Ž .Journal of Immunological Methods 230 1999 141–147www.elsevier.nlrlocaterjim

Improved cell line development by a high throughput affinitycapture surface display technique to select for high secretors

Paul Holmes, Mohamed Al-Rubeai )

Animal Cell Technology Group, School of Chemical Engineering, The UniÕersity of Birmingham, Edgbaston, Birmingham B15 2TT, UK

Received 2 July 1999; received in revised form 25 August 1999; accepted 20 September 1999

Abstract

A novel process is described which permits rapid and objective selection of rare cells from a heterogeneous populationbased on quantity of secreted target protein. The process involves construction of an immobilised affinity surface displaymatrix that specifically binds secreted target product which is then detected using a fluorescent labelled ligand. Cells withthe highest fluorescence can then be sorted using conventional flow cytometric technology. Overall, the whole process canbe completed in less than 4 h during which time in the region of five million cells can be analysed. Cells are rapidly selectedfor in a quantitative manner compared to traditional methods which can take several months and have a reduced probabilityof finding low abundance high secretors due to practical limitations imposed on the number of cells which can be screened.q 1999 Elsevier Science B.V. All rights reserved.

Keywords: Affinity capture surface display; Secretion assay; Cell selection; Cell enrichment; High throughput screening; FACS

1. Introduction

Selection of high producing cell lines is an impor-tant first step in the development of any bioprocess.For protein production from animal cells, such se-lected cell lines have traditionally been delivered byrounds of limiting dilution cloning followed by prod-uct analysis. However, there are several drawbacksto this route. In the first instance, it is both labourintensive and costly. Secondly, the whole process istime consuming because two rounds of cloning arerequired to improve the theoretical confidence of

) Corresponding author. Tel.: q44-121-414-3888; fax: q44-121-414-5324; e-mail: m.al-rubeai@bham.ac.uk

achieving clonality. Clonality is important to avoidovergrowth of high producers by low productivity

Žvariants Frame and Hu, 1990; Ozturk and Palsson,.1990 which usually have higher growth rates than

Ž .high producers Richieri et al., 1991 . As such, thewhole process can take in excess of 8 months tocomplete and even then there is no guarantee that thecloned cell line will be stable and so useful forindustrial bioprocessing. Finally, selection of thehighest producers can be compromised by practicallimitations on the number of cells that can bescreened thereby potentially reducing the efficiencyof selection of low abundance, high productivitycells. Clearly there is a need to develop an objective,rapid and simple method to select for high producingcell lines.

0022-1759r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved.Ž .PII: S0022-1759 99 00181-7

( )P. Holmes, M. Al-RubeairJournal of Immunological Methods 230 1999 141–147142

Ž .Flow cytometry FC has made it easier to moni-tor productivity and to isolate cells with specific

Ž .characteristics Al-Rubeai, 1999 . Important advan-tages of FC include the ability to screen large num-bers of cells rapidly, the capability to distinguish cellsub-populations and the ability to efficiently selectlow abundance cells demonstrating the desired char-acteristics. There are two possible approaches for theselection of high productivity cells using FC, both ofwhich have been successfully applied to the selectionof hybridoma cells. The first approach is based onthe cell surface antibody content; hybridoma cellsdisplaying an increased amount of cell surface anti-body can be identified and recovered through the useof fluorescent labelled antibodies to the hybridomaproduct. Antigen specific hybridomas, isotype switch

Ž .variants Dangl et al., 1982 , higher avidity variantsŽ .Martel et al., 1988 and bispecific hybridomasŽ .Jantscheff et al., 1993 have all been identified andcloned based on characterisation of cell surface im-munoglobulin. While these groups have shown aqualitative correlation between cell surface and se-creted antibody, a quantitative correlation has notbeen broadly documented. Indeed, a kinetic analysisof hybridoma clones reported no correlation betweenthe amount of cell surface antibody and the amount

Žof antibody in cell culture supernatants Meilhoc et.al., 1989 . However, it is possible that this discrep-

ancy in quantitative analysis was due to the authorsoverlooking the effects of cell surface area. Highsurface antibody content may be related to the largersurface area of cells during S and G2 cell cyclephases and may give false values for cells that areactually low producers. Normalisation of cell surfaceantibody concentration per unit area and synchroni-sation of cells so that they are examined in the samecell cycle phase represent approaches for optimisa-tion which may enable the application of this simple

Žmethod of product monitoring and selection Cherlet.et al., 1995 .

Recently, approaches have been developed to se-lect cells based on secreted antibody as an alternativestrategy to circumvent some of the limitations of cellsurface antibody selection. Two broad approacheshave been developed, namely the affinity matrix and

Žthe gel microdrop Gray et al., 1995; Kenney et al.,.1995; Manz et al., 1995 . The former method is

based on creation of an artificial affinity matrix,

specific for the secreted product of interest. Secretedmolecules bind to the affinity matrix on the surfaceof the secreting cell and are subsequently labelledwith specific fluorescent reagents for flow cytomet-ric analysis and cell sorting. The matrix itself iscreated by direct linkage of avidinated specific cap-ture antibodies to the previously biotinylated cellsurface. Secreted antibodies are retained directly onthe surface of the originating cell and product crossfeeding is prevented by using a medium of lowpermeability. Thus, it is possible to analyse andselect cells on the basis of secretory activity. Thepotential efficiency of this technique may be dictatedby a combination of the affinity constant and capac-ity of the cell surface matrix for the secreted product,and the product diffusion rate in the low permeabil-

Ž .ity medium Frykman and Scrienc, 1998 .Microdrop encapsulation involves, as its name

implies, complete encapsulation of single cells inŽagarose beads Gray et al., 1995; Kenney et al.,

.1995 . These beads contain specific capture antibod-ies and so simultaneously capture secreted productand prevent cross feeding of product between cells.Again the quantity of secreted product is determinedby binding of a fluorescent ligand and the cells aresorted or cloned by FC while still encapsulatedwithin the microdrop. The ability to quantify secre-tion at the single cell level is a unique and com-pelling feature of these technologies, since it is nototherwise possible to analyse and rapidly sort indi-vidual cells quantitatively for secretory activity.

However, these techniques are not without short-comings. Microdrop encapsulation is frequently de-scribed as being user unfriendly, needing much opti-misation of conditions for each cell line used andalso requires dedicated equipment and extra time forencapsulation. Additionally, to ensure single cell oc-cupancy of the microdrops only about 5% of thedrops created actually contain a cell, consequentlyaround 95% of the particles analysed by the sorterare devoid of cells and are in effect wasted. Also,decapsulation can reduce the viability of some celltypes commonly used for expression of recombinantproteins, most notably myeloma NS0 cells. The di-rect binding approach addresses some of these prob-lems. For example, it is simpler and requires neitherencapsulation or decapsulation of the cells, but itdoes require chemical modification of some of the

( )P. Holmes, M. Al-RubeairJournal of Immunological Methods 230 1999 141–147 143

reagents and the product saturation limit is theoreti-cally an order of magnitude lower than for encapsu-

Ž .lated cells Frykman and Scrienc, 1998 . Despitethese shortcomings the simplicity of direct bindingmakes it the approach most likely to achieve theaims highlighted above. Additionally, incorporationof some form of correction for cell size effect, whichis a key factor in determining secretion levels for

Ž .mammalian cells Holmes and Al-Rubeai, in press ,would be desirable but has yet to be considered byeither of the methods described above.

Clearly there is interest in developing a generictechnology for selecting high producing cells basedon secreted product. Such a technology needs to becheap, simple enough to become routine, highthroughput, very rapid, objective and adaptable tomany different secreted products. The developmentof such a technology was the purpose of this work.

2. Materials and methods

2.1. Cell line

Ž .NSO 6A1 100 -3, a myeloma cell line expressingb72.3, a chimeric antibody specific to the breasttumour antigen TAG73, was given to this laboratoryby Lonza Biologics. This cell line utilises the power-ful GS promoter system for high level expressionand was originally cloned by the limiting dilution

Ž .method Bebbington et al., 1992 . These cells wereroutinely maintained by passaging every third day

Ž .into fresh GMEM Gibco, UK supplemented withŽ .5% Foetal Calf Serum Gibco , 500 mM glutamic

acid, 500 mM asparagine, adenosine, guanosine, cy-tidine and uridine all at 30 mM, 10 mM thymidine, 1mM sodium pyruvate, 1= non-essential amino acids

Žand 10 mM methionine sulphoximine from Sigma.unless stated . Cells were grown at 378C in 100 ml

Duran bottles agitated by magnetic followers.

2.2. Affinity surface display matrix construction

For construction of the affinity surface displaymatrix 107 cells were washed in PBS adjusted to pH8 with 1N NaOH, resuspended in 1 ml of filter

Žsterilised NHS-LC-biotin 1 mgrml in PBS, pH 8;

.Pierce, UK and incubated at room temperature for20 min. Cells were then washed twice in 25 ml PBSŽ .pH 7 and resuspended in 1 ml of medium contain-

Ž . Žing gelatine 10% wrv; Sigma, UK , neutravidin 4.mg; Pierce, UK , 26 ml biotinylated rabbit anti hu-

Ž .man IgG4 FC specific; Sigma, UK . The cells wereallowed to secret product for up to 80 min at 378Cafter which the gelatine was removed by washing

Ž .twice in 25 ml PBS pH 7 at 378C. Cells wereresuspended in 1 ml of warm medium and boundproduct detected using mouse anti-human kappa light

Ž .chain fluorescein conjugate Sigma, UK at a finaldilution of 1r500.

2.3. Flow cytometry

All cell sorting was done using a Becton Dickin-son 440 with 488 nm and 515 nm excitation anddetection wavelengths, respectively. Cells were se-lected based on integral fluorescence to forwardscattered light ratio, as this allows for some correc-tion for cell size. The 10% of the cells with thehighest fluorescence ratio were selected as for thehigh secretors. Selected cells were washed and platedout at 2=105 cellsrml. After 2 h, cells’ supernatantwas taken and product concentration was analysedby enzyme-linked immunosorbent assay as describedbelow.

2.4. Product analysis

Nunc Maxisorp 96-well plates were coated withŽ .mouse anti-human IgG4 FC specific antibodies

Ž .Sigma, UK overnight at 48C. After washing fiveŽ .times with PBS-Tween 0.05% , unoccupied sites on

Žthe plates were blocked with skimmed milk 1%.wrv in PBS for 1 h. The blocking buffer was then

removed and replaced with analysate at a suitabledilution in blocking buffer and incubated at 378C for1 h. The plates were washed five times and boundchimeric antibody was detected using mouse anti-hu-man kappa light chain horseradish peroxidase conju-

Ž .gate Sigma, UK for 1 h. Colour development wasachieved using 100 ml o-phenylenediamine dihydro-

Žchloride 0.4 grl in 0.05 M citric acid, 0.1 MŽ .di-sodium phosphate, 0.004% vrv hydrogen perox-

.ide per well as a chromogen. The reaction was

( )P. Holmes, M. Al-RubeairJournal of Immunological Methods 230 1999 141–147144

stopped after 30 min using 100 ml 2.5 M sulphuricacid and absorbance values at 495 nm were deter-mined using a SLT Spectra ELISA plate reader.Product concentration was determined by linear re-gression against external standards.

3. Results and discussion

3.1. Receptor localisation

The specific and essentially irreversible bindingof avidin and biotin has great potential not only forimmobilisation of ligands to cell surfaces but also foramplification, due to avidin having a valency forfour biotin molecules. Additionally, biotinylation ofcells and proteins is a relatively straightforward pro-cedure and there are also many biotinylated antibod-ies readily available commercially at relatively lowcost. We designed our surface display capture matrixas shown in Fig. 1 where cells are surface biotinyl-

˚ated, the biotin moiety incorporating a 24 A spacerarm to reduce chances of steric hindrance. The bio-

Fig. 1. Illustration of the affinity surface display matrix showinghow biotinylated capture antibodies are bound to biotinylated cellsthrough neutravidin. A fluorescent detection antibody then quanti-fies captured secreted product. Cross-feeding of product betweenthe cells is prevented by using medium with artificially elevated

Ž .viscosity not shown .

tinylated cells were then mixed with biotinylatedcapture antibody and neutravidin. We used neutra-vidin because it has less non-specific binding thanavidin. This process allowed the capture antibodyand cell to become irreversibly linked via their re-spective biotins through the neutravidin moiety.After allowing the cells to secrete product, whilepreventing product cross feeding between cells byartificially elevating the viscosity of the medium, inthis case with 10% gelatine, detection of the boundproduct was done utilising commercially availableFITC conjugated antibodies. However, in the ab-sence of such conjugates being commercially avail-able, indirect staining would be an appropriateoption. Construction and detection of this immo-bilised surface display matrix did not, based on batchgrowth profiles of biotinylated compared to control

Ž .cells, adversely affect cell viability data not shown .Clearly our aims of producing a generic process

are satisfied here because any cell type with freeamino groups at its surface should be amenable tobiotinylation. As such, this technique should be ap-plicable not only to all kinds of animal cells but alsoto plant and microbial cells if so required. The keydifference in receptor localisation here and in thedirect binding method was how the biotinravidinsystem was used. Our system provides at least twodistinct advantages, namely amplification and sim-plicity. By moving to reagents which are commer-cially available we completely alleviated any need

Ž .for reagent modification antibody avidination andthus ensured that the system is very rapid and alsosimple enough to become a routine operation in anylaboratory using materials which are relatively cheap,easily acquired and permit better quality assurance ofthe process. Simplicity and timeliness were demon-strated by yielding cells ready for analysis within3 h.

We used 10% gelatine to increase the viscosity ofthe medium and prevent product transfer betweencells because this agent had been previously reported

Žas suitable for the purpose Manz et al., 1995; Fryk-.man and Scrienc, 1998 . However, its bovine source

is likely to undermine FDA validation of this processfor selecting cells for subsequent use in bioprocess-ing. Consequently, suitable alternatives to gelatinewould be desirable as long as they fulfil certaincriteria. Firstly, they should be non-toxic to the cells.

( )P. Holmes, M. Al-RubeairJournal of Immunological Methods 230 1999 141–147 145

Secondly, to avoid contamination of the cells withbiological agents, they should not be from bovinesources and preferably not sourced from animals atall. Finally, they should be viscous but not solidifiedat 378C, easily removed by thorough washing andpreferably autoclavable. Potential candidates include,among others, methyl cellulose, PEG and a varietyof gums sourced from plants.

The two antibody system shown in Fig. 1 issuitable for quantification of secreted product. How-ever, when developing hybridoma cell lines, anti-body specificity for the target epitope is initially themost important consideration. By using a fluorescentconjugated epitope, instead of labelled antibodies, itshould be possible to rapidly obtain antibodies topredefined epitopes. Potentially this technology couldbe used to rapidly acquire banks of diagnostic mono-clonal antibodies to multiple epitopes along a wholelength of a protein of interest or to rapidly acquirepairs of antibodies to the same epitope but which aremutually exclusive in terms of post-translationalmodification. Such pairs of antibodies could particu-larly find application in medical research one spe-cific example being diseases such as Alzheimer’sdisease where the pathology is manifested as hyper-phosphorylation of certain proteins compared to thenon-diseased state.

The principle of this technology for rapid selec-tion of antibodies with the desired specificity couldbe extended further to screening and selection ofdesired clones from random combinatorial antibodylibraries expressed in mammalian cells. This couldbe an alternative to selection of binding activities

Žexpressed in phage surface display libraries Cwirlaet al., 1990; McCafferty et al., 1990; Scott and

.Smith, 1990 then subcloning sequences encodingthe desired binding characteristics into animal cellexpression vectors. Potentially, the whole process ofgetting the product to the market place would begreatly expedited. Also, problems that may be asso-ciated with potentially useful sequences being missedbecause they are not compatible with the phage lifecycle would be avoided.

3.2. Saturation kinetics

Rapid system saturation could potentially be alimiting factor with this method and so was investi-

gated by following the amount of secreted productover time. An isotherm of saturation of this system isshown Fig. 2. Saturation was achieved between 30and 40 min which, according to specific productivity

Ž .data by ELISA see Fig. 3 corresponds to at most0.2 pg or 8=105 moleculesrcell. Considering thathigh producing cells typically secrete product in theregion of tenths of picograms per hour, our systemremains suitable for cells with secretion rates of oneor possibly two orders of magnitude greater thanthis. Thus demonstrating that this technique is usefulover what are reasonably expected secretion rates.

3.3. Cell sorting

Gates were set to collect the 10% of cells havingthe highest ratio of FITC fluorescence to forward

Ž .scatter FS as shown in Fig. 3. Collected cells weresubcultured and supernatant analysed, showing thatcompared to the original cloned population thesesorted cells had 30% higher specific productivityŽ .Fig. 3 . Thus, the potential for enrichment of highlevel secretors from a heterogeneous population butwith only a small variation between high and low

Fig. 2. Saturation isotherm of the affinity surface display matrix.Saturation is reached after about 40 min, which equates to approx-imately 0.2 pg of product per cell.

( )P. Holmes, M. Al-RubeairJournal of Immunological Methods 230 1999 141–147146

Fig. 3. Sorting of high producers from a cloned myeloma cell lineŽ .expressing a chimeric antibody. Top 10% of cells with the

Žhighest FITCrFS ratio were sorted and then subcultured. Bot-.tom ELISA analysis of the supernatant of the sorted cells shows

that the productivity of the sorted cells is 30% greater than theoriginal cloned population.

producers is demonstrated. The time from beginningthis protocol to having cells ready for subculturingwas in the region of 4 h and during this time nearlyfive million cells were analysed thus demonstratingthe high throughput potential of this technique.

Producing cells were mixed with non-producingNS0 cells in a ratio of 1 to 4 and gates set to selectthe 10% of cells with the highest FITCrFS ratio.Selected cells were plated out onto 96-well platesand after 2 weeks analysed for product. Of the wells

containing cells, 160 out of 288 wells, 94% of thesewells were positive for product thereby demonstrat-ing the use for this technique for sorting producingfrom non-producing cells. A typical application herecould be to sort hybridomas secreting product to adesired epitope from the heterogeneous population ofa hybridoma fusion using a fluorescent conjugatedligand. A second round of screening as describedabove to select for high producing cells would fol-low this.

Decapsulation of CHO cells after sorting by themicrodrop system does not influence cell viability.However, some important cell types, including NS0which we successfully sorted here, can be problem-atic resulting in loss of viability. Ability to sort thesemore ‘fragile’ cell types may make our techniqueparticularly suitable for sorting primary cell linesaccording to the quantity of secreted product.

Modern FC facilities can include equipment suchas autoclone which could be advantageous by allow-ing concurrent selection and a cloning of high pro-ducing cells by directing the sorted cell to a specificwell of a 96-well plate. Laboratories without auto-clone could easily sort cells by the procedure de-scribed above and then clone the selected high pro-ducers by dilution cloning. As we have shown abovethe vast majority of these cells will be not onlyproducers but also high producers. However, it wasour stated aim at the outset to produce a techniquethat could be readily adopted by any laboratory.Though a powerful and useful technique, FC is notnecessarily an obligate requirement for this work.Laboratories without access to FC could select forcells by reverting to separation technologies used inthe early development of the phage display libraries.For example, they could use detection ligands boundto solid phases to select for positive cells and usesimple cheap techniques such as affinity panningŽ .Scott and Smith, 1990 or magnetic separationŽ .Manz et al., 1995 to select for high producing cellswhere FC is not available.

Acknowledgements

This work was funded by the EC framework IVprogramme.

( )P. Holmes, M. Al-RubeairJournal of Immunological Methods 230 1999 141–147 147

References

Al-Rubeai, M., 1999. Monitoring of growth and productivity ofŽ .animal cells by flow cytometry. In: Jenkins, N. Ed. , Animal

Cell Biotechnology: Methods in Biotechnology, Vol. 8. Hu-mana Press, Totawa, NJ, USA, pp. 145-153.

Bebbington, C.R., Renner, G., Thomson, S., King, D., Abrams,D., Yarranton, G.T., 1992. High level expression of a recom-binant antibody from myeloma cells using and amplifiableselectable marker. Biotechnology 10, 169–175.

Cherlet, M., Kromenaker, S.J., Srienc, F., 1995. Surface IgGcontent of murine hybridomas: direct evidence for variation inantibody secretion rates during cell cycle. Biotechnol. Bioeng.47, 535–540.

Cwirla, S., Peters, E.A., Barrett, R.W., Dower, W.J., 1990. Pep-tides on phage: a vast library for identifying ligands. Proc.Natl. Acad. Sci. U. S. A. 87, 6378–6382.

Dangl, J.L., Parks, D.L., Oi, V.T., Herzenberg, L.A., 1982. Rapidisolation of cloned isotype switch variants using fluorescenceactivated cell sorting. Cytometry 2, 395–401.

Frame, K.K., Hu, W.S., 1990. The loss of antibody productivity incontinuous culture of hybridoma cells. Biotechnol. Bioeng. 45,469–476.

Frykman, S., Scrienc, F., 1998. Quantitating secretion rates ofindividual cells: design of secretion assays. Biotechnol. Bio-eng. 59, 214–226.

Gray, F., Kenney, J.S., Dunne, J.F., 1995. Secretion capture andreport web: use of affinity derivatised agarose microdropletsfor the selection of hybridoma cells. J. Immunol. Methods182, 155–163.

Holmes, P., Al-Rubeai, M., in press. Cell cycle synchronisation.Ž .In: Spier, R. Ed. , Encyclopedia of Cell Technology. Wiley,

New York, USA.

Jantscheff, P., Winkler, L., Karawajew, L., Kaiser, G., Bottger,V., Micheel, B., 1993. Hybrid hybridomas producing bispe-cific antibodies to cea and peroxidase isolated by a combina-tion of hat medium selection and fluorescence-activated cellsorting. J. Immunol. Methods 163, 91–97.

Kenney, J.S., Gray, F., Ancel, M.H., Dunne, J.F., 1995. Produc-tion of monoclonal antibodies using a secretion capture reportweb. Biotechnology 13, 787.

Manz, R., Assenmacher, M., Pfluger, E., Miltenyi, S., Radbruch,A., 1995. Analysis and sorting of live cells according tosecreted molecules, relocated to a cell surface affinity matrix.Proc. Natl. Acad. Sci. U. S. A. 92, 1921–1925.

Martel, F., Bazin, R., Verrette, S., Lemieux, R., 1988. Characteri-zation of higher avidity monoclonal-antibodies produced bymurine b-cell hybridoma variants selected for increased anti-gen-binding of membrane IgG. J. Immunol. 141, 1624–1629.

McCafferty, J., Griffiths, A., Winter, G., Chiswell, D.J., 1990.Page antibodies: filamentous phage displaying antibody vari-able domains. Nature 348, 552–554.

Meilhoc, E., Wittrup, K.D., Bailey, J.E., 1989. Application offlow cytometric measurement of surface IgG in kinetic analy-sis of monoclonal antibody synthesis and secretion by murinehybridoma cells. J. Immunol. Methods 121, 167–174.

Ozturk, S.S., Palsson, B.O., 1990. Loss of antibody productivityduring long-term cultivation of a hybridoma cell-line in lowserum and serum-free media. Hybridoma 9, 167–175.

Richieri, R.A., Williams, L.S., Chau, P.C., 1991. Cell-cycle de-pendency of monoclonal-antibody production in asynchronousserum-free hybridoma cultures. Cytotechnology 5, 243–254.

Scott, J.K., Smith, G.P., 1990. Searching for peptide ligands withan epitope library. Science 249, 386–390.