The complete amino-acid sequence of CEA was acquired from the UniprotKB/Swiss-Prot database. The secondary structure of CEA amino acid sequence was analyzed by Garnier–Osguthorpe–Robson (GOR4) (Garnier et al., 1996), Self-Optimized Prediction Method (SOPMA) (Geourjon and Deléage, 1995), and Protein Secondary Structure (PSS) provided on the EXPASY server. Hydrophilicity, polarity, flexibility, accessibility, and antigenicity of CEA protein were analyzed by the methods of Hopp and Woods (1981), Zimmerman et al. (1968), Emini et al. (1985), and Jameson and Wolf (1988), respectively. Transmembrane domains were analyzed by TransMembrane prediction using Hidden Markov Models (TMHMM). Combined with the prediction results, the antigenicity index established by Jameson and Wolf (1988) was used to comprehensively evaluate the immunodominant B-cell epitopes of CEA and preliminarily select the target peptide for screening. Subsequently, the target peptide was analyzed with the NetCTL 1.2 webserver to find the dominant cytotoxic T-lymphocyte (CTL) epitope.

A phage display library was constructed as described previously (Xue et al., 2016). Affibody proteins were selected from a phage display combinatorial library containing 13 randomized amino acid residues in helices 1 and 2 of the Z domain. A wild SPA-Z scaffold was used as a template for polymerase chain reaction (PCR) with the random primers designed based on the corresponding sequences to the helical regions. Then, the gene fragments were digested by restriction endonucleases Sfi I and Not I (Thermo Fisher Scientific, Boston, MA) and cloned to the phagemid vector (pCANTAB5E) to construct the recombinant phagemid vector (pCANTAB5E/SPA-N). The recombinant phagemids were transformed into competent Escherichia coli TG1 cells, with a library complexity of 1 × 10⁹ and 100% diversity in SPA-Z scaffold. Then, for an assessment of the affibody library capacity, phage stocks were resuspended and stored in sterile phosphate-buffered saline (PBS)/glycerol (20% v/v) at −80°C.

A synthetic peptide (Shanghai Bootech BioScience & Technology Co., Ltd., Shanghai, China) with the dominant immune epitope of CEA 148–175 amino acids was used as a target for three rounds of biopanning and enzyme-linked immunosorbent assay (ELISA). In brief, phage selection of the binders to the CEA (148–175 amino acids) peptide was performed in an immunotube as previously described (Aavula et al., 2011). ELISA was used to further measure their binding affinities to the target peptide as described by Xue et al. (2016). After DNA sequencing, the sequences of the inserted fragments in the selected phages were designated as potential affibodies with high binding affinity and selectivity to CEA recombinant protein.

The DNA sequences encoding the selected CEA-binding affibody molecules (Z_CEA539, Z_CEA546, and Z_CEA919) and wild SPA-Z scaffold (Z_WT) were subcloned into E. coli expression vector pET21a (+) with the addition of His6 tag. The recombinant plasmids were transformed into E. coli BL21(DE3) and induced by 1 mM isopropyl β-D1- thiogalactopyranoside (IPTG, Sigma-Aldrich, St. Louis, MO) for protein expression. Recombinant proteins were purified by affinity chromatography using Ni-NTA Sepharose column (Qiagen, Hilden, Germany) and were dialyzed for 2 h in PBS using Slide-A-Lyzer (Pierce, Rockford, IL). The molecular weight and purity of the fusion proteins were confirmed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and Western blotting using an anti-His-tag monoclonal antibody (MultiSciences Biotech Co., Ltd., Hangzhou, China). The concentrations of the proteins were determined by the bicinchoninic acid (BCA) kit (Beyotime, Beijing, China) and protein quantitation method. The purified proteins were aliquoted and stored at −20°C.

To analyze the biospecific interaction between the selected affibody molecules and CEA protein, surface plasma resonance (SPR) was performed on a Biacore T200 (GE Healthcare, Uppsala, Sweden). The full-length CEA protein serving as a ligand was successfully immobilized onto a sensor chip CM5 (GE Healthcare) according to the manufacturer’s instructions. Then, four different concentrations were prepared (5, 2.5, 1.25, and 0.625 μM) and injected over the immobilized sensor chip (GE Healthcare) surface to monitor the protein–ligand interaction. SPR data were determined using the 1:1 binding model in Biacore T200 evaluation software.

The CEA+ cell lines HT-29, MKN-45, and CEA-HeLa229 were bought from American Type Culture Collection (ATCC). These cell lines (HT-29 and MKN-45) express high levels of CEA and were used to study Z_CEA binding selectivity toward the target protein. All cell lines were cultured in either RPMI-1640 medium or Dulbecco’s modified eagle medium (DMEM), supplemented with 10% fetal bovine serum and antibiotics (penicillin-streptomycin100 units/mL and 0.1 mg/mL) that were bought from Gibco. According to the supplier’s recommendation, the number of passages for cell lines were limited to five to eight passages after purchase.

To assess the targeting specificity of Z_CEA affibodies in vitro, an indirect immunofluorescence analysis was performed. In brief, HT-29, MKN-45, and HeLa229 cells were seeded in a 24-well plate at a density of 2 × 10³ cells/well, stored in 5% CO₂, and incubated at 37°C. Then, the cells were incubated with 100 μg/mL of Z_CEA affibodies or Z_WT for 3 h. The cells were washed with PBS, fixed with 4% paraformaldehyde, permeabilized in PBS containing 0.3% Triton X-100 (Sigma-Aldrich, St. Louis, MO) for 10 min at 37°C, blocked with blocking buffer for 2 h, and then incubated with mouse anti-His monoclonal antibody at 4°C overnight. Subsequently, the cells were stained for 1 h at 37°C with FITC-conjugated goat anti-mouse IgG (H + L). Cell nuclei were counterstained with Hoechst 33342 (blue) (Beyotime Biotech Co. Ltd., China), and the images were visualized using a confocal fluorescence microscope (Nikon C1-i, Japan).

A variety of human gastric tissues (three to five cases) were obtained from surgically removed specimens, fixed with 4% formalin for 24 h, embedded in paraffin, and stored at 4°C. All experimental protocols were reviewed and approved by the Ethical Committee of the First Affiliated Hospital of Wenzhou Medical University. After routine deparaffinization and rehydration, slides were immersed in 3% H₂O₂ for 10 min, and the antigen was retrieved by heating the slides in 0.01 M sodium citrate buffer (pH 6.0) at 96°C for 15 min. The slides were blocked with 5% normal goat serum (Cell Signalling Technology) for 1.5 h at 37°C. Then, the slides were probed with affibodies at 4°C overnight and were incubated with the anti-His monoclonal antibody for 0.5 h at 37°C, followed by horse radish peroxidase (HRP)-conjugated goat-anti mouse antibody for 0.5 h at 37°C. Polyclonal rabbit anti-CEA serum (prepared in-house) was used as a positive control. Finally, the slides were stained with DAB and counterstained with hematoxylin.

The in vivo biodistribution and targeting ability of Z_CEA affibody was detected by near-infrared fluorescence imaging. Female Balb/c nude mice aged 4- to 6-weeks and purchased from the Nanjing Biomedical Research Institute in China were used to establish the HT-29, MKN-45, and HeLa229 mouse xenograft models. Then, we subcutaneously injected the nude mice in the right forearm region with 1 × 10⁷ cells/100 μL PBS (n = 5 per group). When the tumor volume reached approximately 200–500 mm³, mice were taken for imaging. DyLight755 (Thermo Fisher Scientific) has an excitation peak of 754 nm and an emission peak of 776 nm, which was used to label Z_CEA546 and Z_WT affibody according to the manufacturer’s instructions. Then, 100 μmol of Dylight-755-labeled Z_CEA546 or Z_WT dissolved in 150 μL PBS were injected intravenously through the tail vein under brief anesthesia (1.4% isoflurane for 5 min) and imaged using NIR imaging system (CRi Maestro 2.10, Perkin Elmer, Waltham, MA) at different time points (5 min, 30 min, 1 h, 1.5 h, 2 h, 4 h, 6 h, 8 h, and 24 h). After injection, tumor/skin tissue signal intensity was analyzed at different time points using GraphPad Prism software. Then, the tumor-bearing mice were sacrificed to collect the main organs such as the heart, liver, spleen, lung, and kidney, and their fluorescence intensity were measured as describe above.

Two-tailed, single Student’s t-tests were performed to determine statistical significance between groups, with p < 0.05 denoting statistical significance. All graphs were acquired using GraphPad Prism.

Three cycles of panning were performed to select the positive clones from the phage display library. Then the phage clones were screened using ELISA to identify high-affinity Z_CEA binding clones and were analyzed through DNA sequencing. Through sequencing, we were able to identify a total of 31 clones (31/120 or 25.8%). Three potential Z_CEA affibodies (Z_CEA539, Z_CEA546, and Z_CEA919) were highly expressed and purified. This affibody framework region is highly homologous but diverse in the helical regions (Figure 1A). The affibody DNA sequences were successfully inserted into the expression vector pET21a (+) with His6 tag fusion protein (Figure 1B). Then, the recombinant plasmids were transformed into E. coli BL21(DE3) and were induced by IPTG for protein expression (Figure 1C). The expressed proteins were detected by SDS-PAGE analysis (Figure 1D). Western blotting further confirmed that the fusion proteins could react with the monoclonal antibody specific to six histidine tags (Figure 1E).

SPR studies were carried out to measure the binding affinity of Z_CEA539, Z_CEA546, and Z_CEA919 to the target protein CEA using BIAcore T200 biosensor. Z_CEA affibodies of various concentrations (5, 2.5, 1.25, and 0.625 μM) were prepared to flow over the immobilized CEA target protein. Our results demonstrated a concentration-dependent increase in resonance signals and strong binding affinity of three selected Z_CEA affibodies to ligand (Figures 2A–C). In contrast, Z_WT did not show the protein–ligand binding interaction (Figure 2D). The sensogram shows the maximum concentration (5 μM) of ZCEA affibodies and Z_WT (Figure 2E). We then used BIAcore analysis software 3.0 (Biacore) to measure the binding kinetics of Z_CEA protein–ligand interaction, and sensograms were fitted to a 1:1 Langmuir model. The equilibrium dissociation constant (KD) for the binding of Z_CEA539, Z_CEA546, Z_CEA919, and Z_WT were 5.059 × 10⁻⁶ mol/L, 1.733 × 10⁻⁷ mol/L, 5.512 × 10⁻⁶ mol/L, and 1.628 mol/L, respectively (Supplementary Table S1). Our SPR results showed a strong high-affinity binding of Z_CEA affibodies to CEA target protein, while Z_WT did not display any binding interaction toward the target protein.

Immunofluorescence assays were employed to further confirm the specific binding of Z_CEA affibodies to tumor cells that highly express CEA protein. Bright green fluorescence staining was clearly observed at the cell membrane region of CEA+ cell lines (HT-29 and MKN-45) that were treated with Z_CEA affibodies (Figures 3A,B), while no bright green fluorescence staining was noticed in cells treated with Z_WT. Furthermore, CEA cell line (HeLa229) treated with Z_CEA affibodies or Z_WT did not show any fluorescence signal (Figure 3C). This result further confirms that Z_CEA539, Z_CEA546, and Z_CEA919 affibodies have strong binding affinity and specificity to target protein (CEA) expressed in HT-29 and MKN-45 cell lines.

Histological analysis of the tumor tissue was performed to further investigate the binding specificity of Z_CEA affibodies. Immunohistochemistry was performed on formalin-fixed paraffin-embedded human gastric cancer tissue. The Z_CEA affibodies (Z_CEA539, Z_CEA546, and Z_CEA919) stained tissue sections from human gastric cancer tissue [showed in brown color (upper layer)], which was similar to the staining pattern observed in polyclonal anti-CEA serum; nevertheless, human gastric cancer tissue stained with Z_WT and PBS did not exhibit any obvious signal (Figure 4). Furthermore, normal gastric tissue did not display any changes in sections incubated with Z_CEA affibodies and polyclonal anti-CEA serum (lower layer). Our result proved that Z_CEA affibodies bind specifically to native CEA expressed in the tumor tissue.

The in vivo biodistribution of Z_CEA affibodies is shown in Supplementary Figure S1. The Z_CEA affibodies were quickly distributed into most parts of the body within 30 min post-injection, excreted through the kidneys, and rapidly cleared from the body within 24 h.

We further explored the tumor targeting selectivity of Z_CEA546 affibody in CEA tumor-bearing mice. After injection of DyLight755-labeled Z_CEA546 into the tail vein of CEA-tumor bearing mice, fluorescence signals were observed at the site of the tumor 1 h after injection and maximum fluorescence intensity was observed at 4 h, which remained for 24 h (Figures 5A,B). However, in HeLa229 tumor-bearing mice, DyLight755-labeled Z_CEA546 tumor-specific fluorescence signal could not be detected in the xenograft model, and DyLight755-labeled Z_WT also did not show fluorescence signal in xenograft tumor models (Figures 5C–E).

To verify whether in vivo fluorescence imaging retention is only present at the tumor site, the mice were sacrificed to isolate tumors and major organs for ex-vivo analysis at 24 h post-injection. After intravenous injection of DyLight755-labeled Z_CEA546, strong fluorescence signals from the tumors derived from HT-29 and MK-45 cell lines were clearly detected; however, no obvious fluorescence signal was detected in HeLa229 cell line (Supplementary Figure S2).