Mammalian display to secretion switchable libraries for antibody preselection and high throughput functional screening

ABSTRACT

Recently, there has been a co-evolution of mammalian libraries and diverse microfluidic approaches for therapeutic antibody hit discovery. Mammalian libraries enable the preservation of full immune repertoires, produce hit candidates in final format and facilitate broad combinatorial bispecific antibody screening, while several available microfluidic methodologies offer opportunities for rapid high-content screens. Here, we report proof-of-concept studies exploring the potential of combining microfluidic technologies with mammalian libraries for antibody discovery. First, antibody secretion, target co-expression and integration of appropriate reporter cell lines enabled the selection of in-trans acting agonistic bispecific antibodies. Second, a functional screen for internalization was established and comparison of autocrine versus co-encapsulation setups highlighted the advantages of an autocrine one cell approach. Third, synchronization of antibody-secreting cells prior to microfluidic screens reduced assay variability. Furthermore, a display to secretion switchable system was developed and applied for pre-enrichment of antibody clones with high manufacturability in conjunction with subsequent screening for functional properties. These case studies demonstrate the system’s feasibility and may serve as basis for further development of integrated workflows combining manufacturability sorting and functional screens for the identification of optimal therapeutic antibody candidates.

KEYWORDS:

• Antibodies
• display secretion switch
• function first screen
• mammalian libraries
• microfluidics

Introduction

In the field of therapeutic antibody discovery, recent years have seen a co-evolution of various microfluidic technologies, next-generation sequencing (NGS) in combination with high-capacity machine learning, and mammalian libraries for effective identification of optimal biotherapeutics. While other researchers have elegantly shown the applicability of combining microfluidics with NGS and deep learning,^1,2 we sought to evaluate the potential of integrating mammalian libraries in both display and secretion modes with microfluidic function first screens.

Mammalian libraries are often generated in human embryonic kidney (HEK293) or Chinese hamster ovary (CHO) cells by means of lentiviral delivery,^3–8 vaccinia virus,⁹ nucleases,^10–12 recombinases,^13–18 transposases,¹⁹ or episomal approaches²⁰ in conjunction with somatic hypermutation²¹ or unidirectional gene conversion.²² Such libraries enable the preservation of full immune repertoires similar to phage or yeast display libraries. They produce hit candidates in close to final production host cells and often final therapeutic formats such as bispecific T-cell engager (BiTE) or IgG-VHH fusion bispecific antibodies, potentially lowering attrition in preclinical development. Construction of mammalian libraries beyond 100 million diverse clones can be comparably sophisticated, but smaller library sizes are feasible and represent a perfect fit to microfluidic workflow throughputs of mostly 50,000 to a few million events. Libraries can be designed to display the candidates on the surface of the cell, secrete them into the supernatant, or enable simultaneous display and secretion.^21,23 Mammalian display has been used for affinity maturation,^6,12,19–21 to screen for enhanced manufacturability,²⁴ or in an autocrine binding for a desired function.⁴ Conversely, secretion libraries have been applied in combination with microfluidics-assisted screens.^25–27 Diverse microfluidic screening approaches are based on compartmentalization in droplets,²⁸ nanostructures such as nanopens,²⁹ or microcapillaries^30–32 using mostly fluorescence-based methods for binding to recombinant targets or cells,³³ for internalization or functional screening.^1,34–36

Here, we report several case studies exploring the combination of mammalian libraries with rapid microfluidic screening for three different modes of action. We show that cell synchronization could reduce assay variability and high viability of robust applied CHO cells enabled prolonged incubation for longer functional assay readouts, allowing for comprehensive characterization of antibody candidates. Moreover, this combined approach facilitated the screening of diverse final therapeutic antibody formats, including IgG1, BiTEs and IgG-VHH fusions. Proof-of-concept spiking studies demonstrated the applicability of the function first approach in various screening scenarios, including screening for effective T-cell engagers, identification of targeted agonists to a tumor necrosis factor receptor superfamily member (termed herein TNFRSF) and screening for internalization of antibodies for antibody-drug conjugate (ADC) applications. While these setups comprised antibody secretion with reporter cell droplet co-encapsulation, the enrichment rate could be further enhanced by the utilization of autocrine setups, where recombinant target expression is coupled with antibody candidate secretion in the same cell. This innovative configuration enabled robust high-throughput droplet-based microfluidic sorting for internalization.

To fully exploit the advantages of pre-sorting mammalian libraries for manufacturability in display mode and antibody secretion for microfluidics-assisted functional screens, we established and optimized an inducible non-covalent display to secretion switchable system for sufficient autocrine genotype-phenotype-coupled display. Differential surface display of published reference antibodies correlated with good versus poor manufacturability properties and enabled enrichment of desired antibodies from spiking mixtures via flow cytometry sorting. The successful establishment of both an inducible antibody display system amenable to manufacturability sorting and an internalization functional screen in secretion mode laid the foundation for an integrated workflow combining these two steps. Pre-sorting for manufacturability was followed by microfluidic enrichment for target-specific internalization and yielded a distinct enrichment of functional clones from spiking mixtures, exemplifying the capabilities of this integrated approach.

By leveraging the advantages of microfluidics, mammalian libraries, autocrine setups, and function first screens, the case studies for integrated approaches presented herein offer a promising strategy and basis for further development of methodologies for accelerated discovery of therapeutic antibodies with optimal desired functional properties.

Results

Autocrine screen for T cell engaging function

We previously described approaches of autocrine binding of a secreted antibody to a target of interest recombinantly expressed on the surface of the same cell,³⁷ as well as function first screens for T cell activation with a green fluorescent protein (GFP)-inducing reporter cell line.³⁸ Combining these two approaches, we designed a proof-of-concept study by spiking CHO cells expressing epidermal growth factor receptor (EGFR) and producing an anti-EGFR-SP34 BiTE molecule at a 1:100 ratio into CHO cells producing a non-targeted SP34 control molecule for co-encapsulation with Jurkat-GFP reporter cells. Anti-EGFR-SP34 BiTE trans-binding to EGFR on CHO and CD3 on reporter cells induced activation, resulting in a green peak signal (Figure S1a+b). Over 250,000 cells could be screened and specific BiTE-producing cells could be enriched to 29% (Figure S1d) as assessed by BiTE-specific analytical PCR analysis, principally confirming the applicability of such a setup for high throughput function first screenings.

Screen for in-trans conditional TNFRSF agonism

Similar to the aforementioned setup, a two-cell three-component recombinant assay system was established to allow a microfluidic proof-of-concept high throughput functional screening for targeted TNFRSF activation mediated by a bispecific antibody. For this, Jurkat cells were engineered to induce GFP expression upon stimulation of a TNFRSF member (termed herein TNFRSF). In addition, two CHO cell lines were established expressing recombinantly a cancer-related antigen of interest (CRA) in parallel to secretion of an anti-CRA-anti-TNFRSF bispecific antibody or a non-targeted anti-HEL isotype control bispecific antibody (Figure 1a). Pre-evaluation in bulk cell mixtures indicated conditional activation of TNFRSF reporter cells only in the presence of CRA-positive and anti-CRA-anti-TNFRSF secreting cells (Figure S2b). Targeted bispecifics spiked at 1:100 into non-targeted antibody secreting cells were sorted in a droplet-based microfluidics run for specific agonistic reporter signal and could be enriched to 17.7% (Figure 1b–c), essentially validating such an assay setup for function first screens.

Figure 1. Proof-of-concept microfluidic functional screening for in-trans TNFRSF agonism or autocrine target-specific internalization. (a) CHO cell lines (red) are pre-labeled and express either a cancer related antigen (CRA) in parallel to secretion of an anti-CRA-anti-TNFRSF bispecific antibody or a non-targeted anti-HEL isotype bispecific antibody (not illustrated). A TNFRSF member on engineered Jurkat cells can be activated in trans to produce a green signal via GFP that can be sorted for. (b) microfluidic sorting of a 1:100 mixed CRA-targeted to non-targeted red pre-labeled antibody secreting cells for green reporter signal. Note unspecific background signal below the sorting gate of droplets lacking secreting cells but containing leaky and/or multiple reporter cells. (c) Statistics for throughput, PCR recovery and hit rates. (d) Scheme of exemplary anti-B7-H3 reference antibody m276 binding to and internalizing with recombinant B7-H3 in autocrine fashion. pH-dependent detection antibody signal can be harnessed to sort for internalization in red peak versus average mode. (e) Cyto-Mine sort plot of 1:100 m276:anti-HEL secreting cell mixture for internalization signal and resulting hit rate. (f) High PCR recovery rates indicate robustness of CHO cells; high m276 hit rate indicates robustness of the autocrine assay setup.

Schematics and data for two microfluidic functional antibody screening approaches.

Autocrine functional screening for internalization

In light of the thriving development of cytotoxic and immune-stimulatory ADCs requiring strong internalization efficacy, broad function first screens for this property could speed up ADC hit discovery. To accommodate this need, an internalization assay based on pH-dependent secondary detection antibodies was developed (Figure 1d) and successfully applied in autocrine and co-encapsulation setup. Bulk testing of CHO cells secreting m276, a well-known internalizing anti-B7-H3 antibody that binds and internalizes autocrine to recombinantly co-expressed surface B7-H3, showed good fluorescence signals and assay window related to a non-targeted anti-HEL isotype control antibody (Figure S2a). Sorting of B7-H3 internalized antibody secreting cells from a 1:100 spiked cell mixture yielded a 100% m276 positive recovery rate. Of 191 dispensed droplets, 50 contained both m276 and anti-HEL antibody secreting cells, while the other 141 only showed m276-specific PCR amplicons, equaling a 74% monoclonal hit recovery rate (Figure 1e–f).

When applied in a co-encapsulation setup including tumor cells as presented below, enrichment was observed but was less pronounced, likely due to variability in secretion to target cell ratio (Figure 3d).

Enhanced secretion by cell synchronization

In comparison to microfluidic antibody hit discovery interrogating primary plasma cells that naturally secrete antibodies, cells of recombinant mammalian libraries undergo cell divisions leading to a certain share of non-secreting cells in short duration assay setups. Cell synchronization by double thymidine blockage almost doubled the percentage of cells secreting antibodies at early time point such as one hour (Figure S3) and was applied for microfluidic screening assays of 2–4 h duration such as sorting for cellular binding.

Display to secretion switch

Proof-of-concept studies with defined spiking mixtures reported herein indicate the possibilities for microfluidic high throughput functional screens, but library cloning and mammalian library generation in reality yield significant shares of non-full-length antibodies. A full display to secretion switch would thus be desirable to enable flow cytometry sorting to yield reduced library sizes of higher quality and fit to lower throughput microfluidic interrogation for cellular binding or function. To achieve this, an inducible fusion of duplicated Protein A ZZ domain with the PDGFR transmembrane domain (ZZ-PDGFR-TMD or ZZ-PDGFR) under a tetracycline/doxycycline promotor was stably integrated prior to application of CHO cells for proof-of-concept constructs or library generation. Two of the 12 evaluated monoclones were selected for further evaluation based on high display stability and retained viability over 5 weeks (Figure S5a). The kinetic optimum for induction and decay was determined initially with a saturating doxycycline concentration of 1 µg/ml and capturing of the stably integrated reference antibody, anti-EGFR cetuximab. Sufficient display could be observed after 24 h (Figure 2c) with minor further increase after 48 h (data not shown). In the absence of doxycycline, display levels of autocrine produced cetuximab declined over time and reached background levels after 6 to 7 days (Figure S5b). A high percentage of autocrine antibody display is desired to reduce background from paracrine antibodies present in a library cultivation supernatant. To achieve this, induction was further optimized to allow distinct display, but free ZZ-PDGFR capacities were minimized by reducing doxycycline concentrations during the fixed 24-h induction time. Induction with 0.05 µg/ml doxycycline yielded a balanced display and autocrine saturation rate (Figure S5c). However, as unsaturated ZZ-PDGFR remained to varying amounts for several reference antibodies with different secretion rates, Protein A was added to induction and screening media to block binding sites of paracrine antibodies in the supernatant. Addition of 173nM Protein A was found optimal to significantly reduce background by paracrine antibodies and retain display levels enabling flow cytometry sorting (Figure 2a–b).

Figure 2. Inducible antibody display with soluble Protein a addition retains genotype-phenotype correlation and differential display allows sorting for good manufacturability properties. (a) analytical flow cytometry analysis of 1F11 ZZ-PDGFR-TMD CHO cells displaying CS06 antibody (lambda) or bococizumab (kappa) in a 1:10 mixture after 24 h co-cultivation mimics a library situation. Considerable amounts of cells were positive for both lambda and kappa antibody display in Q2 due to paracrine background signal of bococizumab on CS06-displaying cells. (b) setup as in (a) with addition of soluble Protein a as decoy for paracrine antibodies restores exclusive display of autocrine produced antibody and generates a genotype-phenotype coupling sufficient for sorting. (c) differential induced CHO surface display levels of cetuximab (light blue) versus bococizumab (red) of approximately 3-fold as indicated by normalized MFI values. Uninduced non-displaying cells as reference are shown in orange. (d) Individual stains of differential display of CS06 antibody (blue) versus bococizumab (red) indicate the option to sort for manufacturability in a 1:100 mixture (orange) applying the indicated gate. Analytical lambda versus kappa flow cytometry analyses of this 1:100 mixture input (e) and sort output (f) indicate enrichment of CS06 in Q1 from approximately 1% to 10%. Events in Q3 represent bococizumab, while non-display-induced cells are observed in Q4.

Six graphs proving inducible differential antibody display allows sorting for manufacturability.

Differential antibody display enables manufacturability sorting

In addition to pre-enrichment of large libraries for target-specific antibodies, mammalian display systems can be utilized to screen for favorable antibody properties such as low aggregation propensity. The inducible system described here, albeit representing a non-covalent capture of autocrine antibodies, enables differential display as confirmed for proof-of-concept mixtures of anti-PCSK9 bococizumab with either cetuximab (Figure 2c) or anti-c-MET antibody CS06 (Figure 2d and Figure S6b) plus anti-CD3 clone SP34 (Figure S6c). When sorting a 1:100 mixture of CS06 and bococizumab for display (Figure 2d), CS06 could be enriched by approximately 10-fold as assessed by analytical anti-lambda versus -kappa flow cytometry (Figure 2f).

Integrated manufacturability sorting and function first screening

Successful establishment of both an inducible antibody display system amendable to manufacturability sorting (Figure 2) and an internalization functional screen in secretion mode (Figure 1) paved the way toward an integrated workflow combining these two steps. Two lambda antibodies, anti-c-MET clone CS06 and anti-CD3 clone SP34, were chosen due to confirmed high display, whereas bococizumab kappa antibody served as a poor manufacturability profile reference.^17,39 The clone choice enabled ease of evaluation of the two exemplary populations by analytical lambda versus kappa flow cytometry. Applied in 1:1:98 ratio of mixed CS06, SP34, and bococizumab cells and re-analyzed prior to sorting, 1.8% lambda clones could be confirmed in the input mixture (Figure 3a). Sorting solely for high display via anti-IgG-AF488 increased the share of lambda clones to 14.7%, equaling a one-step enrichment of approximately 8-fold (Figure 3b–c). After 7 d cell recovery and display decay, droplet-based microfluidic screening for c-MET-specific internalization on EBC-1 tumor cells increased the prevalence of CS06 to 25.8% (48 of 186 dispensed clones; analyzed by CS06-specific PCR amplicons; Figure 3d). CS06 enrichment in a two-step sorting workflow from 0.9% to 25.8% therefore was approximately 29-fold.

Figure 3. Integrated sorting for manufacturability and internalization via display to secretion switching. (a) a defined 1:1:98 mixture of 1F11 ZZ-PDGFR-TMD CHO cells secreting anti-c-MET lambda antibody CS06 (blue), anti-CD3 lambda SP34 (green) or anti-PCSK9 kappa antibody bococizumab (red) was incubated for 24 h prior to additional 24 h display induction to mimic potential cross-contamination of undesired paracrine library antibodies, here exemplarily bococizumab. (b) the induced clone mixture was sorted for high display via anti-IgG-AF488 and cells were recovered by cultivation without doxycycline induction for 7 d. (c) a split of sort output was then induced for analytical lambda versus kappa flow cytometry evaluation as indicator for enrichment of clones with good versus poor manufacturability properties. (d) Finally, the non-induced sort output was co-encapsulated with c-MET positive EBC-1 tumor cells and screened for c-MET-specific CS06 internalization using the Cyto-Mine microfluidic device. 6167 events were sorted and 192 dispensed. 48 of 186 analyzed single dispensed droplets yielded a CS06 specific PCR amplicon, equaling 25.8% hit rate. Enrichment of CS06 in this two-step screening approach was therefore approximately 30-fold.

Scheme and data showing integrated antibody sorting for manufacturability and internalization via display to secretion switching.

Discussion

The continued success of biotherapeutics’ clinical development has fueled technology advances in all fields of the pharmaceutical R&D value chain. Co-evolution of microfluidic antibody selection technologies and mammalian libraries for their optimization recently allowed combination of these promising approaches for added value hit discovery. Proof-of-concept studies presented herein were chosen to exemplify method diversity of mammalian library in microfluidic screen approaches and their utility for identification of novel biotherapeutics. While the commercially available droplet-based Cyto-Mine device was used, similar or even more flexible droplet-,³ nano-pen-,²⁹ or microcapillary³⁶ -based microfluidic systems would be suited to support similar or more assay setups and successful hit discovery campaigns.

Mammalian libraries, mainly generated by lentiviral delivery,⁸ nucleases,¹¹ recombinases,^14,15,17,18 and episomal approaches,²⁰ in contrast to primary plasma cell screening, enable preservation and repeated interrogation of full immune repertoires including both memory B and plasma cells of human individuals or immunized rodents. While large mammalian library generation can be laborious, library sizes of 1–100 million are feasible and usually capture most of an immune diversity^40,41 or the combinatorial space to be screened for optimal bispecific antibody identification. Unlike primary plasma cells, mammalian antibody secreting cells undergo cell division leading to temporary secretion reduction and therefore assay variability or false-negative screening results. Cell synchronization prior to microfluidic screens was shown to reduce this undesired effect (Figure S3) and accordingly applied to short runtime assays such as screening for cellular binding. Although effects level out after two hours when applying strongly secreting reference antibodies, non-pre-enriched diverse libraries could comprise cells secreting less antibody, and could therefore yield false-negative results in short assays. Second, non-droplet microfluidics allow consecutive short runtime assays that would benefit from low failure-rates due to consistent expression of synchronized cells. Generally high PCR recovery rates across experiments (Figures 1c–f, 3d, Figure S1d) indicate high viability and robustness of applied CHO cells and represent a considerable advantage over primary cells for complex longer runtime functional screening setups.

Functional screening of secreted antibodies for intracellular signaling modulation¹ or agonistic properties^3,38 recently highlighted the opportunities of microfluidic methodologies when combined with suitable, mostly fluorescence-based cellular reporter assays. We recently reported an autocrine screening setup utilizing CHO cells for recombinant target expression and final therapeutic format antibody secretion for robust specific cellular binding selection.³⁷ Addition of an appropriate reporter cell line to the screening system yielded a two-cell three-component setup amendable for targeted agonism screening (Figure S1, Figure 1d). Two variations were evaluated in proof-of-concept studies: (1) An EGFR-targeted SP34 scFv BiTE-secreting cell was spiked at 1:100 into cells producing non-targeted SP34 control antibody and screened in co-encapsulation with Jurkat-GFP reporter cells for trans-mediated CD3 agonism (Figure S1a). EGFR-targeted BiTE secreting cells could be enriched in a single screening cycle to 29% (Figure S1d). (2) CRA-expressing and anti-CRA x anti-TNFRSF IgG-VHH bispecific antibody-secreting cells were diluted 1:100 into cells producing non-targeted isotype bispecifics of the same therapeutic format (Figure 1d). Co-encapsulation with a Jurkat-GFP reporter cell recombinantly expressing TNFRSF enabled microfluidic sorting for in-trans TNFRSF activation. Approximately 250,000 cells could be sorted in one round and anti-CRA x anti-TNFRSF IgG-VHH bispecific antibody-secreting cells were enriched from 1% to 18% (Figure 1f). These examples represent, to our knowledge, the first report harnessing CHO cells in such manner for high throughput function first screens. While the setup shares similarities with previously reported mammalian autocrine functional selections,^4,18,42 the approach applied here includes screening in the final secreted format and in higher throughput than plate-based screens.⁴³ Although secreting and reporter cell lines were of mammalian origin, identifying applicable cultivation conditions needed assay optimization, highlighting the complexity of such screening setups. GFP reporter cell-based screens yielded enrichment factors comparable to similar recent reports,⁴² but background of non-antibody secreting cell (ASC)-containing droplets indicated at least partially leaky reporter systems (Figure 1e and Figure S1b) as a critical aspect of assay setup potentially leading to lower sorting efficiencies. Furthermore, the applied system is constrained to homogeneous fluorescence-based readouts and 488 nm laser excitation setups. Lastly, the dependency on Poisson-distributed co-encapsulation of secreting and target cells leads to strong assay variability and therefore represents an inherent drawback (see also Figure S7 in Gaa et al. ³⁸). Development of alternative or novel droplet-based microfluidic systems enabling 1:1 ASC to reporter cell ratios and reagent addition via droplet fusion would be desirable and potentially enhance success rates.

Internalization is an essential function of antibodies designed for ADC applications. Setting up a microfluidic workflow, the effectiveness in a two cell versus autocrine one cell setup was compared. Spiking experiments with an antiB7H3 reference internalizing antibody m276 in autocrine setup with surface co-expressed B7H3 target generated distinct assay signals and consequently yielded high enrichment results (Figure 1e, f). Application of the same detection system in a co-encapsulation setup for screening of CS06 antibody internalization into EBC-1 tumor cells, however, resulted in notable but comparatively lower enrichment rates (Figure 3d), likely due to Poisson-distribution related signal variability. The autocrine one cell per droplet setup fully exploited the high throughput of droplet-based microfluidics and target co-expression that could facilitate identification of hits against complex targets such as membrane spanning receptors, omitting the need for somewhat artificial recombinant target protein or membrane-like particles application for hit discovery. Note that despite the large assay window in comparison to a non-targeted control (Figure S2), differential enrichment of high versus medium internalizing antibodies remains to be confirmed. In addition, differences in recombinant target expression caused by, for example, lentiviral delivery random integration may add to read-out variability and lead to lower success rates.

Recent publications have indicated the applicability of mammalian display approaches^17,24 and antibody secretion setups for microfluidic hit discovery.^37,42 Extending the toolbox, simultaneous display and secretion^21,23 or an integrated switchable display to secretion system could utilize display for pre-enrichment of clones with high manufacturability (via fluorescence-activated cell sorting (FACS)²⁴ or full-length antibodies (via anti-LC-magnetic-activated cell sorting) leading to focused libraries suitable for subsequent microfluidic functional screening. To accommodate this missing link, a tetracycline/doxycycline-inducible non-covalent display system was developed and optimized to support library sorting for manufacturability. Display was achieved via autocrine antibody binding to a stably pre-integrated membrane-tethered dual Protein A Z-domain fusion protein, referred herein as ZZ-PDGFR-TMD. Single cell clones were screened for stable display (Figure S5a) and kinetics for display induction and decay were determined applying the reference antibody cetuximab (Figure S5b). Finally, induction strength was optimized to balance display to be sufficiently high for sorting, yet sufficiently low to achieve autocrine saturation (Figure S5c). As paracrine antibody background levels from cell mixture supernatant remained, Protein A addition acted as decoy similar to comparable reports.⁴⁴ Titration of Protein A to balance retained display for sorting but significantly reduce background yielded an optimum of 173 nM in the applied setup. A relatively broad low three-digit nanomolar Protein A concentration range was appropriate, but the potential need to optimize for other architectures and single cell clones established in other laboratories should be noted.

Differential display via ZZ-PDGFR-TMD of poor to well manufacturable reference antibodies was not as pronounced as via direct fusion (Figure 2c and Figure S6d-f compared to Figure 1 in Gaa et al.,³⁷ but sufficient for manufacturability sorting via droplet-based microfluidics (Figure 3b). In a proof-of-concept case study, well manufacturable lambda antibody SP34- and CS06-secreting cells were spiked into kappa antibody bococizumab-producing cells representing the majority of library candidates with undesired physicochemical properties and specificities. Sorting solely for high IgG display increased the share of lambda clones to 14.7%, with the eight-fold enrichment principally confirming this approach (Figure 3). In an integrated second microfluidic sorting step for c-MET-specific internalization on EBC-1 tumor cells, c-MET-specific reference antibody CS06 could be enriched to 25.8% (Figure 3d). Enrichment of a well manufacturable and target-specifically internalizing antibody in a two-step sorting workflow from 0.9% to 25.8% was approximately 29-fold. This indicated the applicability of such integrated workflows for fast and successful identification of optimal therapeutic antibody candidates.

While case studies presented herein were performed with defined spiking mixtures and single cell antibody gene PCR recovery for the sake of facile analytics, actual hit discovery campaigns would differ in workflows both upstream and downstream of flow cytometry and microfluidic sorting. Preceding library generation could be realized through subcloning of: (1) Phage display pre-enriched clones;¹⁷ (2) light-chain shuffled or natively-paired diversities enriched for specific binders via immunizations of humanized rodents; (3) single-domain antibody and common light-chain diversities omitting the need for light-chain pairing;³⁷ or (4) combinatorial libraries of known binder panels for bispecific antibody optimization.¹⁸ Downstream workflows could be further enhanced by pre- versus post-sort NGS analyses to identify suitable functional hits by output prevalence,¹ enrichment or high-capacity machine learning.^2,39,45

Whereas a full display shutdown was essential in droplet-based screening setups for cellular binding or internalization relying on fluorescence peak versus average signals, dual display/secretion architectures could serve other workflows such as for a Beacon device that accepts in-pen displayed antibodies while secreted antibody can be evaluated in the flow chamber above the respective nano-pens. While not fully applied in this study, a general setup harnessing F2A peptides with attenuated auto-cleavage (Figure S4b) enabled dual display and secretion (Figure S4d+e) and could be amendable in similar non-droplet-based microfluidic workflows.

The case studies reported here represent the first examples for an integrated methodology of two-step sorting for manufacturability followed by high throughput screening for functions such as target-specific tumor cell internalization. These or similar display setups and microfluidic technologies could increase the probability of identifying best possible therapeutic antibody hits with desirable complex mode of action profiles faster.

Materials and methods

Target and reporter cells

c-MET-expressing EBC-1 squamous cell carcinoma cells (Health Science Research Resources Bank, now National Institutes of Biomedical Innovation, Health and Nutrition, Japanese Cancer Research Bank JCRB0920 031496) were cultivated in MEM Eagle Medium (#M2279, Sigma Aldrich) including 2 mM Glutamine (#35050–061, Gibco ^TM GlutaMAX ^TM, Thermo Fisher Scientific) and 10% fetal bovine serum (FBS; #S0615, Sigma Aldrich). Jurkat-GFP reporter cell line (NF-κB/Jurkat/GFP™ Transcriptional Reporter Cell Line) was sourced from SBI System Bioscience, Palo Alto, California, USA and cultured in RPMI 1640 medium (Sigma-Aldrich) with 10% FBS heat inactivated and 1× Glutamax.

Generation of reference antibody secretion and target expressing cells

Reference antibody secreting cells were generated as previously described.³⁷ Respective publicly available reference antibodies sequences were synthesized with codon usage optimized for Cricetulus griseus (GeneArt, Regensburg, Germany) in a linked manner applying 2A peptides⁴⁶ in a genetic setup as follows: signal peptide; heavy chain; P2A; signal peptide; light chain. Additional target co-expression was realized by including the respective sequence in the second cistron, resulting in an exemplary setup: signal peptide; heavy chain; PTA; antigen; T2A; signal peptide; light chain. Sequences are listed in supplementary Table S1.

Cell synchronization

CHO cells were synchronized by double Thymidine block procedure. 2 mM Thymidine (Sigma-Aldrich) was added to cell culture medium for 16 h before removing Thymidine supplied media by centrifugation, supernatant removal, and resuspension for 8 h cultivation without Thymidine, followed by again addition of 2 mM Thymidine for the next 16 h overnight incubation in preparation of subsequent microfluidic or flow cytometry applications.

Generation of a pre-set inducible display cassette

The construct for an inducible ZZ-PDGFR-TMD cassette was designed as genetic fusion of ZZ domain linked by a GGGGS-linker to the PDGFR transmembrane domain as previously reported²⁰ and based on the commercially available pTetOne vector system (Takara Bio Europe/Clontech, Saint-Germain-en-Laye, France). The amino acid sequence is given in Table S1. Preceding generation of stable single cell clones subsequently used for antibody and target integration was based according to manufacturer’s recommendations. Briefly, transfected cells were induced for 24 h with 1 µg/ml doxycycline (Sigma-Aldrich) and stained with 5 µg/ml cetuximab-AF488-tagged antibody (Syngene, Bangalore, India) and sorted using BD FACS Aria III Cell Sorter. Five 96-well plates were seeded at 1 cell/well and imaged at regular time intervals in a Solentim Cell Metric device to ensure initially one cell per well and subsequent continuous monoclonal cell growth. Based on colony growth over 2 weeks, selected clones were analyzed by analytical flow cytometry and scaled up to 6-well plates and further to shake flasks. Twelve clones were subjected to stability studies over 5 weeks as shown in Figure S5.

Flow cytometry sorting

Sorting for high display on cells producing various reference antibodies in defined spiking mixtures was conducted applying a Sony SH800S flow cytometer. Cells were synchronized as described above, mixed and cultivated in presence of 173 nM Protein A before display was induced by addition of 0.1 µg/ml doxycycline (Sigma-Aldrich) for 24 h prior to staining with Alexa Fluor® 488 AffiniPure Fab Fragment Goat Anti-human IgG (H+L) (cat. 109-547-003, Jackson ImmunoResearch Europe LTD, Ely, United Kingdom) following manufacturer’s recommendations. Sorting was conducted applying a 100 µm nozzle in sort mode “purity” with gates adjusted to include approximately 2% events (see Figure S6 for full gating and Figure 3 for sort statistics).

Microfluidics-assisted ASC evaluation and sorting

Cyto-Mine-based sorting was conducted as previously described,³⁸ including ASC staining with 5 µM CellTracker™ Orange CMRA Dye (cat. C34551, ThermoFisher Scientific). During sorting of ASCs for cellular binding via co-encapsulation of target cells, a ratio of 1:6 million ASC to target cells was used to generate 2 million droplets. In autocrine cellular binding runs, 1 million target co-expressing ASCs were encapsulated in 2 million droplets to achieve a high degree of monoclonality. As all binders were recombinantly cloned to include a human IgG1 Fc portion, analysis of human IgG secretion or cellular binding was conducted with Goat Anti-Human IgG Fc-DyLight® 488 (cat. ab97003, Abcam, Cambridge, United Kingdom) detection antibody and Goat F(ab’)2 Anti-Human IgG - (Fab’)2 (DyLight® 594), pre-adsorbed (cat. ab98602, Abcam) or Alexa Fluor® 488 AffiniPure Fab Fragment Goat Anti-human IgG (H+L) (cat. 109-547-003, Jackson ImmunoResearch Europe LTD, Ely, United Kingdom). Maximum fluorescence signals were recorded applying “peak mode” setting.

Functional sorting for CD3 agonism setup was as previously reported.³⁸ EGFR surface-expressing as well as antibody secreting cells were stained as described above and co-encapsulated with Jurkat-GFP reporter cells in the absence of costimulatory signals. Similarly, Jurkat reporter cells recombinantly expressing an TNFRSF receptor of interest optimized for receptor densities to support cell mixture in-trans activation of GFP signal (kindly provided by Daniel Helman, Inter-Lab Ltd., an affiliate of Merck KGaA, Yavne, Israel; see Figure 1a) were mixed 6:1 with pre-labeled CHO target cells secreting targeted or non-targeted bispecific antibodies and incubated after droplet co-encapsulation at 37°C for 18 h to enable antibody secretion and GFP reporter signal development (see also Figure S2 for bulk culture kinetic assessment).

Sorting for internalization was realized by addition of 300 nM of pH-dependent detection antibody Incucyte® Human Fabfluor-pH Antibody Labeling Dye for Antibody Internalization (Sartorius, Göttingen, Germany) into the cell suspension volume applied to Cyto-Mine device for droplet generation.

Single cell mAb gene PCR recovery and confirmation

Positive droplets from Cyto-Mine were dispensed, treated for single cell antibody gene recovery PCR as previously described.³⁸ Briefly, single droplet dispensing in 96- or 384-well plates containing cell lysis buffer was followed by RT-PCR for cDNA generation and PCR analysis applying primers for analytical amplification of all applied V genes (to confirm presence of an antibody in the respective sample and calculate the overall antibody gene PCR recovery rate) as well as reference antibody construct specific primers to assess the prevalence or enrichment of specific functional antibodies in the sorting output panel.

Abbreviations

ADC	=	antibody-drug conjugate
AF	=	Alexa Fluor
ASC	=	antibody secreting cell
CHO cells	=	Chinese hamster ovary cells
CRA	=	a cancer-related antigen
DMEM	=	Dulbecco’s Modified Eagle’s Medium
Doxy	=	doxycycline
EGFR	=	epidermal growth factor receptor
FACS	=	fluorescence-activated cell sorting
Fab	=	antigen-binding fragment
FBS	=	fetal bovine serum
Fc	=	crystallizable fragment
GFP	=	green fluorescent protein
H+L	=	heavy and light chain
HEK	=	human embryonic kidney cells
HTS	=	high throughput screening
IMDM	=	Iscove’s Modified Dulbecco’s Medium
mAb	=	monoclonal antibody
MoA	=	mode of action
PCR	=	polymerase chain reaction
PCSK9	=	proprotein convertase subtilisin/kexin type 9
PDGFR-TMD	=	Transmembrane domain of platelet-derived growth factor receptor
Pen/Strep	=	penicillin-streptomycin
RT	=	reverse transcriptase
TNFRSF	=	a tumor necrosis factor receptor super family member
VH	=	heavy-light-chain variable domain
VK	=	kappa light-chain variable domain
VL	=	lambda light-chain variable domain
ZZ	=	duplicated Z domain of Protein A

Acknowledgments

We thank Dani Helman for provision of TNFRSF reporter cell lines and input on ZZ-PDGFR-TMD design. We are grateful for the support of technology development and case studies by Lars Toleikis and Andreas Menrad. Schemes in several figures were created using BioRender.com.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2251190

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.