Tailoring subtractive cell biopanning to identify diffuse gastric adenocarcinoma‐associated antigens via human scFv antibodies

Summary

Among various solid tumours, gastric cancer (GC) is one of the leading causes of cancer‐related deaths worldwide. Expansion into the peritoneal cavity, which results from dissemination of diffuse cancer cells, is the main cause of mortality in gastric adenocarcinoma patients. Therefore, investigation of putative biomarkers involved in metastasis is prerequisite for GC management. In an effort to discover potential tumour markers associated with peritoneal metastasis of GC, a semi‐synthetic human scFv library (Tomlinson I) was used to isolate novel antibody fragments recognizing MKN‐45, a poorly differentiated diffuse gastric adenocarcinoma cell line. Four rounds of subtractive selection each consisting of extensive pre‐absorption of phage library with NIH‐3T3 murine embryonic fibroblasts and AGS (a well‐differentiated intestinal gastric adenocarcinoma) cell line were carried out prior to positive selection on MKN‐45 target cells. ELISA‐based screening of 192 phage‐displayed scFv clones indicated 21 high‐affinity binders with specific staining of MKN‐45 compared with AGS cells. Diversity analysis of the selected phage‐scFvs resulted in five distinct sequences with multiple frequency. Further analysis by ELISA and flow cytometry verified three clones that specifically recognized MKN‐45 cells. Liquid chromatography‐mass spectrometry analysis of the scFv‐immunoprecipitated proteins has led to identification of c‐Met, HSP90 α and HSP90 β as candidate biomarkers associated with diffuse GC. Immunohistochemistry revealed the capability of purified scFvs to differentiate diffuse and intestinal gastric adenocarcinoma. Taken together, the isolated MKN‐45‐specific scFv fragments and their cognate antigens would be beneficial in screening and management as well as targeting and therapy of the diffuse gastric adenocarcinoma.

A panel of MKN‐45 cell‐specific scFvs antibodies were isolated using a subtractive phage display selection method for biomarker discovery. Three diffuse gastric adenocarcinoma‐associated biomarkers including c‐Met, HSP90 α and HSP90 β were identified by LC‐MS/MS analysis of the scFv‐immunoprecipitated proteins. Two purified scFvs were capable of discriminating the diffuse type of gastric adenocarcinoma from the intestinal type in immunohistochemistry staining.

Abbreviations

2D‐PAGE

two‐dimensional polyacrylamide gel electrophoresis

Fab

fragment antigen binding

FACS

fluorescence‐activated cell sorting

FITC

fluorescein isothiocyanate

GC

gastric cancer

HGFR

hepatocyte growth factor receptor

HSP90 α

heat shock protein 90 alpha

HSP90 β

heat shock protein 90 beta

LC‐MS/MS

liquid chromatography-mass spectrometry

MACS

magnetic‐activated cell sorting

MALDI‐TOF

matrix‐assisted laser desorption/ionization-time of flight

OD

optical density

PFA

paraformaldehyde

scFv

single‐chain variable fragments

SDS−PAGE

sodium dodecyl sulphate−polyacrylamide gel electrophoresis

TMB

tetramethylbenzidine

V_H

variable heavy chain

V_L

variable light chain

Introduction

Gastric cancer (GC) is the fifth most common malignancy and the third most lethal cancer‐related disease worldwide.1 From a histopathological perspective and Lauren’s criteria, GC is categorized into diffuse and intestinal subtypes. Diffuse type adenocarcinomas are poorly differentiated, and are mainly comprised of non‐cohesive tumour cells that invasively infiltrate normal stomach tissue. On the contrary, intestinal subtype is composed of cohesive tumour cells causing discrete tumour masses. Diffuse GC is more aggressive with a higher incidence of metastasis and poor prognosis in comparison to the intestinal counterpart.2 Lauren’s classification is still accepted and used widely by pathologists and physicians for GC management. Therefore, comprehensive understanding of molecular characterization of Lauren’s classification would assist in diagnosis, prognosis and treatment outcomes. In addition, the lack of proper cancer biomarkers for early diagnosis and treatment of diffuse GC complicates disease outcomes, resulting in high incidence and mortality rates. Therefore, it is necessary to apply new molecular biology tools and different recombinant technologies for discovery of potential tumour markers related to diffuse GC.

A large body of evidence has shown that most tumour‐associated biomarkers are cell surface proteins that are expressed differently in normal and cancer cells. Thus, heterogeneity in cell surface proteins would be helpful in determination of carcinogenesis stage and prognosis. However, discovery of cell surface proteins as potential biomarkers for clinical use is a difficult and laborious task. Two‐dimensional polyacrylamide gel electrophoresis (2D‐PAGE) coupled with mass spectrometry (MS) is a common tool to analyse proteome variations among different cell types. Alternatively, recombinant antibody and molecular display technologies provide additional valuable tools in biomarker discovery.3, 4 Combining phage antibody display, a high‐throughput antibody screening tool, with proteome analysis could be promising in simultaneous identification of novel targetable cell surface antigens and antibody therapeutics.5, 6

Over the past decade, phage antibody display has offered an elegant approach to generate specific antibodies against almost any antigen. The basic principle behind this technique is genetic engineering in which genes encoding antibody fragments [e.g. Fab (fragment antigen binding) or scFv (single‐chain variable fragments)] are cloned mostly close to the gene 3 of M13 bacteriophage in order to be displayed on the surface of the phage particles as g3p (minor coat protein)‐scFv fusion protein. The most commonly used antibody format for display is scFv − the smallest functional domain of antibodies − which is designed by linking variable heavy (V _H) to variable light (V _L) chains via a peptide linker (Gly₄Ser)₃. Isolation of specific phage binders from a phage antibody library is achieved by a sequential affinity‐based selection, the so called ‘biopanning’, through incubation of scFv‐displayed phages with target antigen, washing out non‐specific binders, and retrieval of specific phages. Upon different antigen platforms, biopanning can be carried out either on purified antigens or on the whole cell.7

Whole‐cell panning is considered as an ideal selection method for successful isolation of conformational‐specific antibodies, because it allows the target antigen to present itself closest to its native structure. Recently, whole‐cell‐based selection has been extensively used to discover disease biomarkers avoiding prior knowledge about antigenic profile of the cell surfaces, as well as to identify receptor‐specific or cell‐type‐specific antibodies. Moreover, screening could be done against a wide variety of antigens at the same time. However, antigenic complexity of cell surface affects the performance of cell‐based selection in a manner that undesired antibodies against common cell surface antigens would be dominated.8 In order to decrease non‐specific binders and enrich those that are specific, various cell‐based selection strategies are available, including subtraction,9 Pathfinder,10 competitive elution with receptor‐specific ligand,11 cell separation using magnetic‐activated cell sorting (MACS)12 or fluorescent‐activated cell sorting (FACS).13

Subtractive selection (positive/negative selection) includes an initial incubation with unwanted cells (negative selection) to deplete cross‐reactive antibody clones prior to panning against desired target cells (positive selection).14 In the present study, we applied subtractive selection to isolate recombinant antibody fragments recognizing the GC cell line, MKN‐45, to discriminate metastatic and poorly differentiated adenocarcinoma from non‐metastatic and well‐differentiated type, AGS. Using this subtractive phage display selection method, a panel of metastatic GC‐cell‐specific scFvs were isolated. MKN‐45‐specific antigens that may be involved in cell migration and metastasis were successfully retrieved through immunoprecipitation with the selected scFvs and subsequently their cognate candidate antigens were identified by liquid chromatography‐mass spectrometry (LC‐MS/MS). Immunohistochemistry analysis of GC specimens was performed to further validate in vivo specificity of the selected scFv antibodies. The selected scFv fragments can be employed as diagnostic probes, novel therapeutic agents, and for targeted delivery of the drugs and nanoparticles to GC. In addition, the discovered antigens can be used as potential tumour markers in diagnosis and screening of metastatic gastric adenocarcinoma.

Materials and methods

Phage antibody library and bacteria strains

A semi‐synthetic human scFv library (Tomlinson I) of approximately 1·47 × 10⁸ transformants (Source Bioscience, Nottingham, UK), KM13 helper phage (Source Bioscience), Hyperphage (PROGEN, Heidelberg, Germany), Escherichia coli strains TG1 for phage propagation and HB2151 for soluble scFv production (Source Bioscience, Nottingham, UK) were utilized in the current study. In the library, genes encoding scFvs were cloned close to the g3p into a phagemid vector, pIT2, with His‐ and c‐Myc tags as an N‐terminal fusion of the scFv sequence.

Cell culture

Human gastric adenocarcinoma cell lines (AGS and MKN‐45) and murine embryonic fibroblast cell line (NIH‐3T3) were purchased from ACECR (Academic Center for Education, Culture and Research, Iranian Biological Resource Center, Tehran, Iran). AGS and MKN‐45 cells were cultured in RPMI 1640 (R4130, Sigma‐Aldrich, St Louis, Missouri) supplemented with 10% and 20% fetal bovine serum (FBS), respectively. Murine embryonic fibroblast cell line NIH‐3T3 was cultured in DMEM (Gibco, Waltham, Massachusetts) supplemented with 10% FBS. All cell lines were maintained at 37° in a humidified atmosphere containing 5% CO₂.15

Whole‐cell panning

All three cell lines (NIH‐3T3, AGS and MKN‐45) were cultured in 75‐cm² cell culture flasks to reach 60% confluency. The cell monolayers were washed twice with phosphate‐buffered saline (PBS) and dispersed by cell dissociation buffer (C5914, Sigma) treatment for 10 min at 37°. Subsequently, the prepared cell suspension and 5 × 10¹² CFU/ml phage‐scFv produced from Tomlinson I library were separately blocked with PBS/3% bovine serum albumin (BSA) for 45 min. Diverse antibody phage‐scFv libraries were prepared from E. coli transformants infected with two different helper phages, KM13 and hyperphage, in parallel. The subtractive panning was carried out using NIH‐3T3 and AGS as negative cells to reduce non‐specific phage binders. For this purpose, the blocked phage‐scFvs were incubated with 5 × 10⁶ NIH‐3T3 cells for 1 hr at room temperature with over‐head rotation. After centrifugation at 250 g for 10 min at room temperature, phage‐containing supernatant was incubated twice with 5 × 10⁶ AGS cells for 30 min. The subtracted phage supernatant was collected after centrifugation and used for positive selection through incubation with 10⁷ MKN‐45 cells for 1·5 hr. After five times washing with PBS, cell‐bound phages were eluted with TBSC (10 mm Tris pH 7·4, 137 mM NaCl, 1 mm CaCl₂) containing trypsin from bovine pancreas (1 mg/ml; Sigma‐Aldrich) for 30 min at room temperature with over‐head rotation. The eluted phages were collected and amplified by infection of E. coli TG1 at mid‐logarithmic growth phase, as indicated with optical density of 0·4 at 600 nm (OD₆₀₀ = 0·4), for further rounds of selection according to the library’s manual protocol.

Polyclonal phage ELISA

Whole‐live‐cell ELISA was performed to determine the binding activity of phage‐scFvs to MKN‐45 and AGS cell lines.16 To monitor enrichment of specific binders, a suspension of 5 × 10⁵ cells/well in 2% MPBS (PBS containing 2% non‐fat dried skimmed milk) was distributed in 96‐well cell culture plates and incubated for 1 hr at room temperature. After centrifugation at 500 g for 5 min, both plates containing MKN‐45 and AGS cells, designated, respectively, as target and background plates, were incubated with 100 µl/well 2% MPBS containing 10 µl polyclonal phages that were amplified after each round of selection using either KM13 or Hyperphage. For detection of specific cell‐binding phage particles, the phage‐cell mixture was sequentially incubated with 100 µl/well anti‐M13 antibody (GE Healthcare) and horseradish peroxidase (HRP)‐conjugated goat anti‐mouse IgG (Thermo Fisher Scientific, Waltham, Massachusetts) at 1 : 5000 dilution prepared in 2% MPBS. The cells were washed four times and stained with 100 µl/well TMB (tetramethylbenzidine) substrate in the dark for 30 min at room temperature. After centrifugation, the supernatant was transferred to a new 96‐well plate containing 50 µl/well 5% H₂SO₄ and then absorbance was read at 450 nm subtracted from 630 nm.

Monoclonal phage ELISA

As a primary screening of binding specificity, whole‐cell ELISA was performed to analyse the binding of individual phage‐scFv clones to AGS and MKN‐45 cells. Randomly picked bacterial colonies related to KM13 and hyperphage were inoculated into 96‐well cell culture plates with 100 µl/well 2xTY‐AG (100 µg/ml ampicillin and 4% glucose) and grown overnight at 37° with vigorous shaking. A 5‐µl aliquot from the overnight cultures was transferred to 200 µl 2xTY‐AG and incubated at 37° with shaking until OD₆₀₀ = 0·4. Subsequently, production of phage‐scFvs was induced by super‐infection of the bacterial cultures with approximately 10⁹ PFU (plaque forming unit) of KM13 or hyperphage and incubation overnight at 30°.17 The supernatants of individual colonies were transferred to 96‐well plates containing 4% MPBS to block phage‐scFvs and then used for live whole‐cell ELISA as described before. The detection and staining process was performed using the same method as polyclonal phage ELISA.

ScFv diversity and sequence analysis

To determine diversity of the clones, double‐stranded phagemid DNA of ELISA‐positive clones was extracted using Qiaprep Spin Miniprep Kit (Qiagen, Zist Baran, Tehran, Iran). Polymerase chain reaction (PCR) was carried out using plasmid‐specific primers LMB3 (5'‐CAG GAA ACA GCT ATG AC‐3') and pHEN (5'‐CTA TGC GGC CCC ATT CA‐3') during 35 cycles under denaturation, annealing and extension at 94°, 55° and 72°, respectively. DNA sequencing was performed by DNA sequencer using LMB3 primer and, then, sequence analysis was accomplished by Chromas 2·33 software (Technelysium Pty, Queensland, Australia). Alignment of amino acid sequences was done in VBASE2 database to identify clone diversity.18

Expression and purification of soluble scFv fragments

To express soluble scFvs, scFv‐encoding phagemids of ELISA‐positive clones were transformed into the E. coli strain HB2151. For this purpose, individual scFv clones were grown in 200 ml 2xTY‐AG shaking at 37° to an OD₆₀₀ of 0·9. Induction of soluble scFvs was done at 30° for 5 hr by addition of 1 mm IPTG (isopropyl‐thiogalactopyranoside). To extract soluble scFvs from the periplasmic fractions, the bacterial pellets were harvested by centrifugation at 3000 g for 10 min and incubated with 1/20 ice‐cold TES [50 mm Tris pH 8·0, 1 mm EDTA and 20% (w/v) sucrose] for 1 hr on ice. After centrifugation at 20 000 g for 30 min at 4°, the supernatant containing scFvs was collected and dialysed against PBS buffer overnight.

Purification of His‐tagged soluble scFvs was performed by IMAC affinity chromatography using Ni‐NTA agarose (Clontech, Takara Bio) according to the manufacturer’s instructions. Briefly, 1 ml of the resin was incubated with dialysed samples related to each scFv for 30 min at room temperature. After washing twice with 5 ml wash buffer (50 mm Na₃PO₄ and 300 mm NaCl, pH 7·0), the suspension was passed through the column and the bound scFvs were eluted using wash buffer containing 150 mm imidazole. The purified scFvs were dialysed using Maxi Pur‐A‐Lyzer Dialysis tubes with 12 000 Da cut‐off (Sigma‐Aldrich CHEMIE GmbH, Steinheim, Germany) and stored at −20° for further analysis. The concentrations of purified scFvs were determined at the OD 280 nm.19

SDS−PAGE and Western blotting

The evaluation of the expression and purification was accomplished by both sodium dodecyl sulphate−polyacrylamide gel electrophoresis (SDS−PAGE) and Western blotting based on Bio‐Rad recommendations. The expressed and purified samples were run on two separate gels in parallel. The gels were either stained with Coomassie Brilliant Blue R‐250 or transferred to a nitrocellulose membrane using semi‐dry trans‐blotting system (Bio‐Rad, Munich, Germany). The membrane was blocked in 5% MPBS‐Tween 20 (0·05%) overnight at 4° and then incubated with anti‐c‐Myc monoclonal antibody (mAb; Sc‐40, Santa Cruz Biotechnology, Santa Cruz, CA; 1 : 1000) and HRP‐conjugated goat anti‐mouse IgG (1 : 3000) diluted in 2% MPBS for 1 hr at room temperature, respectively. All incubations were accompanied by three times washing with PBST (PBS 0·05% Tween 20) for 5 min at room temperature. Finally, the blot was visualized by ECL Western blotting kit (GE Healthcare) and X‐ray film.

Binding analysis of scFv fragments by ELISA and flow cytometry

The binding activity of the purified scFv antibodies to MKN‐45, AGS and NIH‐3T3 cell lines was evaluated by whole‐cell ELISA and flow cytometry. ELISA analysis was performed by using 2 µg of the purified scFvs as described before, except that HRP‐conjugated protein A was used for detection of the specific scFv binders.20, 21 Binding specificity of the selected clones was further verified by flow cytometry. Following washing, the cells were blocked in 3% BSA/PBS, 0·03% NaN₃. Cell binding analysis was performed with incubation of 5 × 10⁵ cells with 20 µg/ml scFv for 30 min on ice. After washing, the cells were incubated with 100 µl anti‐c‐Myc mAb diluted 1:100 in FACS buffer for 30 min on ice to detect scFv binders. Then, the cells were stained with 100 µl of fluorescein isothiocyanate (FITC)‐labelled rat anti‐mouse IgG‐ (Biolegend) diluted 1 : 100 in FACS buffer for 30 min on ice. All incubations were followed by washing with FACS buffer (1% BSA/PBS, 0·03% NaN3) and centrifugation at 300 g for 3 min at 4°. Finally, the cells were resuspended in 500 µl PBS and analysed using FACs calibre.

Internalization assay by flow cytometry

The modified internalization assay to detect scFv fragments capable of internalizing into MKN‐45 cells was performed by pre‐incubation of the scFvs with anti‐c‐Myc mAb in a total volume of 100 µl RPMI for 1 hr at room temperature. The scFv‐tag antibody complexes were incubated with pre‐blocked MKN‐45 cells for 2 hr at 37°. To remove surface‐bound scFvs, the cells were washed twice with cold PBS, and three times for 10 min with stripping buffer (50 mm glycine, 500 mm NaCl, pH 2·8). After two additional washing steps with PBS, the cells were fixed in 4% paraformaldehyde (PFA), permeabilized with cold methanol for 10 min and blocked with PBS/1% BSA for 30 min at room temperature. Isotype controls without the scFvs were also included in the experiments. Finally, the samples were incubated with 1 µg rat (FITC)‐labelled rat anti‐mouse IgG for 1 hr at room temperature. After washing, internalization of the scFvs was detected by flow cytometry.22, 23

Immunoprecipitation

MKN‐45 cells were grown on 10‐cm² cell culture plates to 90% confluence. Cells were washed three times with ice‐cold PBS (pH 8·0) and labelled with 1 mg/ml sulpho‐NHS‐LC‐biotin (Apex Bio) freshly dissolved in PBS for 30 min at 4° under rotation. Afterward, the cells were washed three times with PBS supplemented with 100 mm glycine to quench unreacted biotin reagent. The biotinylated cells were collected and resuspended in lysis buffer (1% NP40, 150 mM NaCl, 1 mM EDTA, 10 mm Tris pH 8·0) plus protease inhibitors (Sigma) overnight at 4°. Insoluble debris was removed by centrifugation at 21 000 g, and supernatant containing proteins were pre‐cleared by incubation with 100 µl of Protein A‐agarose beads (Santa Cruz) for 2 hr at 4°. Depleted cell lysates were incubated with 26 µg scFv for 2 hr and immune complexes were captured with 100 µl Protein A‐agarose beads. The beads were pelleted by centrifugation at 250 g at 4° for 2 min and washed four times in PBS. Elution of agarose‐bound immune complexes was performed by resuspending in 50 µl reducing 1 × SDS sample buffer and heating at 95° for 5 min. After centrifugation, the supernatant containing immunoprecipitates was run on 8% SDS−PAGE in duplicate. One gel was transferred onto nitrocellulose membrane and stained with streptavidin‐HRP conjugate (Biolegend) to verify proper immunoprecipitation. The other gel was stained with Coomassie Brilliant Blue R‐250, and protein bands corresponding to target antigen were excised and analysed with LC‐MS/MS.

LC‐MS/MS analysis

For proteomics sample preparation, the protein bands were cut from the gel and digested with trypsin. Briefly, proteins were reduced with dithiothreitol (DTT) to the final concentration of 10 mm for 30 min at room temperature and alkylated with 50 mm iodoacetamide (IAA) for 30 min at room temperature in the dark. Modified sequencing grade trypsin (Promega) was then added (0·5 µg) followed by overnight incubation. The peptides were cleaned using StageTips (Thermo) according to the manufacturer’s protocol. Samples were dried using SpeedVac centrifugal evaporator (DNA120, Thermo) and, prior to LC‐MS/MS analysis, 0·1% formic acid solution (Fluka) was added to the samples. Protein digests were analysed in a randomized order by LC‐MS/MS. Samples were loaded onto a 50‐cm column [EASY‐Spray, 75 µm internal diameter (ID), PepMap C18, 2 µm beads, 100 Å pore size] connected to a nanoflow Dionex UltiMate 3000 UPLC system (Thermo) with buffer A (0·1% formic acid in water) and eluted with a 60 min gradient reaching from 4% to 26% of buffer B (98% ACN, 0·1% FA, 2% H₂O) at a flow rate of 300 nl/min. Mass spectra were acquired with an Orbitrap Q Exactive mass spectrometer (Thermo Fisher Scientific) in the data‐dependent mode at a nominal resolution of 17 500, in the m/z range from 375 to 1500. Peptide fragmentation was performed via higher‐energy collision dissociation (HCD) with energy set at 28 NCE. The ion selection abundance threshold was set at 0·1% with exclusion of singly charged ions.

The raw data from MS were analysed by MaxQuant, version 1.5.6.5.24 The Andromeda search engine matched MS/MS data against the Uniprot complete proteome database (human, version UP000005640_9606, 92 957 entries). Cysteine carbamidomethylation was used as a fixed modification, while methionine oxidation was selected as variable modifications. No more than two missed cleavages were allowed. A 1% false discovery rate was used as a filter at both protein and peptide levels. Label‐free quantification of peptides and proteins was performed. Match between runs function was enabled. All the other parameters were set to default. After quantification, the contaminants were removed, and only proteins with at least two peptides were considered.25

Immunohistochemistry

Immunohistochemistry was carried out on 5‐µm sections of gastric adenocarcinoma biopsies obtained from regular paraffin blocks. Tissue sections were deparaffinized in xylene, rehydrated in graded alcohol series, and antigen retrieval was performed in 95° prewarmed citrate buffer 10 mm, pH 9·0. Endogenous peroxidase was quenched by incubation with 3% H₂O₂ in methanol for 10 min at room temperature. After washing with PBS, sections were blocked in 20% normal goat serum and incubated with purified scFvs (1:5) overnight at 4°. After washing with PBS, sections were incubated with anti‐c‐Myc mAb (1:50 Santa Cruz) and HRP‐conjugated goat anti‐mouse IgG for 2 hr at room temperature, respectively. Detection of bound HRP was performed by 3,3‐diaminobenzidine (DAB) for 5 min at room temperature.

Results

Subtractive cellular panning against MKN‐45

Semi‐synthetic human Tomlinson library I was used to search for novel antigens related to diffuse gastric adenocarcinoma. A strategy was devised for the isolation of MKN‐45‐specific phage particles comprised from two sets of selection with phage library rescued by either KM13 or Hyperphage in parallel, in order to expand potential diversity of specific binders. To eliminate cross‐reactive clones related to common cell surface antigens, subtractive panning was performed through three pre‐absorption steps with NIH‐3T3 and AGS cells followed by one step of positive selection with MKN‐45. Overall, four successive rounds of subtractive panning were carried out. After four rounds, the enrichment factors were 743 and 1465 for the library rescued by KM13 and Hyperphage, respectively.

Polyclonal phage ELISA was performed to further analyse specific phage enrichment. To this aim, amplified phages after each round were incubated with both MKN‐45 and AGS cells. In terms of surface proteome, AGS is a closer cell line to MKN‐45 than NIH‐3T3 cells, and so it was considered as a stringent negative control in this study. For both selection with KM13 (Fig. 1a) and Hyperphage (Fig. 1b), an increase in the specific binding activity to MKN‐45 cells was initiated after round 3 and completed at round 4. Signal intensities for MKN‐45‐specific binders showed about three‐ and four‐fold increase over the negative control cells (AGS).

Figure 1.

Affinity selection of MKN‐45‐specific phage‐scFvs. Representative polyclonal phage ELISA was applied to evaluate the enrichment of specific binders with phage library rescued by (a) KM13 or (b) Hyperphage. Polyclonal phages amplified after each round were incubated with 5 × 10⁵ cells/well MKN‐45 (black bars) and AGS (striped bars), representative of target and negative control cells, respectively. Detection of cell binding phages was performed with anti‐M13 antibody and horseradish peroxidase (HRP)‐conjugated goat anti‐mouse IgG. The values were calculated as absorbance at 450 nm (A₄₅₀) subtracted from background absorbance at 630 nm. Each value represents standard error of the mean (SEM) of two independent experiments (means ± SD). Monoclonal phage ELISA was used to screen specific phage‐scFvs clones rescued by (c) KM13 or (d) Hyperphage. The chart represents individual phage clones with at least threefold higher absorbance for binding to MKN‐45‐positive cells (black bars) in comparison with AGS‐negative control cells (stripped bars).

Screening and diversity analysis

Individual phage clones were randomly selected from the fourth round of panning and analysed for their binding ability to MKN‐45 and AGS cells as well as clone diversity. Based on immunostaining with ELISA, phage‐scFv isolates with > threefold absorbance value over background were considered as positive clones. Of 192 clones analysed, 14 candidates for KM13 (Fig. 1c) and seven candidates for Hyperphage (Fig. 1d) with a strong affinity to MKN‐45 and low cross‐reactivity to AGS cells were selected for diversity analysis. As shown in Table 1, homology analysis and CDR regions comparison based on database VBASE2 showed five different sequences, among which one sequence related to KM13 helper phage was over‐represented 13 times, and another sequence that belongs to Hyperphage was over‐represented four times. Overall five representative clones with respect to their binding specificity (including B12 and F10 for KM13 helper phage, and B10, G1 and G11 for Hyperphage) were selected for further analysis.

Table 1.

Sequence analysis of selected phage‐scFv clones

Selected phage‐scFvs	% Frequencya (number of occurrences)	V H			V L			Yield (mg/ml)
		CDR1	CDR2	CDR3	CDR1	CDR2	CDR3
B12	4·7 (1/21)	GFTFSSYA	IDGGGTGT	AKTDSGFDY	QSISSY	TAS	QQDAAAPAT	0·28
F10	62 (13/21)	GFTFSSYA	ITATGATT	AKSAAGFDY	QSISSY	YAS	QQTNYDPTT	0·16
B10	4·7 (1/21)	GFTFSSYA	ISSSGSNT	AKDSSYFDY	QSISSY	NAS	QQDSDNPTT	0·37
G1	19 (4/21)	GFTFSSYA	ISATGNST	AKSGGTFDY	QSISSY	AAS	QQAYDDPST	0·33
G11	9·5 (2/21)	GFTFSSYA	ISGSGSYT	AKDDTNFDY	QSISSY	SAS	QQYTTDPTT	0·53

scFv, single‐chain variable fragments; V _H, variable heavy chain; V _L, variable light chain.

Diversity in amino acid sequences of CDR regions was determined based on database VBASE 2.

^aFrequency was calculated as the percent of the individual antibody fragment clones occurred in the total number of ELISA‐positive antibody clones.

Specificity analysis of soluble scFv fragments

The phagemid vector of the selected phage antibodies were used to infect E. coli strain HB2151 to express soluble scFv fragments. Periplasmic fraction of bacterial cell lysates containing 6xHis‐tagged scFvs were extracted and purified by IMAC affinity chromatography using Ni‐NTA agarose. SDS−PAGE (Fig. 2a,b) and Western blot analysis (Fig. 2c,d) indicated successful expression of soluble scFvs with an expected molecular size of ~28 kDa. Additionally, visualization of the single bands indicated homogeneity of the scFvs and purity above 90%.

Figure 2.

Sodium dodecyl sulphate−polyacrylamide (SDS−PAGE) and Western blot analysis of expression and purification. The samples related to expression and purification steps were analysed by 12% SDS−PAGE (a, b) followed by immunoblotting (c, d). The gels were stained with Coomassie blue R250. The blots were probed by anti‐c‐Myc mAb and horseradish peroxidase (HRP)‐conjugated goat anti‐mouse IgG, followed by incubation with ECL and appearing on X‐ray film. The arrows denote the approximate molecular weight of the scFvs. Molecular weight standards in kDa, M; uninduced total cell lysate, UN; induced total cell lysate, IN; uninduced periplasmic extracts, UNP; induced periplasmic extracts related to the selected clones; periplasmic extraction, P; flow‐through, F; wash fraction, W; eluted fractions of each scFv clones.

To verify the specificity of the soluble scFv fragments, the purified scFvs were analysed both by whole‐cell ELISA and flow cytometry using MKN‐45, AGS and NIH‐3T3 cell lines. As shown in Fig. 3(a), ELISA analysis showed that B12 and G1 scFvs selectively and strongly bind to MKN‐45, while binding activity of F10 to MKN‐45 was moderate. The unrelated scFv as negative control revealed no evidence of binding to the three cell lines.

Figure 3.

Binding specificity analysis of the purified scFvs. In ELISA test (a), 5 × 10⁵ MKN‐45 (black bars), AGS (stripped bars) and NIH‐3T3 (grey bars) cells were sequentially incubated with purified scFvs (2 µg/well), anti‐c‐Myc mAb and horseradish peroxidase (HRP)‐conjugated goat anti‐mouse IgG. For all scFvs, absorbance values that represent standard error of the mean from duplicate experiments were in favour of MKN‐45 cells. In flow cytometry analysis (b), MKN‐45, AGS and NIH‐3T3 cells were incubated with purified scFv fragments (2 µg) for 30 min on ice. Bound scFvs were detected by anti‐c‐Myc mAb and fluorescein isothiocyanate (FITC)‐labelled rat anti‐mouse IgG. All plot histograms represent analysis of live cells by excluding PI‐positive cells (dead cells). Fluorescence intensity of the scFv staining (green line) in contrast to the corresponding isotype controls (filled grey peaks) were analysed by FACSCalibur and CellQuest Pro software. All scFvs bind preferentially to MKN‐45 and not to the NIH‐3T3 and AGS (negative control cells). Unrelated scFv was used as negative control.

In addition, flow cytometry analysis (Fig. 3b) was conducted to evaluate the reactivity of the purified scFv fragments. B12, F10 and G1 scFvs were positive with MKN‐45 as illustrated by the right shift in the related peak without any shift for both AGS and NIH‐3T3 (as negative cells) compared with the respective isotype controls. Unrelated scFv was unreactive with the three cell lines. G11 and B10 scFvs were neglected from further analysis due to non‐specific or weak binding.

Cell binding and internalization behaviour

To ensure potential application of B12, F10 and G1 scFv fragments in targeted delivery of the drugs and nanoparticles, we examined the internalization behaviour of the selected scFvs by flow cytometry. To this aim, we mixed anti‐c‐Myc mAb with purified scFvs prior to incubation with MKN‐45 cells. Histogram plots (Fig. 4a) illustrated uptake of all three scFv fragments into MKN‐45 cells. Significant fluorescence signal in FL1 for 37° test samples in comparison to no fluorescence signal for 4° sample with stripping (as a negative control) implied internalization of the scFv fragments. Additionally, FITC staining of the 4° sample without stripping (as a positive control) reflected surface‐bound scFvs without intracellular uptake. As shown in Fig. 4(b), the flow cytometry results were statistically quantified by CellQuest Pro software after normalizing with respective isotype controls (without scFvs treatment). The results revealed 63·52%, 57·9% and 35·14% intracellular uptake for B12, F10 and G1 scFvs, respectively. The percentage of shifted cells in the negative control was the same as isotype controls with only 8·4% uptake indicating successful striping of the surface‐bound scFvs, while the percentage of shifted cells in the positive control was 58% indicating that surface‐bound scFv did not enter the cells in the presence of NaN₃ at 4°.

Figure 4.

Internalization and immunoprecipitation assay. For internalization assay, MKN‐45 cells were incubated with preformed scFv‐c‐Myc complexes for 2 hr at 4° and 37° to analyse surface binding and internalization behaviour of the selected scFvs. Staining of the cells was carried out with fluorescein isothiocyanate (FITC)‐labelled rat anti‐mouse IgG. Histogram plots (a) illustrate fluorescent signals of isotype control (filled grey line, the cells without the scFvs treatment) and test samples (green lines, the cells treated with B12, F10 and G1 scFvs). The chart (b) illustrates the percentage of the surface‐bound and internalized scFvs. For 37° samples, the fluorescence signal after removal of surface‐bound scFvs via glycine treatment shows internalization of all scFvs. All samples were normalized to isotype controls, and then binding and internalization of the scFvs are represented as the percentage of shifted cells in FL1. For 4° samples, the presence of significant shift in FL1 for positive control along with minor shift in FL1 for negative control reflects cell binding reactivity of the scFv in the absence of glycine treatment. The result is obtained from triple independent experiments. Identification of target was done by scFv immunoprecipitation (c). The biotinylated MKN‐45 cell lysate was precipitated through incubation with scFv fragments and pulling‐down by protein A‐agarose. Following separation by 8% sodium dodecyl sulphate−polyacrylamide gel electrophoresis (SDS−PAGE) and blotting, membranes were probed with streptavidin‐horseradish peroxidase (HRP) conjugate. The arrows indicate molecular weight of protein bands immunoprecipitated by designated scFv clones above each lane. Non‐related scFv (NR) was used as negative scFv control.

Identification of target antigens

Showing the most specificity for MKN‐45 cells, the selected scFvs (i.e. B12, F10 and G1) were used to identify corresponding target antigens. To this aim, immunoprecipitation was performed to retain 6xHis‐tagged scFvs and their binding partners in MKN‐45 cell lysate. To increase the efficiency of the immunoprecipitation procedure, surface molecules were biotinylated prior to cell lysis and selected scFvs were used to immunoprecipitate antigens from cell extracts. Western blot analysis of immunoprecipitates with F10 scFvs revealed two protein bands approximately 100 and 140 kDa (Fig. 4c). G1 and B12 scFvs also immunoprecipitated protein bands at 100 and 170 kDa, respectively. The protein bands were subjected to in‐gel digestion, and the resulting peptides were analysed by LC‐MS/MS. Table 2 represents an overview of the target antigen candidates and their characteristics. Hepatocyte growth factor receptor (HGFR), c‐Met (UniProtKB accession number: P08581) was found as a cognate antigen precipitated by F10 scFv. Sequence analysis of the peptides for both 100‐ and 140‐kDa protein bands unambiguously identified eight unique peptide sequences that matched the MET gene. Heat shock protein (HSP) 90 alpha, HSP90AA1 (UniProtKB accession number: P07900) and HSP90 beta, HSP90AB1 (UniProtKB accession number: P08238) were proposed as candidate antigens for G1 scFv. The LC‐MS/MS analysis identified three and eight unique peptides for HSP90AA1 and HSP90AB1, respectively. MS analysis of the gel slice related to B12 scFv did not match to any appropriate protein candidate.

Table 2.

Overview of candidate target antigens

scFv	Antigen	Accession number (UniProtKB)	Mol. weight (on SDS−PAGE)	Predicted mol. weight	No of peptides read out in LC‐MS/MS	No of unique peptides in LC‐MS/MS	Localization
F10	HGFR (MET)	P08581	140 kDa, 100 kDa	85·74–155·54 kDa	8	8	Cell membrane/cytoplasm
G1	HSP90 α (HSP90AA1)	P07900	100 kDa	84·659–98·16 kDa	9	3	Cell membrane/cytoplasm
	HSP90 β (HSP90AB1)	P08238	100 kDa	83·263 kDa	15	8	Cell membrane/cytoplasm

HGFR, hepatocyte growth factor receptor; HSP90 α, heat shock protein 90 alpha; HSP90 β, heat shock protein 90 beta; LC‐MS/MS, liquid chromatography‐mass spectrometry; scFv, single‐chain variable fragment; SDS−PAGE, sodium dodecyl sulphate−polyacrylamide gel electrophoresis.

Immunohistochemistry analysis

To confirm the relevance of MKN‐45‐specific scFv fragments (F10 and G1) to diffuse gastric adenocarcinoma biopsies, immunohistochemical staining was performed on paraffin‐embedded tissue sections from gastric adenocarcinoma as well as normal gastric tissues. As shown in Fig. 5, both F10 and G1 scFvs could specifically stain four (40%) and six (60%) of 10 diffuse adenocarcinoma biopsies, respectively. The result revealed no sign of staining in intestinal and normal sections indicating specificity of these two antibody fragments.

Figure 5.

Immunohistochemical staining of gastric cancer (GC) tissues to verify scFv specificity. Ten biopsies of diffuse and intestinal gastric adenocarcinomas as well as normal gastric slides were stained with F10 and G1 scFv fragments. After deparaffinization and antigen retrieval in prewarmed citrate buffer, tissue sections were sequentially incubated with scFv fragments, anti‐c‐Myc mAb and horseradish peroxidase (HRP)‐conjugated goat anti‐mouse IgG. scFv binding was visualized by 3,3‐diaminobenzidine (DAB) and nuclei were counterstained with haematoxylin. Both F10 and G1 revealed strong membranous/cytoplasmic staining on diffuse gastric adenocarcinoma biopsies without any cross‐reactivity to intestinal gastric adenocarcinoma and normal gastric tissues. Magnification, 400 ×.

Discussion

Increasing interest for interrogation of novel cell surface biomarkers as tumour‐related diagnostic and therapeutic targets has encouraged the development of novel technologies. Proteomics‐based approaches including 2D‐PAGE, MS techniques (i.e. LC/MS, MS/MS) and matrix‐assisted laser desorption/ionization time of flight (MALDI‐TOF), as well as protein and antibody arrays have been implemented for biomarker discovery in cancer research.26 Phage antibody display is another approach that has made an important impact on cancer biomarker discovery and proteomics studies.27 This promising technique along with improved selection and screening methods have allowed to generate high‐affinity antibodies against tumour‐associated antigens and, at the same time, to identify novel diagnostic and therapeutic targets. Cell surface membrane proteins that are differently expressed between healthy and tumour cells are the main targets for isolation of antibody molecules. However, selection of tumour‐cell‐specific antibodies is experimentally problematic mainly due to low abundance of target antigen in the complex protein context of the cell surfaces.3 As a result, isolation of antibodies with desired tumour cell specificity may be obscured by clones that bind to the irrelevant common antigens that may dramatically affect the selection outcomes. To this end, subtractive selection including depletion steps on irrelevant cells prior to positive selection on a given target cell can be applied in order to remove common and non‐specific antibody binders.

Among various solid tumours, GC is one of the leading causes of cancer‐related mortality worldwide. Based on Lauren’s classification, two distinct histological types of GC (i.e. intestinal and diffuse type) have been described. The intestinal adenocarcinoma is characterized by good differentiation and expansive growth, while diffuse type is poorly differentiated with infiltrative growth and expansion into the peritoneal cavity.2 Metastasis that mainly occurs due to peritoneal dissemination is considered as the leading cause of mortality in gastric adenocarcinoma patients. This process is accompanied with alterations in the biology and molecular characteristics of the tumour and, therefore, identification of the biomarkers underlying cell migration and metastasis is of utmost importance in diagnosis, therapeutic and prognosis.28, 29

To isolate a panel of diffuse GC cell reactive antibodies from Tomlinson I library, we selected two gastric adenocarcinoma cell lines, MKN45 and AGS, as representatives for Lauren’s diffuse and intestinal subtypes, and applied a subtractive‐based selection procedure. To reduce the number of irrelevant phage‐scFvs and increase specific binders, we performed four rounds of biopanning each consisting of three negative steps with NIH‐3T3 and AGS followed by one step of positive selection on MKN‐45. Panning of phagemid‐based antibody libraries is challenging because only a small proportion of rescued phages are functional, which may limit the performance of the selection step. Therefore, to improve the efficiency of the panning process we utilized two different phages, KM13 and Hyperphage, in order to expand clone diversity and scFv displaying phages represented in the library.30 Polyclonal phage ELISA demonstrated enrichment of specific binders that initiated from round 3 and completed in round 4 (Fig. 1a,b). Screening of 192 clones confirmed 21 high‐affinity binders (14 candidates for KM13 and seven candidates for Hyperphage; Fig. 1c,d). Results obtained from amino acid alignment in VBASE2 database indicated five distinct scFv sequences with multiple frequency (Table 1). Specificity analysis of the purified scFvs through ELISA and flow cytometry verified that three out of five clones (i.e. B12, F10 and G1) were reactive with diffuse gastric adenocarcinoma cells (MKN‐45), but without or weak binding to intestinal gastric adenocarcinoma (AGS) and normal cells (NIH‐3T3). Clone G11 reacted with both MKN‐45 and AGS without any binding to NIH‐3T3 cells. Presumably, this clone might recognize a common antigen related to gastric adenocarcinoma. Clone B10 did not show any binding to any of the three cell lines. As we intended to identify discriminatory antigens, both G11 and B10 were neglected from further analysis.

To demonstrate the potential application of the isolated scFvs in site‐specific delivery of drugs, we also investigated whether binding of the scFv fragments activates receptor‐mediated internalization. In the first experiment, incubation of the scFvs with target cells did not result in intracellular uptake, probably in part due to monovalent format of the scFv fragments (data not shown). Based on previous investigations, we hypothesized that bivalent nature of the scFv fragments is a prerequisite for internalization via receptor dimerization.31 In order to provoke internalization, we preformed scFv‐c‐Myc complexes prior to incubation with MKN‐45 cells. As expected, all scFv fragments underwent endocytosis after 2 hr incubation at 37° without any uptake at 4° in control sample.

Succeeding in immunoaffinity purification is largely dependent on the fact that the isolated scFvs are capable of recognizing the antigens in the native structure. So far, only a few studies have accomplished in de novo discovery of cell surface antigens via scFv fragments obtained from phage display technology.32 Currently, to retrieve the respective target antigens contributing to the growth and migration of MKN‐45 cells, we performed immunoprecipitation using protein A‐agarose beads. Based on immunoaffinity purification results, F10 and G1 scFvs were able to precipitate three protein bands at molecular weights of 100 and 140 kDa. MS analysis of the 100‐ and 140‐kDa protein bands immunoprecipitated by F10 scFv led to the identification of c‐Met, receptor tyrosine kinase (RTK) [also known as hepatocyte growth factor (HGF) receptor], which is a proto‐oncogene often implicated in cancer.33 The MET (c‐Met encoding) gene is located on human chromosome 7 (7q21‐q31), encodes for a disulphide‐linked heterodimer consisting of a 50‐kDa extracellular alpha chain and a 145‐kDa transmembrane beta chain.34 c‐Met receptor and its ligand, HGF, play a crucial role during cancer development via activation of the signal transduction, mediating epithelial cell dissemination, epithelial remodelling and invasive growth.35 Rajadurai et al. 35 illustrated that c‐Met inhibition in MKN45 cells, which show constitutive genomic amplification of MET and production of cortactin‐positive actin‐rich protrusions, prevents formation of invadopodia. Interestingly, aberrant c‐Met signalling pathway as a result of c‐Met receptor dysregulation or HGF overexpression is accompanied with an aggressive phenotype. Extensive evidence has demonstrated the involvement of the c‐Met signalling in the progression and spread of various cancers including gastrointestinal types.36 Moreover, the expression of the c‐Met in diffuse type adenocarcinomas is stronger than the intestinal type.37

Mass spectrometry analysis of the 100‐kDa target protein precipitated by G1 scFv proposed HSP90 α (HSP90AA1) and HSP90 β (HSP90AB1) as other candidate biomarkers for the metastatic type of MKN‐45 gastric adenocarcinoma cell line. Heat shock proteins are highly conserved molecular chaperones with an approximate molecular weight of 90 kDa that regulate protein folding and prevent protein aggregation in all cell types. HSP90 α and HSP90 β that are encoded by different genes constitute 1%−3% of total protein content with 85% identity in amino acid sequences.38, 39 Besides implication in physiological process, HSPs are increased in diverse pathological conditions such as cancer, and their upregulation is strongly associated with metastasis.40 In the study conducted by Zuo et al.,41 the expression rate of HSP90 α in patients with GC was remarkably higher than that in patients with gastritis. Additionally, HSP90AB1 overexpression has a pivotal role in tumour growth and metastasis of GC cells. In the study performed by Wang et al.,42 injection of MKN‐45 cells with stable overexpression of H90AB1 into the tail vein of nude mice resulted in more lung and peritoneal metastasis compared with those with the control cells (MKN‐45 cells without overexpression of Hsp90AB1). HSP90 is required in maintaining the stability and activity of signalling proteins involved in cancer. Interestingly, it has been demonstrated that c‐Met, a HSP90‐dependent kinase, interacts with HSP90 to stabilize its activity.43 Immunohistochemistry of gastric adenocarcinoma tissue sections further confirmed that F10 and G1 scFvs were able to discriminate the diffuse type of gastric adenocarcinoma from the intestinal one.

In summary, we successfully isolated a panel of scFv fragments that can discriminate Lauren’s diffuse gastric adenocarcinoma cell line, MKN‐45, from intestinal ones, AGS, via whole‐cell panning of phage antibody library. Furthermore, c‐MET, HSP90 α and HSP90 β with known pivotal roles in tumour growth, GC metastasis and poor prognosis were discovered as promising molecular biomarkers for diffuse type gastric adenocarcinoma cells line, MKN‐45. Overall, the isolated antibodies may have potential biomedical applications in targeted therapeutic approaches, either in naked forms as blocking agents of ligand‐receptors interactions or in conjugated form with other cytotoxic agents as a carrier in targeted delivery of small molecules, radioactive agents and bacterial toxins.

Disclosures

The authors declare no conflict of interest.