Selection of novel affinity-matured human chondroitin sulfate proteoglycan 4 antibody fragments by yeast display

Abstract

Chondroitin sulfate proteoglycan 4 (CSPG4) is a promising target for cancer immunotherapy due to its high level of expression in a number of malignant tumors, and its essential role in tumor growth and progression. Clinical application of CSPG4-targeting immunotherapies is hampered by the lack of fully human high-affinity CSPG4 antibodies or antibody fragments. To overcome this limitation, we performed affinity maturation on a novel human CSPG4 single-chain Fv fragment (scFv) using the random mutagenesis approach and screened for improved variants from a yeast display library using a modified whole-cell panning method followed by fluorescence-activated cell sorting. After six rounds of panning and sorting, the top seven mutant scFvs were isolated and their binding affinities were characterized by flow cytometry and surface plasmon resonance. These highly specific, affinity-matured variants displayed nanomolar to picomolar binding affinities to the CSPG4 antigen. While each of the mutants harbored only two to six amino acid substitutions, they represented ~270–3000-fold improvement in affinity compared to the parental clone. Our study has generated affinity-matured scFvs for the development of antibody-based clinical therapeutics targeting CSPG4-expressing tumors.

Introduction

Human chondroitin sulfate proteoglycan 4 (CSPG4) is a type I transmembrane protein expressed as a 250 kDa glycoprotein or a 450 kDa proteoglycan. It is considered a promising therapeutic target due to its high level of expression in a number of malignant cancers, including melanoma, glioblastoma, triple-negative breast carcinoma, squamous cell carcinoma and malignant mesothelioma (Wang et al., 2010; Price et al., 2011; Rivera et al., 2012; Zhang et al., 2014). CSPG4 is essential for the growth and progression of tumors (Price et al., 2011). It functions as a scaf-fold protein and affixes signaling molecules to form a complex that facilitates sustained, high-level activation of key survival pathways. These pathways involve receptor tyrosine kinases signaling through the mitogen-activated protein kinase cascade and integrin signaling through focal adhesion kinase activation (Price et al., 2011). In addition, the expression of CSPG4 has been detected on stem cell populations in melanoma and glioblastoma, and it is believed to be associated with resistance to radiation and chemotherapies (Price et al., 2011; Svendsen et al., 2011; Zhang et al., 2014).

The clinical potential of CSPG4-targeting immunotherapies has been demonstrated by previous studies. In a phase I clinical trial for patients with neoplastic meningitis, I¹³¹-labeled mouse F(ab)₂ fragment was successfully used to treat a 46-year-old female. The patient exhibited no radiographic evidence of disease four years post-treatment (Cokgor et al., 2001). In addition, administration of the CSPG4-specific murine monoclonal antibody, 225.28, reduced lung metastasis in a xenograft model of breast cancer (Wang et al., 2010), and inhibited tumor growth and recurrence in melanoma xenograft models (Hafner et al., 2005). Antibody fragments directed towards CSPG4 have also been developed into other forms of cytotoxic agents. These include a mouse α-melanoma-associated CSPG Pseudomonas exotoxin A (MCSP-ETA’) immunotoxin (Schwenkert et al., 2008; Brehm et al., 2014), a cytolytic fusion protein, which replaces the toxin portion of the αMCSP-ETA’ immunotoxin with a mutated version of the microtubule-associated protein (MAP) tau (αCSPG4(scFv)-MAP) (Amoury et al., 2016), and a bi-specific mouse antibody (r28M) targeting CD28 and CSPG4 (Spiesberger et al., 2015). Multiple mouse monoclonal antibodies targeting CSPG4 (e.g. 225.28 S, TP41.2, 149.53) have also been successfully utilized for the construction of chimeric antigen receptors (CARs) (Beard et al., 2014). Despite the encouraging data these studies have generated so far, clinical application of CSPG4-targeting immunotherapies is hampered by the lack of high-affinity fully human CSPG4 antibodies or antibody fragments.

To overcome this limitation, we aimed to develop fully human single-chain variable region antibody fragments (scFvs) with high-binding affinity and specificity to CSPG4. In this article, we describe the affinity maturation of a human anti-CSPG4 scFv using the random mutagenesis approach. The mutant anti-CSPG4 scFv library was screened by yeast-display using a modified whole-cell panning procedure combined with fluorescence-activated cell sorting (FACS). The mutant clones were characterized by both flow cytometry and surface plasmon resonance (Biacore) for their binding affinities.

Materials and Methods

Yeast strains, cell lines and plasmids

The Saccharomyces cerevisiae EBY100 (ATCC, Manassas, VA) yeast strain was utilized for the surface display of the anti-CSPG4 scFvs. H350, a human melanoma cell line expressing CSPG4, was maintained in our laboratory. HEK293, a cell line derived from human embryonic kidney cells, which lacks expression of CSPG4, was obtained from ATCC. pYD1 plasmid (Thermo Scientific, Waltham, MA) was used for cloning the random mutagenesis scFv library to be displayed on the yeast surface. scFvs selected from the yeast display library were subcloned into a pET28a(+) vector (EMD Millipore, Billerica, MA) for expression of the scFvs as inclusion bodies in the Escherichia coli strain One Shot BL21 Star (DE3) (Thermo Scientific).

CSPG4-D2A antigen preparation

The extracellular domain of CSPG4 was divided into smaller subdomains to facilitate the preparation of the soluble CSPG4 antigen for the development of a fully human CSPG4 scFv (Price et al., 2011). The gene for subdomain D2A (CSPG4-D2A), corresponding to amino acids 641–1233, was synthesized and cloned into the EcoRI and NotI sites of pET43.1a(+) vector (EMD Millipore) by Genewiz (South Plainfield, NJ). The plasmid was transformed into E. Coli BL21 Star (DE3). Proteins accumulated in the inclusion bodies were solubilized and refolded using a previously described protocol (Kato et al., 2010). The refolded protein sample was dialyzed and loaded on to a Ni Sepharose Excel column (GE Healthcare Bio-Sciences, Pittsburgh, PA) and proteins were eluted with a gradient from 100 to 500 mM imidazole using an AKTAPurifier (GE Healthcare Bio-Sciences). The target protein was further purified by size exclusion chromatography on a TSKgel SuperSW3000 column (TOSOH Bioscience, Tokyo, Japan) to greater than 95% purity, as determined by SDS-PAGE.

Random mutagenesis by polymerase chain reaction

A random mutagenesis library was constructed by error-prone polymerase chain reaction (PCR) using a clone that we have previously isolated, which bound to CSPG4 with low affinity. The gene of this parental clone was PCR-amplified using PfuUltra high-fidelity DNA polymerase (Agilent, Santa Clara, CA) with pYD1-EcoRI-F-1H10 and pYD1-NotI-R-1H10 primers listed in Table I. The amplification product was digested with EcoRI and NotI (New England Biolabs, Ipswich, MA) and ligated into the pYD1 vector linearized by the same restriction enzymes, resulting in the pYD1–1H10 recombinant plasmid. Next, a DNA library of diverse, randomly mutated scFv sequences was generated by error-prone PCR using the Diversify PCR random mutagenesis kit (Clontech, Mountain View, CA) following the manufacturer's protocol. Polymerization was performed by the low-fidelity Taq polymerase with the primers pYD1-EcoRI-F-1H10 and pYD1-NotI-R-1H10. The product recovered from the error-prone PCR was further amplified using PfuUltra high-fidelity DNA polymerase with the same pair of primers. To verify the mutation rate and initial library diversity before yeast transformation, the amplified mutant DNA library was double digested by EcoRI and NotI, and ligated to the pYD1 vector. An aliquot of the ligation reaction was used to transform Max Efficiency DH5α chemical competent E. coli cells (Thermo Scientific). Twenty colonies were randomly picked up from the LB agar plates (5 g NaCl, 5 g Tryptone, 2.5 g yeast extract, 7.5 g agar in 1 l dH₂O containing 100 μg/ml ampicillin). Plasmids were extracted using the Qiagen plasmid miniprep kit (Qiagen, Germantown, MD) and were subsequently sequenced (Genewiz).

Table I.

Primers used in the study

Name	Nucleotide Sequence (5′−3′)
pYD1-EcoRI-F-1H10	5′-CAGGATCCAGTGTGGTGGAATTCGAGGTGCAGCTGGTGGAGTC
pYD1-NotI-R-1H10	5′-CTCTAGACTCGAGCGGCCGCCAACCTAAAACGGTGAGCTGGGTC
pET28a-EcoR-F-UC8	5′- CAGGATCCAGTGTGGTGGAATTCGAGGCGCAGCTGGTGGAGTC
pET28a-NotI-R-UC	5′-CTCTAGACTCGAGCGGCCGCAACCTAAAACGGTGAGCTGGGTC

Construction of the yeast display library

The pYD1 vector linearized by EcoRI and NotI double digestion was mixed with the random mutagenesis DNA library at a molar ratio of 1:5, and transformed into EBY100 yeast cells using the LiAc/SS carrier DNA/PEG (polyethylene glycol) method described by Gietz and Schiestl (2007). The transformed yeast cells were pooled into SDCAA media (20 g dextrose, 6.7 g Difco yeast nitrogen base without amino acid, 5 g bacto casamino acids, 5.4 g Na₂HPO₄, 8.56 g NaH₂PO₄-H₂O in 1 l dH₂O containing 50 μg/ml kanamycin and 30 μg/ml tetracycline). The library was cultured at 30°C and 250 rotations per minute (r.p.m.) from a starting optical density of 0.2 at 600 nm (OD₆₀₀ = 0.2) for 48–72 h to amplify yeast clones successfully transformed with plasmids containing the library sequences. To estimate the initial library size, an aliquot of the SDCAA culture (OD₆₀₀ = 0.2) was serially diluted and plated onto an SDCAA agar plate (1 l SDCAA media plus 182 g d-sorbitol and 15 g agar). The over-grown library (round 0 library) was stored in 15% glycerol at −80°C, with at least 10-fold more cells than the library size calculated from colony counting.

Selection by whole-cell panning

A frozen aliquot of the round 0 library was inoculated to 0.2 at OD₆₀₀ in the SDCAA medium. The amount of yeast cells inoculated was 10- to 100-fold more than the library size. The culture was allowed to grow overnight and passaged again to ensure the complete elimination of dead cells. The next day, the SDCAA culture was diluted in SGCAA media (galactose instead of dextrose in SDCAA media) to 0.5 at OD₆₀₀ for an induction of the scFv display on yeast cells. The yeast cells were incubated at 18°C and 180 rpm for 48–72 h for full display of the scFv on the yeasts. A fraction of the culture was then pelleted and incubated with a 10 nM soluble, purified CSPG4-D2A protein at room temperature (RT) for 1 h, and then with a mouse anti-Nus tag IgG1 antibody for another hour at 4°C. After three rounds of washes, a mouse anti-V5 tag antibody conjugated to allophycocyanin (APC) (Abd Serotec, Raleigh, NC), which detected the V5 tag at the 3′ end of the scFv displayed on the yeast surface, and a goat anti-mouse IgG1 cross-adsorbed secondary antibody conjugated to R-phycoerythrin (PE) (Pierce, Waltham, MA), which detected the anti-Nus tag antibody on the yeasts that bound to the CSPG4-D2A protein, were added to the yeast cells and incubated for 1 h at 4°C. FACSCalibur (GE Healthcare Bio-Sciences) was used to analyze the fluorescent PE and APC signals from the labeled yeasts to confirm the expression of scFvs on the yeast surface and binding of the library clones to the CSPG4-D2A antigen.

After induction in SGCAA media, the round 0 library was used to perform whole-cell panning. H350 and HEK293 cells were seeded in tissue culture flasks with a growth area of 175 cm² and cultured overnight. The next day, the cells were fixed with 0.25% glutaraldehyde (Sigma, St. Louis, MO). An aliquot of the induced round 0 library covering 40 times of the library size was pelleted and resuspended in PBSF buffer (8 g NaCl, 0.2 g KCl, 1.44 g Na₂HPO₄, 0.24 g KH₂PO₄, 1 g bovine serum albumin in 1 l dH₂O, pH 7.4). The resuspended yeast cells were added to the HEK293 cells fixed on the flask and incubated at RT for 1 h with gentle shaking. The supernatant was removed and added to the fixed H350 cells for a 2 h incubation at RT with gentle shaking. After discarding the supernatant, the cell layer was washed five times with phosphate-buffered saline (PBS; 1×PBS, pH 7.4) by gently rocking the flask 20 times for each wash cycle. The H350 cells and the yeasts bound to them were harvested using a cell scraper. The mixture was resuspended in SDCAA media and incubated at 30°C, 250 rpm until OD₆₀₀ reached around four. Then the culture was passaged again before it was diluted in SGCAA media for induction at 18°C, 180 rpm for 48–72 h. The process was repeated for the second and third rounds of biopanning, except at the end of the third round, in which the wash step with 1×PBS was extended to an overnight wash at 4°C on a shaker. After each round, a fraction of the library after induction was pooled, incubated with 10 nM CSPG4-D2A protein and stained with the antibodies described earlier for detection of the scFv expression and binding to the antigen.

Selection by FACS

The yeast library recovered from the third round of the whole-cell panning (round 3 library) was incubated with the soluble, purified CSPG4-D2A protein and anti-Nus tag IgG1 antibody, and then labeled with the anti-V5 antibody and anti-mouse IgG1 antibody as described above. Yeast clones exhibiting higher signals from both APC and PE channels (top 0.8% of the population for the first round of sorting, 0.6% for the second round of sorting, 0.2% for the third round of sorting) were sorted on FACSCalibur using a published protocol (Chao et al., 2006). The sorted yeasts were collected and pooled down to SDCAA media and incubated at 30°C, 250 rpm. The culture was passaged once before it was used for the next round of FACS screening. FACS screening progressed with decreasing concentrations of CSPG4-D2A proteins (from 2.5 to 0.5 nM). The final screened library (round 6 library) was plated on SDCAA plates, and individual colonies were randomly selected for DNA sequencing. Yeast clones with unique DNA sequences were identified and their binding to CSPG4-D2A proteins was validated by flow cytometry.

Expression, refolding and purification of clones

The genes of selected mutant scFvs with unique DNA sequences were PCR-amplified from the pYD1 vector using the forward primer pYD1-EcoR1-F-1H10 and reverse primer pET28a-NotI-R-UC (Table I). In the case of the mutant clone UC8, forward primer pET28a-EcoR-F-UC8 was used instead of pYD1-EcoR1-F-1H10. The mutant scFvs were sub-cloned into a pET28a(+) vector, and plasmids were transformed into BL21 Star (DE3) E. Coli. The scFvs were expressed as inclusion bodies, solubilized, refolded and purified using the method described in ‘CSPG4-D2A antigen preparation’ section. The purity of the scFvs was determined by SDS-PAGE analysis.

Affinity measurement by FACS

Purified scFvs were directly conjugated to Alexa Fluor 488 (AF488) using the AF488 TFP ester (Thermo Fischer Scientific), and excess dye was removed using MagneHis magnetic beads (Promega, Madison, WI). Of note, 3 × 10⁵ cells (H350 or HEK293) were resuspended in 500 μl of 1×PBS containing 5% fetal bovine serum (FBS) (5% FBS–PBS). Serially diluted antibodies were incubated with the cells and analyzed on FACSCalibur. Mean fluorescent intensity values were obtained from the histograms and were used to plot the binding curves. A standard approach of nonlinear regression using the one-site-binding hyperbola available in Graphpad Prism 5 (Graphpad Software, La Jolla, CA) was used to fit the curves, and c values were calculated by the program. The binding specificity of the scFvs were determined by incubating the AF488-labeled scFvs with HEK293 cells.

Affinity measurement by biacore

Binding kinetics and affinity of purified scFvs against soluble CSPG4-D2A antigen was determined by surface plasmon resonance using Biacore 3000 (Biacore, Pittsburgh, PA). Each antibody was assayed at three different concentrations (75, 150 and 300 nM) against immobilized CSPG4-D2A protein at a flow rate of 30 μl/min in binding buffer (10 mM HEPES, 0.15 M NaCl, 3.4 mM EDTA and 0.005% p20 surfactant at pH 7.4). The curves were fitted using the 1:1 Langmuir-binding model, and rate constants were calculated with BIAevaluation Software (Biacore).

Results

Construction of the anti-CSPG4 human scFv random mutagenesis library

The parental clone 1H10 is a fully human anti-CSPG4 scFv previously isolated in our lab from a naive phage display library against the sub-domain D2A, corresponding to amino acids 641–1233 of the human CSPG4 protein (CSPG4-D2A). Due to its relatively weak binding to the antigen (K_D = 3.83 × 10⁻⁶ M; Table II), we performed random mutagenesis to improve its affinity. Mutations were introduced randomly throughout the full length of the scFv using low-fidelity Taq polymerase. We restricted the complexity of the mutagenesis library by performing error-prone polymerase chain reaction (PCR) at a targeted mutation rate of approximately three to five nucleotide changes (or two to four amino acid changes) per scFv. To verify the quality of the initial random mutagenesis, the amplified mutant library was transformed into E. Coli and 20 clones were randomly picked up for sequencing (data not shown). All clones displayed unique sequences different from 1H10. The average number of nucleotide mutations and non-synonymous amino acid substitutions per scFv was determined to be 3.55 and 3.18, respectively. Both numbers were within the targeted range. The ideal ratio of A/T bases converted to G/C bases and vice versa is 1.0, and in our experiment, the conversion ratio had a value of 1.33. This confirmed that the diversity of the random mutagenesis DNA library was well within the desired range.

Table II.

Affinity of selected mutant scFv clones from round 6 random mutagenesis yeast display library

Clone	Amino acid change(s)a				Inc.f	Affinity by flow cytometry KD (M)	Affinity by biacore
	Num.b	Reg.c	Pos.d	Mut.e			KD (M)	kassoc (1/Ms)	kdissoc (1/s)
1H10	–	–	–	–	0	(3.83 ± 0.46) × 10−6	(2.57 ± 0.43) × 10−6	2.02 × 104	5.20 × 10−2
UC3	2	VH FR4	113	L-R	3	(8.31 ± 0.56) × 10−9	(3.45 ± 0.07) × 10−9	3.28 × 106	1.13 × 10−2
		VL CDR3	239	Y-H
UC4	6	VH CDR2	57	E-V	1	(3.70 ± 0.45) × 10−9	(2.51 ± 0.24) × 10−9	4.58 × 105	1.15 × 10−3
		VH FR3	77	S-G
		VH FR3	78	T-A
		Linker	131	S-P
		VL CDR1	166	S-R
		VL FR3	226	D-V
UC5	5	VH FR1	25	S-F	3	(8.18 ± 2.06) × 10−9	(4.56 ± 1.00) × 10−9	7.08 × 105	3.23 × 10−3
		VH FR3	77	S-G
		Linker	132	G-S
		Linker	139	T-A
		VL CDR3	239	Y-H
UC8	3	VH FR1	2	V-A	1	(1.17 ± 0.35) × 10−8	(5.64 ± 0.73) × 10−9	5.18 × 105	2.92 × 10−3
		VH FR4	113	L-P
		VL CDR3	239	Y-H
UC10	2	VH FR3	59	T-S	3	(7.15 ± 1.44) × 10−9	(9.31 ± 0.67) × 10−9	3.22 × 105	3.00 × 10−3
		VL CDR3	239	Y-H
UC12	3	VH CDR2	57	E-V	3	(3.37 ± 1.55) × 10−9	(7.25 ± 0.15) × 10−10	4.59 × 105	3.33 × 10−4
		VL FR3	201	N-K
		VL CDR3	239	Y-H
UC16	3	VH FR2	39	Q-L	2	(7.97 ± 0.34) × 10−9	(3.72 ± 0.80) × 10−9	1.46 × 106	5.43 × 10−3
		VH FR3	79	A-V
		VL CDR3	239	Y-H

^aNon-synonymous amino acid mutations compared to the parental 1H10 scFv.

^bTotal number of non-synonymous amino acid mutations in each clone.

^cRegion where mutation was found.

^dPosition of the mutation.

^eSpecific amino acid mutation at each position, presented in the format of ‘before-after’. Single-letter amino acid codes were used.

^fIncidence among 28 mutant clones randomly picked up and sequenced from round 6 library, all of which had fluorescent signal stronger than the parental clone 1H10.

Selection of clones with improved affinity for CSPG4 by whole-cell panning and FACS

After transformation of the mutant DNA library into the yeast, the initial yeast library size was determined to be 3 × 10⁷. All further passages of the yeast culture were performed by 10–100-fold of oversampling of the yeast library size in order to reduce the probability of losing unique clones. Three rounds of whole-cell panning were performed against the human melanoma cell line H350 expressing the target antigen, CSPG4, followed by three rounds of FACS. Results before and after each selection round were monitored by labeling an aliquot of the yeast library and analyzing its binding to 10 nM of the soluble, purified CSPG4-D2A protein by flow cytometry. As screening progressed, the population of scFv clones exhibiting higher phycoerythrin (PE) fluorescence signals (which demonstrate the affinity to CSPG4-D2A protein) increased gradually, as represented by the percentage of yeast clones within the gated region (Fig. 1). Whole-cell panning generated a relatively minor increase in the percentage of high-affinity clones, from 0.59% for the round 0 library (yeast library prior to panning, Fig. 1A), to 2.46% for the round 2 library (yeast library after round 2 whole-cell panning, Fig. 1C). Even with an extensive overnight wash performed during round 3 panning, high-affinity clones only accounted for 8.17% of the total population after the third round (Fig. 1D). However, after the first round of FACS (round 4 library) using 2.5 nM of CSPG4-D2A antigen, the high-affinity clones were significantly enriched from 8.17 to 26.30% (Fig. 1E), demonstrating the efficiency of single-cell sorting to isolate the top binders of very low abundance (<1%) in the population. For the next two rounds of FACS, an increased stringency of selection was applied by using 1 and 0.5 nM of antigen, respectively. Substantial enrichment of high-affinity clones occurred after the second round of sorting, after which the percentage of high-affinity clones increased to 77.12% (Fig. 1F). Sorting was stopped after the third round, with an overall enrichment of 86.15% of high-affinity binders in the final library (Fig. 1G). An aliquot of the round 6 library was then plated and 28 yeast clones were randomly selected for sequencing. We identified 14 unique sequences from these clones, each containing one to six non-synonymous amino acid substitutions (Fig. 2). Yeasts displaying the single clones of each of these 14 variants all showed improved binding to the CSPG4-D2A protein by flow cytometry, compared to the parental clone 1H10 (Supplementary Fig. S1).

Fig. 1.

Bivariate dot-plot of random mutagenesis library screened by whole-cell panning and FACS. Total of 10 000 yeasts were recorded by FACS Calibur for each analysis plot. About 10 nM of soluble, purified CSPG4-D2A protein was used as the antigen. The diagonal window represents the population of yeasts displaying scFvs from the library and showing improved binding affinity to the CSPG4-D2A antigen. Numbers above the diagonal window indicate the percentage of yeasts within the gated region, as compared to the whole population analyzed. (A) Round 0: initial yeast library before whole-cell panning. (B–D) Yeast library after first (B), second (C) and third (D) round of whole-cell panning. (E–G) Yeast library after the first (E), second (F) and third (G) round of FACS.

Fig. 2.

Amino acid sequence alignment of 14 affinity-matured anti-CSPG4 scFv variants with parental clone 1H10. Framework regions (FR) and complementarity-determining regions (CDR) are indicated on the top. The different residues are presented as one letter amino acid code and the same residues are presented as dots (.). The two most common mutations (Y239H in the V_L CDR3 region, and E57V in the V_H CDR2 region) are highlighted.

Binding affinity and specificity analysis of the CSPG4 scFv variants by flow cytometry

We selected the top seven mutant scFv yeast clones (UC3, UC4, UC5, UC8, UC10, UC12 and UC16) based on their mean PE intensity values for further characterization (Supplementary Fig. S1). The parental and mutant scFvs were expressed in E. Coli as inclusion bodies, reduced, refolded, and purified using Ni Sepharose Excel affinity chromatography and size exclusion chromatography to at least 90% purity by SDS-PAGE (Supplementary Fig. S2).

To demonstrate the specificity of the mutant scFvs binding to CSPG4, flow cytometry analysis was performed using 350 nM of the mutant scFvs directly labeled with AF488 against H350 (CSPG4 antigen-positive) or HEK293 (CSPG4 antigen-negative) cells. A significant increase in the AF488 mean fluorescence intensity was observed against H350 cells expressing high levels of CSPG4. No binding was seen against the antigen-negative HEK293 cells, demonstrating that the binding of the mutant CSPG4 scFvs was highly target-specific (Supplementary Fig. S3).

A summary of the position and the amino acid substitutions for each of the seven selected mutant clones, the number of incidences of each variant among the 28 clones randomly picked from the round 6 yeast display library, and the average affinity values (K_D) measured by flow cytometry is presented in Table II. The sequencing results showed that each of the top seven mutant scFvs incorporated two to six non-synonymous amino acid substitutions. In addition, the binding affinities of all seven mutant clones were significantly improved by ~300–1000-fold compared to the parental clone 1H10, which had a K_D value of 3.83 × 10⁻⁶ M by flow cytometry analysis. Except for UC8, which had a binding affinity of 1.17 × 10⁻⁸ M, all other mutant clones had K_D values in the nanomolar range. Clone UC12 displayed the highest affinity, with a K_D value of 3.37 × 10⁻⁹ M. A representative plot of affinity analysis and the average K_D values are presented for the parental 1H10 clone and individual mutant scFv in Fig. 3.

Fig. 3.

Apparent affinity analysis of soluble, purified 1H10 and mutant anti-CSPG4 scFvs by flow cytometry. A representative plot is presented for each scFv, and the average K_D values derived from at least three independent experiments are indicated on the graphs.

Affinity analysis of the CSPG4 scFv variants by biacore

The binding kinetics of the parental and mutant scFvs were also determined by Biacore. The purified scFvs were applied to sensor chips coated with the purified CSPG4-D2A antigen. The K_D values for individual scFvs are summarized in Table II, and a representative sensorgram for each clone is presented in Fig. 4. Biacore gave similar values of K_D compared to the results from flow cytometry affinity analysis. The parental clone 1H10 had a weak-binding affinity of 2.57 × 10⁻⁶ M, as determined by Biacore, due to its slow association rate and fast dissociation rate. Significantly, the mutant clones exhibited 270–3000-fold of improvement in K_D values compared to the parental clone 1H10 by Biacore. All seven mutant clones had K_D values in the nanomolar range except for the clone with the highest affinity, UC12, which had a K_D of 7.25 × 10⁻¹⁰ M. Due to the high response units (RU) observed in the sensorgrams of UC8 and UC12, we repeated the analysis of these two clones by decreasing the immobilization level of the surface antigen and increasing the concentration titration of the scFvs from three concentrations to nine concentrations ranging from 300 to 1.17 nM (Supplementary Fig. S4). Under these experimental conditions, the affinity values of UC8 and UC12 remained similar to the K_D values obtained at high Rmax (RU ≫ 100); K_D = 5.67 × 10⁻¹⁰ M for UC12 (compared to 7.25 × 10⁻¹⁰ M at RU ≫ 100), and a K_D = 9.02 × 10⁻⁹ M for UC8 (compared to 5.64 × 10⁻⁹ M at RU≫100).

Fig. 4.

Affinity analysis of soluble, purified 1H10 and mutant anti-CSPG4 scFvs by Biacore. Purified CSPG4-D2A antigen was immobilized on the chip, while anti-CSPG4 scFvs were serially diluted in the solution. A representative sensorgram was shown for each scFv, and the average kinetic values from at least two individual experiments were indicated on the graphs. The three lines in the sensorgrams correspond to 300, 150, and 75 nM of scFvs.

Discussion

Due to its high-expression levels on the cell surface and its essential role in growth and progression of tumors, CSPG4 is a promising therapeutic target for a variety of cancers. The clinical potential of antibody-based therapies targeting CSPG4 in cancer treatment has been demonstrated by various studies. One of the most critical requirements for an effective CSPG4-targeted immunotherapy is a good antibody or antibody fragment (e.g. an scFv) with high binding affinity and specificity. These antibodies or antibody fragments can then be combined with peptides, proteins, or drugs to be developed into various formats, including bispecific antibodies, fusion antibodies and antibody-drug conjugates. They can also be utilized to direct CARs to cancer cells expressing the antigen. In these scenarios, the use of fully human antibodies might be preferable over murine antibodies, since the immunogenic responses to the murine antibodies could potentially impact both safety and pharmacokinetic properties, resulting in low utility and efficacy of the drugs. Unfortunately, most CSPG4-targeted immunotherapies currently available involve the use of murine antibodies and antibody fragments. There is a lack of fully human high affinity CSPG4 antibodies. As a result, we aimed to develop fully human scFvs with high affinity and specificity against CSPG4 using the random mutagenesis approach for affinity maturation.

In the current study, we performed affinity maturation on a parental anti-CSPG4 scFv using the random mutagenesis approach and screened the yeast display library using a modified whole-cell panning method combined with FACS. Seven unique scFvs harboring two to six amino acid substitutions were isolated and characterized by flow cytometry to determine their binding specificity to CSPG4 expressed on H350 cells. We further determined the affinity of these mutant scFvs by flow cytometry and surface plasmon resonance. The affinity-matured scFv variants were highly specific and had nanomolar to picomolar binding affinities to the CSPG4 antigen. The mutant scFvs exhibited ~270–3000-fold affinity improvement over the parental clone.

Random mutagenesis has become an increasingly important tool in optimizing protein properties. A critical aspect of a random mutagenesis library is its mutational frequency. The optimal mutation rate is a tradeoff between accumulating enough beneficial mutations and avoiding too many deleterious mutations (Zhang and Shakhnovich, 2010). While hypermutated libraries might give rise to clones exhibiting greatest affinity improvements (Daugherty et al., 2000), since our yeast display library only had a small size of 3×10⁷, we targeted a mutation rate of 0.4–0.7% (three to five DNA mutations per gene) during random mutagenesis. Indeed, the final seven unique clones with improved binding affinities all harbored two to six DNA mutations per 750 kb gene, corresponding to an actual mutational frequency of 0.26–0.8%. The highest affinity clone, UC12, had four nucleotide mutations (a mutational rate of 0.53%), which resulted in three non-synonymous amino acid substitutions and more than 1000-fold improvement of K_D. These results demonstrate that significant affinity improvement can be achieved by targeting at this mutational frequency range.

Both FACS and whole-cell panning have been successfully utilized to screen yeast-display libraries (Chao et al., 2006; Tillotson et al., 2013). Flow cytometry allows for a quantitative assessment of antigen binding and scFv expression on the yeast cell. In situations where soluble antigens are not readily available, yeast-display scFvs can be panned against monolayers of mammalian cells, or incubated with mammalian cells in solution and separated by density centrifugation (Tillotson et al., 2013). Previous attempts have been made to combine both whole-cell panning and soluble target-based screening to ensure that the isolated scFvs are genuine binders for the intact cellular target (Dangaj et al., 2013). In this study, we utilized a modified cell panning method and combined it with FACS. The cell panning performed in our study is different from previous studies in several aspects. First, we started whole-cell panning before cell sorting. Due to the large size of the CSPG4 protein, which resulted in difficulty of expression, we only used a truncated part of the extracellular domain of CSPG4, namely CSPG4-D2A, as the antigen. The CSPG4-D2A antigen was expressed and purified from bacterial inclusion bodies. Since the soluble CSPG4-D2A antigen prepared in bacteria lacked glycosylation and could have had a conformation different from the membrane-bound CSPG4 protein, we proceeded with cell panning first on antigen-negative HEK293 cells, followed by panning on CSPG4-positive H350 cells. This helped to ensure that the library contained mostly true binders to the native CSPG4. Second, we utilized cell culture flasks with 175 cm² growth area, which allowed for a greater volume input of the library instead of using a six-well plates described in the original whole-cell panning protocol (Tillotson et al., 2013). Third, prior to adding the yeast library, we fixed the mammalian cells on the flask using glutaraldehyde. This ensured that the native CSPG4 would not degrade during the incubation and washing steps. It also facilitated the use of more vigorous washing procedures, while reducing the possibility of losing clones that bound the antigen-expressing cells that were detached during the washing process. Fourth, during the round 3 panning, instead of gently washing the plates by rotating them to eliminate the unbound yeasts, we applied a more stringent method which included an overnight wash at 4°C on a shaker. In our case, an overnight wash only modestly improved the enrichment of high-affinity clones from 2.5 to 8.2% (Fig. 1). It is possible that the avidity effect might have played a role in cell panning, potentially leading to modest enrichment of high-affinity clones. Therefore, to overcome the avidity issues, we continued the affinity maturation with FACS after three rounds of cell panning. Unlike cell panning, one advantage of using FACS for screening yeast display libraries is that the variations in the expression levels of scFvs on individual yeasts can be normalized using fluorescent signals from antibodies binding to the epitope tag (e.g., the V5 tag), thus eliminating artifacts due to host expression bias and allowing for a fine discrimination between the mutants. In addition, the sorting stringency of FACS can be easily manipulated by lowering the concentration of the soluble antigen and using a smaller sorting gate. Indeed, three rounds of sorting significantly enriched the percentage of high-affinity clones from 8.17 to 86.15% (Fig. 1). These results demonstrated that the yeast clones enriched from our newly developed screening process were able to bind to both the intact CSPG4 antigen on the cell surface and the truncated, soluble CSPG4-D2A antigen prepared from bacteria.

Compared to the binding kinetics of the parental clone, the enhanced affinities of the seven mutant clones (Table II) resulted from significant improvements in both the association rate constants (faster on-rate) as well as the dissociation rate constants (slower off-rate) (Supplementary Table S1). For the mutant clones UC3, UC5, UC8 and UC16, the fold improvement of the association constant was higher than that of the dissociation constant, suggesting that the increased association constant had a greater contribution to the enhanced affinity (Supplementary Table S1). On the other hand, for UC4, UC10 and UC12, the fold improvement of the dissociation constant was greater than that of the association constant, demonstrating that the improved dissociation constant had a greater contribution to the increased affinities of these three mutant clones. Especially in the case of UC12, the best binder among the mutant clones, there was a 156-fold improvement in its dissociation constant, but only a 23-fold improvement in its association constant.

Analysis of the sequences of the 14 unique mutant anti-CSPG4 scFvs isolated from the round 6 yeast library identified the top two predominant mutations among the 14 unique mutant clones to be: (i) tyrosine to histidine at position 239 in the V_L CDR3 region (Y239H) in 12/14 clones (at an occurrence rate of 86%) and (ii) glutamic acid to valine at position 57 in the V_H CDR2 region (E57V) in 4/14 clones (at an occurrence rate of 29%) (Fig. 5). Interestingly, none of the 20 clones picked from the initial random mutagenesis DNA library prior to cell panning contained the Y239H mutation, indicating that this specific mutation was initially of very low abundance in the mutant library, but was subsequently enriched during the screening process. This suggested that the Y239H mutation might have played a significant role in the improved binding affinity of the mutant clones. In addition, the mutant clone UC1 only had the single amino acid mutation, Y239H (Fig. 2) and displayed apparently improved binding to the soluble CSPG4-D2A protein compared to the parental clone, 1H10 (Supplementary Fig. S1), confirming the importance of the Y239H substitution in the affinity maturation of 1H10 scFv. UC4 and UC9 were the only two mutant scFvs that did not harbor the Y239H mutation among the 14 mutant clones. Instead, they shared the second-most frequent mutation, E57V in the V_H CDR2 region. Unfortunately, among the 14 mutant clones, we did not find an scFv with the E57V mutation alone, thus further study is needed to confirm the contribution of this specific mutation in affinity maturation. Compared to UC12, UC4 had a similar K_D by flow cytometry analysis, but had a 3.4-fold weaker binding to the CSPG4-D2A protein by Biacore. These results suggest that while the Y239H mutation was important for the increased affinity of mutant anti-CSPG4 scFvs, other mutations or combinations of mutations could also enhance the affinity of the parental scFv.

Fig. 5.

Sequence analysis of the 14 mutant clones isolated from round 6 yeast display library. Left Y-axis and colored bars: counts of a specific substitution that occurred within the 14 clones. Right Y-axis and line: percentage of the clones with a specific substitution, among the 14 clones. Amino acid substitutions and their positions were indicated. Single-letter code for amino acids was used. Regions of the mutations were indicated by the colors of the bars and the descriptions of the corresponding colors beneath the graph. Percentages of the most abundant mutations were labeled on the graph.

Wang et al. have described a fully human anti-CSPG4 antibody (scFv-FcC21) with its affinity characterized by a kinetic-binding assay using flow cytometry (K_D = 5 × 10⁻⁸ M) (Wang et al., 2011). While the UC12 antibody fragment reported in our study is monovalent, it exhibited increased affinity (K_D = 3.37 × 10⁻⁹ M measured by flow cytometry, and K_D = 7.25 × 10⁻¹⁰ M measured by Biacore) compared to the bivalent antibody scFv-FcC21. In this regard, the mutant anti-CSPG4 scFvs described herein are novel antibody fragments with greater affinity against CSPG4-expressing tumors.

In conclusion, we have established fully human, high-affinity anti-CSPG4 scFvs through effective utilization of random mutagenesis and yeast-display selection. These affinity-matured anti-CSPG4 scFvs have the potential to be developed into diagnostic agents or targeted therapeutics for the clinical evaluation and treatment of CSPG4-expressing tumors.

Supplementary Data

Supplementary data are available at Protein Engineering, Design & Selection online.

Conflict of interest

None declarerd.

Funding

This work was supported by the following grants from the National Institutes of Health (NIH) of the United States: P01-5P01CA154291 (DDB) and the National Cancer Institute (NCI)—1R35CA197264 (DDB).