based productiontechnology has been explored and demonstrated to be a viablealternative to egg-based technology [7–9]. In contrast to theconventional egg-based production process, cell-based produc-tion technology promises shorter production cycles, greater surgecapacity, greater process control and more reliable and well char-acterized production substrates [10,11]. Furthermore, cell culturebased manufacturing processes are capable of providing largerquantities of vaccine within a shorter period of time which is espe-cially important during a pandemic outbreak when the egg supplyfrom specific pathogen free flocks is limited. Following the success-ful launch of a cold-adapted (ca) Live Attenuated Influenza Vaccine(LAIV) approved for use in 2–49 year old individuals [12] andproduced using conventional egg-based production technology,MedImmune initiated a cell culture based flu vaccine developmentprogram. This new vaccine manufacture process makes use of theadherent Madin–Darby Canine Kidney (MDCK) cells as a produc-tion cell substrate, and grows and infects the cells on microcarrierbeads under a controlled environment in bioreactors [9].

The first generation LAIV cell culture flu vaccine manufacturingprocess produced vaccine bulk at high titers (>8 log10 FFU/ml) for 3

selected virus strains [9]. However, when the process was evaluatedusing additional flu strains, large variations in productivity wereobserved among different viruses, ranging from 7 to 9 log10 FFU/ml.This variability creates a high degree of uncertainty with respectto expected process yields, particularly for novel virus strains that

ccine 2

ltsiitpoimitdoweaoctA3masv

2

2

CCstaob

2

o(TssIu

2

dsdflm5TflbLi

K. Aggarwal et al. / Va

ack sufficient manufacturing history. The result is significant risko the timely production and supply of flu vaccine. Since flu vaccinetrains change frequently and process optimization time is limited,t would be advantageous to develop a robust platform process thats relatively insensitive to virus strain changes and provides consis-ency in productivity for most seasonal and even future unknownandemic flu viruses. To achieve this goal we pursued a strategyf identifying critical process parameters that meet the follow-ng three criteria: (a) broad impact, affecting virus productivity of

ultiple virus strains; (b) minimum change requirement, i.e. mod-fication of these parameters does not result in major alterations ofhe platform process such as change of production substrate, pro-uction medium, use of additional process equipment; (c) shortptimization time (less than 2–3 weeks). In this communicatione describe our efforts at identifying three critical process param-

ters that meet the above criteria (time of infection, harvest timend multiplicity of infection) and demonstrate their utility in devel-ping the platform vaccine manufacture process. Using 25 differenta influenza vaccine strains that belong to multiple types or sub-ypes of flu viruses (Type B, A/H1N1, A/H2N2, A/H3N2, A/H5N1,/H7N3, and A/H9N2 influenza viruses) we show that up to a0-fold improvement in virus productivity can be achieved by opti-izing only a few selected process parameters. In addition, we can

lso reduce the variability in productivity among different virustrains and thus develop a more robust manufacturing process foraccine production.

. Materials and methods

.1. Cell line

Anchorage dependent Madin Darby Canine Kidney (MDCK, ATCCCL-34) cells were originally obtained from American Type Cultureollection (Manassas, VA), cloned by limit dilution and adapted toerum-free growth [11]. One of the MDCK cell clones, 9B9-1E4, wasested for susceptibility for ca reassortant influenza virus infectionnd banked for this study [10]. The cells were cultured in T-flasksr roller bottles before being seeded in bioreactors as describedelow.

.2. Virus strains

Wild type (wt) influenza viruses used in this study werebtained from the Centers for Disease Control and PreventionAtlanta, GA). The corresponding ca reassortant viruses (listed inable 1) were made by reassorting wt virus with LAIV master donortrains, A/Ann Arbor/6/60 or B/Ann Arbor/01/64 following the clas-ical reassortment procedure that was established previously [13].n the work presented here, only the ca reassortant viruses weresed to infect MDCK cells.

.3. Static culture conditions (cell thaw and expansion)

Frozen stock of MDCK cells was recovered by thawing the cellsirectly in a T75 flask (Corning, Lowell, MA) containing pre-warmederum free proprietary growth medium and culturing them for 3–4ays. For further cell expansion, cells were subcultured in T-225asks and roller bottles (Corning, Lowell, MA) in the same growthedium every 3–4 days. Cultivation was carried out at 37 ◦C in a

% CO2 incubator. For passaging, MDCK cells were trypsinized using

rypLETM Select (Gibco, Carlsbad, CA) for 15 min in T-75 and T-225asks or for 20 min in roller bottles. The trypsinization was stoppedy addition of lima bean trypsin inhibitor solution (Worthington,akewood, NJ) as recommended by the supplier. Initial cell seed-ng density of 5 × 104 cells/mL and 6.7 × 104 cells/mL was used for

9 (2011) 3320–3328 3321

T-225 (100 mL) flasks and roller bottles (300 mL/850 cm2), respec-tively. After culturing for 3–4 days, cells were harvested from rollerbottles and used to inoculate 3 L bioreactors.

2.4. Bioreactor culture conditions

MDCK cells were grown on Cytodex 3 microcarriers (GE Health-care, Piscataway, NJ) in a proprietary serum-free growth medium,in fully controlled stirred 3 L glass bioreactors (Applikon Biotech-nology, Foster City, CA). A cell seeding density of 9 × 104 cells/mLand microcarrier concentration of 2 g/L was used. pH was controlledat 7.4 by sparging CO2 and addition of 1 M NaOH as needed. Tem-perature was maintained at 37 ◦C until virus infection. Dissolvedoxygen was allowed to decline from 100% during the early cultureand was maintained at 50% of air saturation by sparging pure oxy-gen during later culture times. The agitation rate was kept constantat 90 rpm.

2.5. Virus infection

To determine the optimal time of infection, MDCK cells wereseeded in a 3 L bioreactor and then infected daily from 1 day postcell seeding (dps) up to 5 dps. At the time of infection, MDCKcell culture was removed from the 3 L bioreactor mother cultureand divided in to 30 mL aliquots to set up daughter cultures in125 mL shake flasks. Prior to infection, medium exchange with freshmedium was performed followed by addition of TrypLETM. Selectto a final concentration of 3% (volume/volume). The daughter cul-tures were then infected by a ca virus strain at specific virus inputs,or multiplicity of infection (MOI). After infection, all shake flaskswere incubated at 33 ◦C and 100 rpm in a 5% CO2 incubator for upto 4 days. The cell culture fluid was collected daily post infectionfrom one or 2 different infections depending on the virus strain andthe virus titer was determined by Fluorescent Focus Assay (FFA), aspreviously described [14].

2.6. Analytical methods

Daily samples were taken from the 3 L bioreactor cultures forviable cell density and metabolite analysis until daughter cultureswere set up in shake flasks for infection. After infection, the shakeflask cultures were sampled for 4 more days for infectious virustiter measurement. Cell number and viability were determinedusing a NucleoCounter (New Brunswick Scientific Co. Inc, NewBrunswick, NJ). Offline pH, nutrient and metabolite concentrationswere obtained using a Bioprofile 400 (Nova Biomedical, Waltham,MA). Infectious virus titer was measured by infecting MDCK cellmonolayers with culture supernatants using FFA [14].

3. Results

3.1. Effect of time of infection on virus replication kinetics

To investigate the effect of time of infection on virus productiv-ity, MDCK cell cultures were seeded on microcarriers and grownfor different lengths of time, varying between 1 and 5 days, beforeinfection with ca A/Wisconsin/67/05 under four different condi-tions: three different MOIs, i.e., 0.01, 0.001 and 0.0001 FFU/celland a constant virus input of 2000 FFU/ml based on the cell cul-ture volume. The constant volume-based virus input was chosen

to determine if a more operation-friendly process was feasible andeliminate the need for calculating the amount of inoculum virusbased on viable cell density (VCD) before each infection, Fig. 1shows a typical virus production time course in cell cultures thatwere infected at a low MOI (0.001 FFU/cell). The upward trend of

3322 K. Aggarwal et al. / Vaccine 29 (2011) 3320–3328

Table 1Cold adapted, Live, Attenuated Influenza Vaccine (LAIV) strains used for studying the effect of MOI on virus titer.

No. LAIV strain Virus type MOIs tested (FFU/cell)

1 B/Florida/7/04 B 10−2, 10−3, 10−4, 10−5, 10−6

2 B/Yamanashi/166/98 B 10−2, 10−3, 10−4, 10−5, 10−6

3 B/Victoria/504/2000 B 10−2, 10−3, 10−4, 10−5, 10−6

4 B/Ann Arbor/1/66 B 10−2, 10−3, 10−4, 10−5, 10−6

5 B/Florida/4/2006 B 10−2, 10−3, 10−4, 10−5, 10−6

6 B/Malaysia/2506/04 B 10−2, 10−3, 10−4, 10−5, 10−6

7 A/Texas/36/91 A/H1N1 10−2, 10−3, 10−4, 10−5, 10−6

8 A/Solomon Islands/3/06 A/H1N1 10−2, 10−3, 10−4, 10−5, 10−6

9 A/Hong Kong/2652/06 A/H1N1 10−2, 10−3, 10−4, 10−5, 10−6

10 A/South Dakota/6/07 A/H1N1 10−2, 10−3, 10−4, 10−5, 10−6

11 A/Beijing/262/95 A/H1N1 10−2, 10−3, 10−4, 10−5, 10−6

12 A/New Caledonia/20/99 A/H1N1 10−2, 10−3, 10−4, 10−5, 10−6

13 A/California/07/2004 A/H3N2 10−3, 10−4, 10−5, 10−6

14 A/Brisbane/10/2007 A/H3N2 10−2, 10−3, 10−4, 10−5, 10−6

15 A/Nepal/921/06 A/H3N2 10−2, 10−3, 10−4, 10−5, 10−6

16 A/Uruguay/716/2007 A/H3N2 10−2, 10−3, 10−4, 10−5, 10−6

17 A/Wyoming/03/2003 A/H3N2 10−2, 10−3, 10−4, 10−5, 10−6

18 A/Wuhan/395/95 A/H3N2 10−2, 10−3, 10−4, 10−5, 10−6

19 A/Panama/2007/99 A/H3N2 10−2, 10−3, 10−4, 10−5, 10−6

20 A/Wisconsin/67/05 A/H3N2 10−3, 10−4, 10−5, 10−6

21 A/Ann Arbor/6/60 A/H2N2 10−2, 10−3, 10−4, 10−5, 10−6

A/ −2 −3 −4 −5 −6

A/A/A/

vsAtttmvlst

Fgoctp78Fttt

22 A/Hong Kong/213/200323 A/Vietnam/1203/200424 A/British Columbia/06/0425 A/Hong Kong/G9/97

irus titer exhibited at early times confirmed that the infection wasuccessful and progeny virus production occurred in all infections.lthough samples were collected and tested immediately following

he infection (0 hpi) the virus titer was too low to be detected andhus not reported. Based on the VCD of the culture at the chosenime of infection and MOI, the virus titer in these samples was esti-

ated to be between 1.3 and 4.0 log10 FFU/ml. Starting from 24 hpi

irus titer increased quickly and reached a peak level by 48 hpi. Fol-owing peak production, virus titer decreased in the cell cultures,uggesting existence of a potential time window for virus harvesthat maximizes the process yield regardless of the time of infection.

Hours post infection (hrs)

120967248240

Vir

us

Tit

er (

log

10F

FU

/ml)

6

7

8

9

ig. 1. Virus replication time course. MDCK cells were seeded on microcarriers andrown for various times. They were infected with ca A/Wisconsin/67/05 at MOIf 0.001 FFU/cell and sampled at the following specified time points: 1 day postell seeding (dps) (©), 2 dps (�), 3 dps (�), 4 dps (�) and 5 dps (�). Virus produc-ion reached the peak level by 48 h post infection (hpi) in all the infections. Theeak titer varied depending on the time of infection with 7.9 log10 FFU/ml for 1 dps,.8 log10 FFU/ml for 2 dps, 7.9 log10 FFU/ml for 3 dps, 8.3 log10 FFU/ml for 4 dps and.6 log10 FFU/ml for 5 dps cultures. The virus titer in each sample was determined byluorescent Focus Assay (FFA) and is presented as Fluorescent Focus Units (FFU) inhe figure. Virus titer at 0 h post infection was too low to be detected. In some cases,he titers at 24 hpi were also below the FFA detection limit and are not presented inhe figure.

H5N1 10 , 10 , 10 , 10 , 10H5N1 10−2, 10−3, 10−4, 10−5, 10−6

H7N3 10−2, 10−3, 10−4, 10−5, 10−6

H9N2 10−2, 10−3, 10−4, 10−5, 10−6

This observed trend and virus replication kinetics was not uniqueto the low MOI infection with ca A/Wisconsin/67/05 virus. This wasalso observed in infections with other virus strains and at differentMOIs (data not shown). These observations indicate that the dura-tion of cell growth on microcarriers, as measured by the time postcell seeding, did not alter the virus replication kinetics. The timeto reach peak virus production for ca A/Wisconsin/67/05 virus wasconsistently around 48 h regardless of the age of the culture or timeof infection.

3.2. Effect of time of infection on the level of virus production

While virus replication kinetics remained stable, the level ofvirus production including the peak virus titer varied accordingto the time of infection. For example, the later the culture wasinfected the higher the level of virus production observed. Thisbecame evident when the peak virus titer determined from thecultures infected at different times of infection was compared.The peak virus titer was generally lower in the cell cultures thatwere infected shortly after seeding on microcarriers (1, 2 or 3 dps)than those that had grown for a longer period of time (4 or 5 dps)(Fig. 2). This was speculated to be related to VCD as it was higherat 4–5 dps compared to 1–3 dps (refer to the VCD curve in Fig. 2).It was interesting to note that peak virus titers in the 5 infectionswith a constant volume-based virus input (2000 FFU/ml) were sim-ilar to those infected at high MOI (0.01 FFU/cell) early after cellseeding (1–3 dps) and then changed to become more like thoseat lower MOI (0.001 FFU/cell and 0.0001 FFU/cell) later after cellseeding (4–5 dps). Calculations based on the cell density and theinput virus titer at specific times revealed that the actual MOI inthe infections with the volume-based virus input was between0.011 and 0.003 FFU/cell for the infection performed on 1–3 dps and0.002–0.001 FFU/cell for the infections performed on 4 and 5 dps,

confirming the actual MOI was high early and low late post cellseeding with constant volume-based virus input. The highest peakvirus titer for ca A/Wisconsin/67/05 was 8.8 log10 FFU/ml and wasobserved at 48 hpi in the MDCK cultures infected on 4 dps at a MOIof 0.0001 FFU/cell.

K. Aggarwal et al. / Vaccine 29 (2011) 3320–3328 3323

48 hpi

Days post seeding (dps)6543210

Vir

us

Tit

er (

log

10F

FU

/ml)

6.6

6.9

7.2

7.5

7.8

8.1

8.4

8.7

9.0

Via

ble

cel

l den

sity

(1x

106

cells

/ml)

0.1

1

10

100

Fig. 2. Effects of the time of infection on peak virus titers. MDCK cells were seededon microcarriers at 9 × 104 cells/mL and grown for 1–5 days followed by infectionwith ca A/Wisconsin/67/05 at 3 different MOIs: 0.0001 FFU/cell (�), 0.001 FFU/cell(©), 0.01 FFU/cell (�) and one constant volumetric virus input of 2000 FFU/ml (�).Infected cell culture fluid was collected from shake flasks daily up to 4 days postivct

3v

ditdvt4lcmihv

FMtaait

ca A/South Dakota/6/07

Hours post infection (hpi)

120967248240

FFA

tit

er (

log

10F

FU

/ml)

4

5

6

7

8

9

10

ca A/Brisbane/10/2007

Hours post infection (hpi)

120967248240

FFA

tit

er (

log

10F

FU

/ml)

3

4

5

6

7

8

9b

a

Fig. 4. Virus production time-course. MDCK cells were grown on microcarriers andinfected with (a) ca A/South Dakota/6/07 and (b) ca A/Brisbane/10/2007 at five differ-

−6 −5 −4 −3

nfection and the virus titer was determined by the Fluorescent Focus assay. The peakirus titer was reached at 48 h post infection and is presented along with the viableell density (×) determined from the mother bioreactor cultures at the specifiedimes of infection.

.3. Effect of multiplicity of infection on ca A/Wisconsin/67/05irus production

While analyzing the effect of time of infection on virus pro-uction we found that another process parameter, multiplicity of

nfection or MOI also impacted virus productivity. This is illus-rated in Fig. 3 which shows the extent of virus amplification atifferent times post infection and at different MOI. The extent ofirus amplification was defined as the ratio of amount of infec-ious viruses obtained at a specific time post infection (in this case,8 hpi) divided by the amount of input virus present in the inocu-

um. When the cell cultures were infected at different times postell seeding the difference in virus titer was relatively small, nor-

ally less than 10 fold between 1 and 5 dps. When the cells were

nfected at different MOI the virus titer differed more than severalundred fold. For example, approximately 847 fold amplification ofirus was obtained by simply reducing the MOI from 0.01 FFU/cell

48 hpi

Days post seeding (dps)

54321

Vir

us

Am

plif

icat

ion

(x1

03 )

1

10

100

1000

10000

ig. 3. Comparison of virus titer obtained from infections performed at differentOIs and different times of infection. MDCK cells were seeded, grown, infected and

ested as described in the legend in Fig. 2. Four different virus inputs were usedt the time of infection: 0.0001 FFU/cell (�), 0.001 FFU/cell (©), 0.01 FFU/cell (�)nd 2000 FFU/ml (�). Virus amplification was calculated by dividing the amount ofnfectious viruses determined at 48 hpi by the amount of input viruses used at theime of infection.

ent virus inputs: 10 FFU/cell (�), 10 FFU/cell (�), 10 FFU/cell (�), 10 FFU/cell

(©) and 10−2 FFU/cell (�). Virus titers were determined from individual infections atthe defined time points. Note the more pronounced increase in virus titers for lowerMOI infections compared to the higher MOI infections between 24 and 72 hpi.

to 0.0001 FFU/cell in the MDCK cells that had grown for 5 days. Con-sistent with this observation, when the cell cultures were infectedat a constant volumetric virus input of 2000 FFU/ml, the extent ofvirus amplification changed as the time of infection varied. Theextent of virus amplification increased at later times (4 or 5 dps)than early times (1–3 dps), presumably because of an increase ofviable cells and the corresponding decrease of MOI at later times.

3.4. Effect of very low multiplicity of infection on virus replicationkinetics and yield

To determine whether the observed MOI dependency of virusproductivity was specific to ca A/Wisconsin/67/05, another A/H3N2virus, ca A/Brisbane/10/2007 and a different subtype virus, caA/South Dakota/6/07 (type A/H1N1) were tested. MDCK cells wereseeded and grown on microcarriers for 4 days before infectionwith the virus at MOI of 10−6, 10−5, 10−4, 10−3 and 10−2 FFU/cell,respectively. Virus production was monitored daily up to 4 dpi.Fig. 4(a) and (b) shows the virus production time course under

these infection conditions. During the first 24 hpi the measuredvirus titers correlated with the amount of viruses in the inoculum,namely higher MOI infection resulted in higher titer virus produc-tion. Starting from 48 hpi, lower MOI infections exceeded higher

3324 K. Aggarwal et al. / Vaccine 29 (2011) 3320–3328

F ith cai otogrt at a la

Mvahi7t(a7ct

wArcIavdi(vtd(sau

citAi

ig. 5. Cytopathic effect of infected MDCK cells. MDCK cells were infected at 4 dps wnfected cells were examined under a light microscope daily and representative phhat excessive floating cells are present in higher MOI (10−2 FFU/cell) infections andre 500 �m.

OI infections in virus productivity as measured by infectiousirus titers. MDCK cells infected with both ca A/South Dakota/6/07nd ca A/Brisbane/10/2007 at MOIs of 10−6 and 10−5 FFU/cellad more measurable infectious viruses than their counterparts

nfected at higher MOIs, namely, 10−3 and 10−2 FFU/cell after2 hpi, suggesting that the viruses underwent greater amplifica-ion and reached higher peak titers in very low MOI infectionsdown to 10−6 and 10−5 FFU/cell). In this study the optimum MOInd virus harvest time were determined to be 10−5 FFU/cell and2 hpi for ca A/South Dakota/6/07 and 10−4 FFU/cell and 72 hpi fora A/Brisbane/10/2007 in the microcarrier-based MDCK cell cul-ure.

The effect of MOI on virus infection and production kineticsas also evident based on cell morphology changes (Fig. 5). In ca/South Dakota/6/07 virus infection at higher MOI (10−2 FFU/cell)esulted in detachment of large numbers of cells from the micro-arriers which floated in the cell culture fluid as early as 48 hpi.n contrast, at low MOI (10−6 FFU/cell) most of the cells remainedttached to the microcarriers at this time point. With progression ofirus infection more cells infected at MOI of 10−6 FFU/cell becameetached and cell debris accumulated in the culture by 72 hpi. It

s interesting to note that although the extent of cytopathic effectCPE) corresponded to MOI changes, it is not always indicative ofirus titer. For example, more severe CPE was observed for infec-ions at higher MOI (10−2 FFU/cell), but higher titer viruses wereetected from the culture infected at lower MOI (10−6 FFU/cell)Figs. 4 and 5), suggesting determination of virus harvest timehould not be based on the appearance of the infected cultureslone and needs to be adjusted individually for specific virus strainssed for vaccine production.

To test how broadly the aforementioned MOI effects were appli-

able to flu vaccine production, we collected a panel of 25 canfluenza vaccines strains and investigated the effect of MOI onheir peak titers. The strains represented type B, A/H1N1, A/H2N2,/H3N2, A/H5N1, A/H7N3, and A/H9N2 influenza viruses and

ncluded both seasonal and pandemic vaccine strains. MDCK cells

A/South Dakota/6/07 at MOIs of 10−6 FFU/cell and 10−2 FFU/cell, respectively. Theaphs taken at the indicated time points are presented (magnification 100×). Noteater time post infection (72 hpi) at lower MOI (10−6 FFU/cell). Horizontal scale bars

were infected at 4 dps at a wide range of MOIs varying between10−6 and 10−2 FFU/cell for each virus (Table 1). Infected cells weresampled daily and analyzed together. Close examination of the rela-tion between MOI and peak virus titer as shown in Fig. 6 revealed3 general patterns: (1) a curved shape that covered the range ofoptimum MOI between 10−4 and 10−6 for many virus strains; (2) adownward trend with higher virus titers obtained in lower MOIinfections; (3) a relatively flat horizontal line for a few viruses,mostly Type B viruses. The curve shape suggested that the optimalMOI to maximize virus productivity was reached within the rangeof MOI tested, while the downward trend confirmed the previousobservation that lower MOI infection led to high titer virus pro-duction in MDCK cells was true for many strains of different typesand/or subtypes of ca influenza viruses. It should be mentioned thatthe optimal MOI for maximum virus production yield may not havebeen detected in this study for a few ca virus strains. Due to practicalreasons such as limitations in virus stock titers, very high virus stockdilutions and small virus stock volumes needed for infection, we didnot test MOIs below 10−6 FFU/cell. We suspect that further produc-tivity improvement could be obtained by lowering MOI below thislevel for specific virus strains. In general, lower virus input used forinfection was found to yield higher peak titers for most vaccinesstrains listed in Table 1. On the other hand, the effect of MOI onvirus titers was less obvious for many type-B virus strains as com-pared to the type-A strains. Only a slight increase in virus titers wasobserved for type-B strains when low MOI was used for infection.This was seen as a flat line in Fig. 6. No specific MOI was determinedto be optimal for the entire panel of different strains of type-B andtype-A (H3N2, H5N1, H7N3, H2N2 and H9N2) virus strains exceptthe H1N1 subtype strains for which MOI of 10−5 FFU/cell providedthe highest peak titers (Fig. 6). Therefore, for practical reasons we

routinely used MOI of 10−4 for type-B virus infection.

The data compiled in Fig. 7 showed that among the 25 testedvirus strains 9 reached peak virus titer production in MDCK cellsinfected at MOI of 10−6 FFU/cell, 11 with peak titer in infectionsat MOI of 10−5 FFU/cell and 5 required MOI of 10−4 FFU/cell to

K. Aggarwal et al. / Vaccine 29 (2011) 3320–3328 3325

Other type-A ca influenza vaccine strains

Virus input (FFU/ml)

10-7 10-6 10-5 10-4 10-3 10-2 10-1

Vir

us

tite

r (l

og

10F

FU

/ml)

6.9

7.2

7.5

7.8

8.1

8.4

8.7

9.0

9.3

Type B ca influenza vaccine strains

Virus input (FFU/ml)

10-7 10-6 10-5 10-4 10-3 10-2 10-1

Vir

us

tite

r (l

og

10F

FU

/ml)

6.5

6.8

7.1

7.4

7.7

8.0

8.3

8.6

8.9

9.2

Type A/H1N1 ca influenza vaccine strains

Virus input (FFU/ml)

10-7 10-6 10-5 10-4 10-3 10-2 10-1

Vir

us

tite

r (l

og

10F

FU

/ml)

6.8

7.1

7.4

7.7

8.0

8.3

8.6

8.9

9.2

Type A/H3N2 ca influenza vaccine strains

Virus input (FFU/ml)

10-7 10-6 10-5 10-4 10-3 10-2 10-1

Vir

us

tite

r (l

og

10F

FU

/ml)

6.3

6.6

6.9

7.2

7.5

7.8

8.1

8.4

8.7

9.0

9.3

a

b d

c

Fig. 6. Effect of MOI on peak virus titers of LAIV strains. Four day old MDCK cells were infected with 25 ca influenza viruses at 5 different MOIs as indicated in the figure. Thepeak virus titer was determined from the virus production time course (1 through 4 dpi) and plotted against the MOI. The experiment results are presented in groups accordingto the influenza virus types and sub-types as the following: (a) type A/H1N1: A/Texas/36/1991 ( ), A/Solomon Islands/3/2006 ( ), A/Hong Kong/2652/2006 ( ), A/SouthDakota/6/2007 ( ), A/Beijing/262/1995 ( ) and A/New Caledonia/20/1999 ( ), (b) type A/H3N2: A/California/07/2004 ( ), A/Brisbane/10/2007 ( ), A/Nepal/921/2006

( ), A/Uruguay/716/2007 ( ), A/Wyoming/03/2003 ( ), A/Wuhan/395/95 ( ), A/Pavaccine strains: A/Ann Arbor/6/1960 (H2N2, ), A/Hong Kong/213/2003 (H5N1, ), AA/Hong Kong/G9/1997 (H9N2, ). (d) type B: B/Florida/7/2004 (�), B/Yamanashi/166/19B/Malaysia/2506/2004 (�).

Optimum MOI (FFU/cell)

10-6 10-5 10-4 10-3 10-2

Nu

mb

er o

f co

ld a

dap

ted

vac

cin

e st

rain

s

0

2

4

6

8

10

12

48 hpi72 hpi

Fig. 7. Distribution of optimal MOIs producing peak virus titers. The number ofvirus strains that reached peak virus titer at a specific MOI and specific time is pre-sented by the bar height. The bar shade indicates whether the peak virus titer wasreached at 48 hpi or 72 hpi. A total of 25 ca influenza virus strains were tested at 5MOIs that ranged from 10−2 to 10−6 FFU/mL. Note none of the higher MOIs (10−2 to10−3 FFU/mL) resulted in peak virus production.

nama/2007/1999 ( )and A/Wisconsin/67/2005( ), (c) other type A influenza/British Columbia/06/2004 (H7N3, ), A/Vietnam/1203/2004 (H5N1, ), and

98 (©), B/Victoria/504/2000 (�), B/Ann Arbor/1/1966 (�), B/Florida/4/2006 (�) and

achieve highest virus productivity. None of these viruses achievedthe peak productivity at MOIs equal to or higher than 10−3 FFU/cell.In addition, the virus production time that was required to reach thepeak titer, i.e. the optimal harvest time, varied with MOI in a strainspecific manner. As observed for ca A/Brisbane/10/2007 infection(Fig. 4(b)), it took a longer time to reach the peak virus titer in cul-tures infected at lower MOIs for most viruses. The peak titer wasusually obtained in 72 h when infection was performed at MOI of10−6 FFU/cell, rather than 48 h after infection at higher MOI.

3.5. Applications of MOI optimization on LAIV vaccine production

Previously we described a MDCK cell-based manufacturingprocess to produce flu vaccine for human clinical trial material pro-duction [9]. Following that procedure MDCK cells were separatelyinfected with 25 vaccine strains at MOI of 10−3 FFU/cell and moni-tored daily for virus production up to 4 days. Fig. 8 summarizes thepeak virus titers for these 25 tested vaccine strains. The averagepeak virus titer was 8.2 log10 FFU/ml and the peak titer variation

range was 7.1–8.8 log10 FFU/ml among the 25 ca vaccine strains.Through several process parameter changes, including adjustingthe MOI from 10−3 FFU/cell to lower MOI as described above, allvirus strains produced higher titer viruses with average peak titerreaching 8.7 ± 0.3 log10 FFU/ml and peak titer varying between 7.7

3326 K. Aggarwal et al. / Vaccine 29 (2011) 3320–3328

Cold adapted vaccine strains

B/Flo

rida/

7/04

B/Yam

anas

hi/166

/98

B/Vic

toria

/504

/200

0

B/Ann

Arbor/1

/66

B/Flo

rida/

4/20

06

B/Mal

aysi

a/25

06/0

4

A/Tex

as/3

6/91

A/Solo

mon Is

lands/

3/06

A/Hong

Kong/265

2/06

A/South

Dak

ota/6

/07

A/Bei

jing/2

62/9

5

A/New

Cal

edonia

/20/

99

A/Cal

iforn

ia/0

7/20

04

A/Bris

bane/

10/2

007

A/Nep

al/9

21/0

6

A/Uru

guay/7

16/2

007

A/Wyo

min

g/03/

2003

A/Wuhan

/395

/95

A/Pan

ama/

2007

/99

A/Wis

consi

n/67/

05

A/Ann A

rbor/6

/60

A/Hong K

ong/213

/200

3

A/Vie

tnam

/120

3/20

04

A/Brit

ish C

olum

bia/0

6/20

04

A/Hong K

ong/G9/

97

Vir

us

tite

r (L

og

10F

FU

/ml)

6.9

7.2

7.5

7.8

8.1

8.4

8.7

9.0

9.3

Cold adapted vaccine strains

B/Flo

rida/

7/04

B/Yam

anas

hi/166

/98

B/Vic

toria

/504

/200

0

B/Ann A

rbor/1

/66

B/Flo

rida/

4/20

06

B/Mal

aysi

a/25

06/0

4

A/Tex

as/3

6/91

A/Solo

mon Is

lands/

3/06

A/Hong K

ong/265

2/06

A/South

Dak

ota/6

/07

A/Bei

jing/2

62/9

5

A/New

Cal

edonia

/20/

99

A/Cal

iforn

ia/0

7/20

04

A/Bris

bane/

10/2

007

A/Nep

al/9

21/0

6

A/Uru

guay/7

16/2

007

A/Wyo

min

g/03/

2003

A/Wuhan

/395

/95

A/Pan

ama/

2007

/99

A/Wis

consi

n/67/

05

A/Ann A

rbor/6

/60

A/Hong K

ong/213

/200

3

A/Vie

tnam

/120

3/20

04

A/Brit

ish C

olum

bia/0

6/20

04

A/Hong K

ong/G9/

97

Vir

us

tite

r (L

og

10F

FU

/ml)

6.9

7.2

7.5

7.8

8.1

8.4

8.7

9.0

9.3a

b

Fig. 8. Peak infectious virus titers obtained from infection at (a) MOI of 10−3 FFU/cell and (b) optimized MOI, for 25 different LAIV strains. The results are presented byg ), typp iple ad nd 8.7i

aoiacaspd

4

ps

rouping the virus strains to influenza type B (©), type A/H1N1 (�), type A/H3N2 (�rocess yield and the associated standard deviations are calculated based on multotted lines, respectively. The average process titers were 8.2 ± 0.5 log10 FFU/ml a

nfections as described in the text, respectively.

nd 9.2 log10 FFU/ml. This represented productivity improvementf 3-fold on average, reduction of yield variation and at least 4-foldmprovement in the lowest titer attained. These accomplishmentsre particularly important for vaccine production since the flu vac-ines require production of an equal amount of each vaccine strainnd large variations in the productivities among the three virustrains constituting the trivalent vaccine force the manufacturer toroduce additional costly batches and may result in delay in vaccineelivery.

. Discussion

Limitations associated with the current egg-based flu vaccineroduction have caused worldwide concern about the flu vaccineupply chain and thus the ability to effectively immunize the gen-

e A/H2N2 (�), type A/H5N1 (�), type A/H7N3 (�) and type A/H9N2 (�). The averagessay results of the samples collected from single infection and shown in bold and± 0.3 log10 FFU/ml when MOI of 10−3 FFU/cell and optimized MOIs were used in

eral population. Partnering with the U.S. federal government, thevaccine industry is exploring alternative vaccine production tech-nologies. Among these new technologies, the cell culture-basedmanufacturing platform offers some significant advantages. Cellculture based technology overcomes many of the limitations inher-ent to egg-based production including lengthy production scale-up,and complex production substrate supply chain and provides, bet-ter characterization of in-process and final products, better controlof raw materials and production processes, and most importantlythe surge capacity to produce additional vaccine doses in the event

of a pandemic outbreak, similar to what we are currently experi-encing. In several recent publications we and others have describedselection of highly productive vaccine production cell substratesand development of manufacturing processes for flu vaccine pro-duction [7,11,15,16]. These efforts have contributed to the solution

ccine 2

otesapudcaflefdi

matAtaa4termtidsticitttvisstedair[pbIitat7h

NfvtWf

K. Aggarwal et al. / Va

f some of the problems unique to egg-based vaccine. However,here still remain a number of technical issues associated with bothgg- and cell culture-based production platforms. For example, thehort production cycle that often leads to limited manufacturingnd process development time and potentially compromises theroductivity and robustness of the developed processes. It is notnusual for the flu vaccine manufacturers to produce virus seeds,evelop individual processes for each of the 3 vaccine strains andomplete technology transfer to the manufacturing facility within2–3 month timeframe. To minimize the impact of the compressedu development cycle on manufacturing process quality we havexamined several process parameters that may influence the per-ormance of flu vaccine manufacturing process. Among them weescribe three factors, time of infection, harvest time and virus

nput or MOI.As viruses use host cells to replicate it is not surprising that

ore progeny viruses are produced when more host cells arevailable, at least during the initial stage following virus infec-ion. Consistent with this phenomenon our study shows that ca/Wisconsin/67/05 and other ca viruses are produced at higher

iters at later times after seeding when there are more cells avail-ble for virus infection than at early times post cell seeding. Forll the viruses that we tested, the preferred time of infection is–5 dps when a higher VCD is achieved and the cells, according tohe cell growth curve, are either in the late exponential phase orntering stationary phase. Since viruses use host cell machinery toeplicate, the physiological state of the cells at the time of infectionay also affect peak virus titer. Although mechanistic interpreta-

ions are beyond the scope of this particular study and were notnvestigated, the consistent virus amplification ratio observed atifferent times post infection suggests that ca flu viruses are moreensitive to VCD than changes in cell physiology associated withhe cell growth at different time post cell seeding. In addition, its important to note that the virus harvest time is fairly short andan be affected by other factors such as MOI. At lower MOIs thenitial virus replication seems to be slow relative to high MOI infec-ions. In general, after reaching peak titers 2–3 dpi, the ca flu virusiter remains constant for less than 24 h under the tested condi-ions and then decreases quickly. This is not unexpected as theseiruses go through a lytic infection cycle and as they are releasednto the cell culture fluid become vulnerable to inactivation out-ide the more protected intracellular environment. Therefore, inpite of the continuing production of progeny viruses, the virusiter does not increase once the production and inactivation reachesquilibrium. After peak titer is reached, virus production slowsown due to continued destruction of host cells and reduction ofvailable cellular machinery for virus replication. Thus, the virusnactivation rate exceeds the virus production rate resulting in aeduction of the measured titer of viruses after peak titer is reached17]. We conclude that both time of infection and harvest timelay important roles in determining virus productivity and shoulde optimized during vaccine manufacture process development.

t is our experience that these two process parameters are eas-ly managed as they do not tend to change significantly amonghe viruses that we tested and their range is relatively narrownd well defined. A manufacturing process with a time of infec-ion between 4 and 5 dps and time of harvest between 48 and2 hpi consistently produced bulk vaccines at high titers in ourands.

However, virus yield fluctuates greatly with changes in MOI.o single optimum MOI has been identified to be applicable

or all the 25 vaccine strains we have analyzed. The optimumirus input is virus strain specific. This factor has been iden-ified as the most impactful process parameter in our study.

e have built an MDCK cell-based flu vaccine production plat-orm by optimizing this parameter without changing cell culture

9 (2011) 3320–3328 3327

medium, cell growth and virus production temperature, micro-carriers concentration, cell seeding density, pH, DO and manyother process parameters from strain to strain and from cam-paign to campaign. Although the vaccine production time maybe extended for an extra 1 or 2 days for some vaccine strainsbecause of the low MOI infection, this loss of time is compen-sated through reduction of process development time as a resultof reduced development efforts associated with optimization ofa smaller set of process parameters. We have successfully pro-duced a high titer vaccine lot with a three week development timebased on a sudden and urgent request for a pandemic vaccinestrain prior to our scheduled seasonal flu production campaign.The strong correlation between peak titer and virus input hasbeen described by several laboratories [15,18,19]. However, muchof the reported work investigates the effect of MOI in a higherrange (between 1 and 1 × 10−5 infectious virus units/cell) and thevirus titer remained unchanged or decreased with use of lowerMOI. In the work reported here, a lower range of MOI (10−6 to10−2 FFU/cell) has been tested and the impact on virus produc-tion of a large panel of virus strains determined (up to 25 differentinfluenza vaccine strains). Irrespective of the type and subtype ofviruses analyzed, virus inputs less than 10−3 FFU/cell consistentlyyielded higher peak titers compared to those obtained at virusinputs above 10−3 FFU/cell. On average, a three fold productivityimprovement (∼0.5 log10 FFU/ml) was observed using the manu-facturing process described here by making a change to a singleprocess parameter, virus input. For specific virus strains such asca B/Florida/07/04 an even greater increase in virus productivity,up to a 30-fold improvement, was achieved. With MOI optimiza-tion, the range of peak titer variation for these 25 vaccine strainshas been decreased from 1.7 log10 FFU/ml (7.1–8.8 log10 FFU/ml)to 1.5 log10 FFU/ml (7.7–9.2 log10 FFU/ml), achieving a reduction of0.2 log10 FFU/ml in productivity variations among different vaccinestrains. Furthermore, the lowest attained peak titer was increasedfrom 7.1 log10 FFU/ml to 7.7 log10 FFU/ml. It is clear that optimiza-tion of MOIs not only increases process productivity but alsoimproves process robustness. Use of smaller amount of virus seedat the time of infection also reduces exhaustion of and thus extendsthe lifespan of the costly master virus banks. The mechanism(s) bywhich lowering of MOI significantly improves ca influenza virusproductivity is not fully understood, however, we speculate thepresence and/or production of defective interfering particles (DIP)in higher MOI infections plays a role in reducing the productionand accumulation of infectious progeny viruses. Krell [20] reportedthat low MOI minimizes the generation of DIP during successivevirus amplification steps and thus increases production of infec-tious viruses. We are currently investigating this hypothesis byanalyzing the ratio of DIP and infectious viruses under differentinfection conditions.

This work confirms the observations made by others forinfluenza and other viruses that time of infection, harvest timeand MOI are critical process parameters that greatly affect theproductivity and robustness of flu vaccine manufacturing pro-cesses. Although the reported experiments are performed usingcold adapted influenza viruses the observations made are appli-cable to other viruses, specifically non-LAIV influenza viruses assimilar observations are also made for non-LAIV viruses such aswild type influenza viruses [20–22] and in other host cells suchas Vero and HEK-293 cells [21,23]. Thus it has broader applica-tions to influenza vaccine production. The work presented herealso demonstrates the feasibility of a process development strat-

egy for converting an existing process into a platform processfor flu vaccine production and calls for optimization of time ofinfection, virus input and harvest time in a virus specific man-ner for the production of all new vaccine virus strains using MDCKcells.

3 ccine 2

A

fRmHctfsg

R

[

[

[

[

[

[

[

[

[

[

[

[

328 K. Aggarwal et al. / Va

cknowledgements

This project is funded in whole or in part with Federal fundsrom the Office of the Assistant Secretary for Preparedness andesponse (ASPR), Biomedical Advanced Research and Develop-ent Authority, under Contract Nos. HHSO100200600010C andHSO100200700036C. The total federal program funding for theseontracts is $221,379,570, representing approximately 92% of theotal amount for the projects. The remaining 8% of the total amountor the projects is anticipated to be financed by nongovernmentalources. We would also like to thank Vaccine Analytical Sciencesroup at MedImmune for performing the FFA assay.

eferences

[1] CDC. Prevention and control of influenza: recommendations of the AdvisoryCommittee on Immunization Practices (ACIP). MMWR 2006;55(RR10):1–42.

[2] Kemble G, Greenberg H. Novel generations of influenza vaccines. Vaccine2003;21(16):1789–95.

[3] Cox MM, Hollister JR. FluBlok, a next generation influenza vaccine manufac-tured in insect cells. Biologicals 2009;37(June (3)):182–9.

[4] Vemula SV, Mittal SK. Production of adenovirus vectors and their use as adelivery system for influenza vaccines. Expert Opin Biol Ther 2010;10(October(10)):1469–87.

[5] Barrett PN, Portsmouth D, Ehrlich HJ. Developing cell culture-derived pandemicvaccines. Curr Opin Mol Ther 2010 Feb;12(1):21–30.

[6] Gerdil C. The annual production cycle for influenza vaccine. Vaccine2003;21(May (16)):1776–9.

[7] Tree JA, Richardson C, Fooks AR, Clegg JC, Looby D. Comparison oflarge-scale mammalian cell culture systems with egg culture for the pro-duction of influenza virus A vaccine strains. Vaccine 2001;19(May (25–26)):3444–50.

[8] Mabrouk T, Ellis RW. Influenza vaccine technologies and the use ofthe cell-culture process (cell-culture influenza vaccine). Dev Biol (Basel)2002;110:125–34.

[9] George M, Farooq M, Dang T, Cortes B, Liu J, Maranga L. Production of cell culture(MDCK) derived live attenuated influenza vaccine (LAIV) in a fully disposableplatform process. Biotechnol Bioeng 2010;106(April (6)):906–17.

[

[

9 (2011) 3320–3328

10] Liu J, Mani S, Schwartz R, Richman L, Tabor DE. Cloning and assessment oftumorigenicity and oncogenicity of a Madin–Darby canine kidney (MDCK) cellline for influenza vaccine production. Vaccine 2010;28(February (5)):1285–93.

11] Liu J, Shi X, Schwartz R, Kemble G. Use of MDCK cells for production of liveattenuated influenza vaccine. Vaccine 2009;27(October (46)):6460–3.

12] Belshe RB, Edwards KM, Vesikari T, Black SV, Walker RE, Hultquist M, et al. Liveattenuated versus inactivated influenza vaccine in infants and young children.N Engl J Med 2007;356(February (7)):685–96.

13] Baez M, Palese P, Kilbourne ED. Gene composition of high-yielding influenzavaccine strains obtained by recombination. J Infect Dis 1980;141(March(3)):362–5.

14] Wei Z, McEvoy M, Razinkov V, Polozova A, Li E, Casas-Finet J, et al.Biophysical characterization of influenza virus subpopulations using fieldflow fractionation and multiangle light scattering: correlation of particlecounts, size distribution and infectivity. J Virol Methods 2007;144(September(1–2)):122–32.

15] Genzel Y, Olmer RM, Schafer B, Reichl U. Wave microcarrier cultivation of MDCKcells for influenza virus production in serum containing and serum-free media.Vaccine 2006;24(August (35–36)):6074–87.

16] Ghendon YZ, Markushin SG, Akopova II, Koptiaeva IB, Nechaeva EA, MazurkovaLA, et al. Development of cell culture (MDCK) live cold-adapted (CA) attenuatedinfluenza vaccine. Vaccine 2005;23(September (38)):4678–84.

17] Teale A, Campbell S, Van Buuren N, Magee WC, Watmough K, Couturier B,et al. Orthopoxviruses require a functional ubiquitin-proteasome system forproductive replication. J Virol 2009;83(March (5)):2099–108.

18] Maranga L, Brazao TF, Carrondo MJT. Virus-like particle production at low mul-tiplicities of infection with the baculovirus insect cell system. Biotechnol Bioeng2003;84(October (2)):245–53.

19] Yuk IH, Lin GB, Ju H, Sifi I, Lam Y, Cortez A, et al. A serum-free Vero productionplatform for a chimeric virus vaccine candidate. Cytotechnology 2006;51(July(3)):183–92.

20] Krell PJ. Passage effect of virus infection in insect cells. Cytotechnology1996;20(1–3):125–37.

21] Ru AL, Jacob D, Transfiguracion J, Ansorge S, Henry O, Kamen AA. Scalable pro-duction of influenza virus in HEK-293 cells for efficient vaccine manufacturing.Vaccine 2010;28(May (21)):3661–71.

22] Mohler L, Flockerzi D, Sann H, Reichl U. Mathematical model of influenzaA virus production in large-scale microcarrier culture. Biotechnol Bioeng2005;90(April (1)):46–58.

23] Youil R, Su Q, Toner TJ, Szymkowiak C, Kwan WS, Rubin B, et al. Comparativestudy of influenza virus replication in Vero and MDCK cell lines. J Virol Methods2004;120(September (1)):23–31.