2.1 Ligand categories

To date, high-throughput screening and display bioengineering have been extensively and effectively used to develop various peptide- and protein-based ligands for molecular imaging. As shown in Figure 1, these ligands can be broadly categorized as linear peptides, cyclic peptides, protein scaffolds, nanobodies, and antibodies.

Linear polypeptides (3–5 kDa) are short, small proteins composed of <50 amino acids attached by covalent bonds. They are characterized by ease of preparation and modification, high cellular uptake/tissue penetration, relatively fast pharmacokinetics, and safe immunological profiles, and furthermore, they can be identified using mature screening platforms, which are benefits for imaging applications [11].

Cyclic peptides, originally designed as alternatives to linear peptides, are mainly formed via disulfide bonds, unnatural amino acids, and carbonic anhydrase catalysis. Cyclic peptides generally have a higher stability and longer half-life than linear peptides. Until now, the Food and Drug Administration (FDA) has approved 20 cyclic peptides either for cancer therapy and diagnosis or as antibiotic and antifungal drugs [12].

Protein scaffolds (4–20 kDa), considered alternatives to antibodies, have a rigid tertiary structure and several randomized sites. Protein scaffolds are generally more stable physicochemically than polypeptides. Representative protein scaffolds, such as affibody, knottin, anticalin, adnectin, and DARPins®, have been extensively investigated in preclinical research for clinical imaging or therapy [13].

The nanobody (12–15 kDa) derived from the camelid antibody VHH domain has a long CDR3 region with a high affinity for hidden epitopes. The nanobody does not require pairing of heavy and light chains, making it suitable for biological display systems. Additionally, the nanobody has demonstrated high potential for drug delivery, disease therapy or diagnosis, in vitro detection kit development, and crystallization research [14].

Antibodies (150 kDa) and antibody derivatives, such as Fab, scFv, and minibody (25–75 kDa), are pioneering affinity ligands that play important roles in targeted therapy or detection. Currently, over 100 monoclonal antibodies and eight antibody fragments have been approved by the FDA for marketing or clinical trials [15].

2.2 High-throughput screening using biological display systems

Biological display systems have been systematically developed for more than 30 years and have been broadly used for high-throughput ligand selection and other fields (Figure 2) [16]. With gene recombinant technologies, the structure of natural ligands can be conveniently imitated and reshaped by incorporating functional tags, groups, and linkers. Additionally, biological display systems ensure strict expression regulation and easy library amplification at low cost as well as rapid information decoding of candidates via polymerase chain reaction (PCR) and sequencing. Currently, representative display platforms are phage, bacteria, yeast, ribosome, mRNA, and mammalian cell display systems [17, 18].

2.3 Signal detection

Monitoring the interactions between ligands and targets provides physical information for candidate selection. Generally, candidates from virus-based phage display platforms interact with solid-phase immobilized target molecules. Subsequently, enzymatic luminescence signals are detected and recorded via a microplate reader. Cell-based selection platforms, such as bacterial and yeast, are relatively convenient. Because monoclonal candidates presented on the cell surface are multivalent, the intensity of the fluorescence signal corresponds to the strength of the interaction between candidates and targets during fluorescence-activated cell sorting (FACS). More importantly, FACS directly collects single-cell signals, which leads to greatly reduced false positives compared with cascade amplification signals of cumbersome enzyme-linked immunosorbent assay (Elisa) [8]. Subsequently, surface plasmon resonance (SPR) or bio-layer interferometry (BLI) platforms are used to accurately analyze dynamic changes in the molecular interaction and measure the ligand affinity. More importantly, the microplates and microarray chips of libraries or target molecules can be integrated with the above physical detection instruments to establish a high-throughput screening platform, which enables rapid and automatic screening of thousands of samples for affinity toward one or more molecules in parallel [19].

3.1 Phage display

In 1985, Smith et al. were the first to present a 57-mer peptide on the surface of a filamentous phage via insertion of a foreign gene into coat protein III of the filamentous bacteriophage. The phage represents a perfect vector for exogenous protein presentation, and because phage display has various advantageous features, including easy library construction via genetic engineering, strict regulation of expression, and high infectivity and propagation of the phage, it has been further developed in many fields since its establishment [20].

Although many phage display systems and formats are available, the filamentous phage and Podoviridae phage are most commonly used because of the high compatibility of the long foreign fragment [21]. The M13 bacteriophage is a rod-shaped nonlytic phage with a 6.4-kB single-stranded genomic DNA and 11 functional proteins. The fusion-protein length and amount may prevent phage rearrangement, infectivity, and propagation if a foreign gene is directly inserted in the genome. The phagemid vector and helper phage combination display modality (3 + 3 and 8 + 8) can be devised for independent display without interference to the phage functionality. The high-copy PVIII capsid protein, which is used for multivalent and small proteins, and the low-copy PIII capsid protein, which is used for low valency and large proteins, are the main presentation display systems of filamentous phage platforms. Because of the differences in the presentation valency, PVIII and PIII systems tend to generate high avidity and affinity ligands, respectively [22]. T7 is a lytic phage with an icosahedral capsid head and a 40-kB double-stranded genomic DNA. A fusion protein gene is commonly inserted at the c-terminal of coat protein gp10B for surface display [23].

The process of hit discovery follows library design and construction, phage recovery, bio-panning with targets, and function testing and sequencing of candidates. As the most popular system for peptide and protein-based ligand development, phage display has achieved great success not only for imaging but also for therapeutic applications [21].

3.1.4 Nanobody

The camelid antibody-derived nanobody is an appealing targeting moiety for imaging and diagnosis owing to its small molecular size, high thermostability, and high affinity. Generally, genes of natural or immunized nanobodies can be directly cloned into a phagemid vector for nanobody library construction. However, humanization optimization may necessitate lowering the immunogenicity at the expense of structural stability [48]. As an alternative, a synthetic nanobody library, with a larger capacity and lower cost, does not require animal immunization and is compatible with toxic and pathogenic antigens. The prerequisites for constructing a synthetic library are a stable scaffold and specific coding complementarity-determining regions (CDRs) for further gene cloning. In this respect, the most intensively investigated scaffolds are mature cAbBCII10 and NbBCII10FGLA, which are already humanized and very stable, with no need for disulfide bonds. Moreover, the length polymorphism of CDR3 expands the library diversity, and the CDR3 coding sequence (3–28 aa) abides by the native nanobody preference to balance the diversity while improving structural stability, solubility, and correct folding [49].

Recently, Li's group identified a CDH17-binding nanobody, E8 (K_D = 70.3 nM), using the native nanobody phage library. E8-IR800 emits the strongest fluorescence signal at the tumor site of an MKN45 mouse model 12 h after injection, which is nearly four-fold greater than that observed in the control group (Figure 3). The E8-PE38 toxin complex also efficiently inhibits tumor growth and prolongs mouse survival [50]. Another PD-L1-specific nanobody, Nb16 labeled with Cy7, exhibits selective and high tumor accumulation in vivo, and the ex vivo imaging signal of this nanobody is 2.7-fold greater than that of the PD-L1 monoclonal antibody [51].

3.1.5 Antibody fragments

The Geyer group designed a phage display Fab library with a framework derived from the HER2 monoclonal antibody trastuzumab with diversified CDR L3 and H3 [52]. Using this library, they acquired an HER3-targeting Fab, which was reshaped to scFv, diabody, scFv-CH3, scFv-Fc, and IgG for further IRDye800CW labeling. In vivo imaging evaluation of these six antibody-based probes demonstrated that the molecular weight difference influences the bio-distribution and clearance time, while the affinity is improved with the bivalent fragment. The smallest probe, scFv, emits a single peak at 4 h post injection and is cleared at 6 h post injection. The other probes emit signals with a minimal background fluorescence at 24 h post injection, and IgG and scFv-Fc emit the strongest signals, which last for 72 h [53].

Using a naïve phage display library, Lebeau's group also selected a FAP-specific scFv, which was engineered to a full-length antibody, B12 IgG. In a prostate cancer xenograft model, B12 IgG-IRDye-800cw emits a specific signal that is more than three times greater than that of the control group 24 h after injection [54]. Another phage-selected CD44-binding scFv was evaluated using PET [55]. Specifically, bivalent scFv-Fc-CD44 was generated with a more than 200-fold improvement in affinity. [⁸⁹Zr]-DFO-scFv-Fc-CD44 localizes to the tumor tissue 1 day after administration with a high uptake (~56% injected dose/g), and the SUV_max ratio (scFv-Fc-CD44 vs. control) increases two- to seven-fold from days 1 to 7 (Figure 4).

3.2 Bacterial display

The phage display system is limited by the potential interference of helper phages, which accounted for 10% in secreted virions, long panning cycle, and lack of ability to monitor the binding strength during the selection process. To tackle these limitations, the bacterial surface display was invented by anchoring ligands to cell membrane proteins.

The OMP and LPXTG motif fusion systems are the mainstream surface platforms for Gram-negative Escherichia coli and Gram-positive staphylococcus bacteria, respectively. Generally, the ligand library gene is inserted into bacterial display vectors and expressed, which is stringently controlled by the arabinose promoter. After incubation and FACS, the candidate genes are then extracted and transformed to new host cells for the next round selection [18]. The biggest advantage of the bacterial surface display is the multivalent presentation, which enables monitoring of the ligand-binding strength via the fluorescence intensity during FACS for better ligand affinity maturation [56]. However, large ligands may display poorly because the two-layer membranes function as a barrier, and fusion ligands may reduce outer membrane integrity, cell vitality, and library size [20]. Despite these disadvantages, full-length antibody presentation of the NlpA-ZZ fusion protein via the bacterial display has been reported [57]. The bacterial surface display is mainly applied to improve ligand affinity for molecular imaging.

Skerra's group employed the bacterial display for the initial selection of human HSP70-binding anticalin primary candidates (K_D = 13 nM). They constructed a mutated bacterial library via error-prone PCR and performed another cycle of selection, which produced the BGG10I clone (K_D = 391 pM), with an affinity improvement of nearly 33-fold. The specific PET signal was detected at the tumor site 24 h after the injection of [⁸⁹Zr]-labeled Anticalin-PAS200 [58]. Löfblom's group combined phage and staphylococcal display systems and selected a HER3 affibody Z₀₅₄₁₆ with nanomolar affinity. Next, they conducted alanine scanning and established a second-generation staphylococcal surface display library. Z₀₈₆₉₈ with an affinity of 21 pM was obtained via FACS [59, 60]. PET of [⁶⁸Ga]-(HE)3-Z₀₈₆₉₈-NODAGA with a hydrophilic tag has an increased tumor uptake but lowered blood and hepatic uptakes in BxPC-3 xenografts [61].

Löfblom's group used the same strategy to obtain several VEGFR2-binding affibodies with nanomolar affinity, which were then engineered to biparatopic affibody construct Z_{VEGFR2_Bp2} (Z_{VEGFR2_22}-(S₄G)-Z_{VEGFR2_40}-(S₄G)₃-ABD₀₃₅), which has an inhibitory potency similar to that of ramucirumab (Figure 5a). [¹¹¹In]-NODAGA-Z_VEGFR2-Bp2, with picomolar affinity (K_D = 33 ± 18 pM), emits a strong SPECT signal in the MS1 tumors of mice at 2 h post injection (Figure 5b). Moreover, the signal ratio of tumor-to-blood, tumor-to-muscle, and tumor-to-brain reached 11, 15, and 78, respectively. An intracranial imaging signal was observed in the glioblastoma orthotopic model at 2 h after the injection of 4 or 40 μg of [¹¹¹In]-NODAGA-Z_VEGFR2-Bp2 and 4 μg of [¹¹¹In]-NODAGA-Z_tag-Ztaq. The 4-μg Z_VEGFR2-Bp2 group shows a two-fold higher tumor-to-cerebellum signal ratio than the others, and the specificity is consistent with the results of macroautoradiography and hematoxylin and eosin staining (Figure 5c,d) [62-64].

3.3 Yeast surface display

The yeast surface display emerged in 1997 as an alternative to the phage display to mitigate the disadvantage of the lacking PTM (post-translational modification) of the prokaryotic system. Following the discovery of glycosylphosphatidylinositol anchored surface proteins, including Agα1p, Aga2p, flocculin Flo1p, Cwp1p, Cwp2p, Sed1p, Tip1p, YCR89w, and Tir1, two typical yeast platforms based on agglutinin Aga1–Aga2 and flocculin Flo1p were established. Commonly, a ligand library gene is cloned into the following sites: (i) either or both terminals of the Aga2 gene of Saccharomyces cerevisiae, (ii) the C-terminal of the adhesive domain of the Flo1p gene, or (iii) the N-terminal of the Flo1p gene, the C-terminal of which contains a GPI anchoring gene [65]. In contrast to the phage display, the yeast display technique not only simplifies the library preparation and quantification but also reduces the selection cycle and improves the accuracy with FACS compatibility. However, the yeast growth cycle is longer than the bacterial growth cycle, and over-glycosylation may impair protein folding and function. Although the yeast display platform is mainly applied to affinity maturation of the antibody–antigen interaction, development and optimization of imaging ligands based on this system have come to the fore [66].

Hackel's group identified a CD276-binding affibody, ABY_B7-H3 (K_D = 310 ± 100 nM), from a yeast display library, the affinity of which was further improved to 0.9–20 mM by employing a helix-walking strategy. The ABY_B7-H3 ultrasound imaging probe modified with microbubbles (MB), MB_ABY-B7-H3, emits a strong in vivo signal (8.4 ± 3.3 a.u.) in breast orthotopic tumors, which is nearly five-fold greater than that observed in the control group [67]. They also engineered an EGFR-binding GP2 molecule via the yeast display platform and FACS. The optimized ligand, GαE35, exhibits an improved charge distribution and high affinity (K_D = 8.8 ± 0.6 nM). ⁶⁴Cu-NODAGA-GαE35 accumulated in A431 cells emits a good PET signal (4.7 ± 0.5%ID/g), and it has a low unspecific uptake of 0.6 ± 0.1%ID/g by MDA-MB-435 cells [68].

Abbott's group developed a yeast display library and identified the scFv-C9 antibody, which specifically binds to bisecting N-glycans of tumor cells. An in vivo study of scFv-C9 magnetic bead complexes showed a selective and relatively stable signal in A1847 xenograft tumors [69]. Willmann's team employed yeast surface display and FACS to select a Thy1-targeting scFv to prepare an ultrasound imaging probe. MBThy1-scFv has a high signal in the orthotopic pancreatic ductal adenocarcinoma (5.32 ± 1.59 a.u.) of mice, which is nearly nine times greater than that in the normal pancreas (0.67 ± 0.71 a.u.) [70].

Cochran et al. constructed a yeast surface display library of knottin peptides based on the EETI-II 2.5D scaffold, and they isolated a predominate candidate, 3–4C, with an affinity of 2 nM toward αvβ3 integrin after 9 rounds of FACS (Figure 6a). Knottin 3–4C was further reshaped to 3–4A with two RGD integrin binding loops. Compared with the parent knottin, 3–4A shows a nearly three-fold increase in binding ability to U87 cells, two-fold increase in binding ability to K562-αvβ5 cells, and similar affinity to K562-αvβ3 cells. Moreover, ⁶⁴Cu-DOTA-knottin 3–4A has properties that are favorable for imaging with a rapid tumor uptake of 3.51% ± 0.83%ID/g at 1 h post injection, tumor-to-blood signal ratio of 27 ± 3, and tumor-to-muscle signal ratio of 31 ± 7 at 4 h post injection in U87MG xenograft mice [71]. Another ⁶⁸Ga-NOTA-3-4A PET probe was further evaluated for its ability to diagnose vulnerable atherosclerotic plaques. This probe has a high specific plaque uptake of 6.67 ± 1.44 and a favorable plaque-to-normal artery signal ratio of 15.88 at 1 h post injection in a mouse model (Figure 6b–e) [72].

Gambhir et al. also developed an αvβ6-binding knottin, R₀1-MG, via yeast surface display selection. No adverse effects were observed for healthy human volunteers who were administered [¹⁸F]FP-R01-MG-F2, which mainly cleared via the kidneys with low lung and liver uptakes (SUV_mean  <  1) (Figure 7a). Moreover, in female patients with pancreatic cancer, [¹⁸F]FP-R01-MG-F2 could be detected throughout the tumor (ROI = ~8000 mm³, SUV_mean = 6.2) and had a low liver uptake (SUV_mean = 0.9). [¹⁸F]FDG only localized to the biliary stent margin (ROI = ~3000 mm³, SUV_mean = 4.1) and showed a liver uptake (SUV_mean = 2.9). Therefore, the knottin-based PET probe has good specificity and high contrast image for tumor diagnosis (Figure 7b,c). Additionally, in a pilot study of six patients with idiopathic pulmonary fibrosis (IPF), the [¹⁸F]FP-R01-MG-F2 probe demonstrated a potential value for the detection of IPF, although more histological data are needed for further verification [73].

3.4 Ribosome display

The ribosome display method was initially developed in 1994 by Mattheakis and coworkers to solve the low transformation efficiency of the phage display system [74]. The ribosome display system comprises three components, namely library gene template synthesis, an in vitro transcription–translation system, and affinity selection. Specifically, the library DNA cassette contains the ligand gene, and the C-terminal without a stop codon ensures the formation of the mRNA-ribosome-ligand complex. The prokaryotic E. coli S30, eukaryotic wheat germ, and rabbit reticulocyte lysate systems are generally used for ligand production with a selection process similar to that of the phage system. mRNA is then eluted and reverse transcribed into cDNA for the next round of library preparation or sequencing [75].

Because the ribosome display is independent of a cell expression system, it excludes the gene transformation step, thereby efficiently expanding the library size and eliminating the amplification bias from the difference in cell growth rates. Moreover, toxic or protease-sensitive ligands are produced, which further ensure the library diversity, while site-directed mutation can be easily realized via error-prone PCR for further molecular evolution. However, some obstacles remain. For example, there is a need to reduce the mRNA instability via an RNAse inhibitor and terminal hair-pin structures. Moreover, the ribosome of the complex may compete with the ligand to interact with the target molecule, and a ligand with a large molecular size may not be efficiently presented [76].

Currently, the ribosome display is popular for scFv selection, and some protein scaffolds developed for imaging have been reported [77]. For example, the thrombus plays a significant role in stroke. Jaroslav's group selected a fibrin clot-binding ABD molecule, D7, using a ribosome display platform and demonstrated the strong binding between fibrin fibers and D7 molecules using confocal and electron microscopy. The D7H2 liposome can penetrate an in vitro flow thrombus model to a depth of 50–100 μm after a 20-min incubation, demonstrating the potential value of D7 for in vivo diagnostic imaging of thrombi [78].

The inflammatory biomarker VCAM-1 is widely expressed on the endothelial cell surface. Kaufmann's team recently reported VCAM-1-binding DARPins® molecules selected from a ribosome display. Compared with MB, MB_{1732_F8} and MB_{1730_C7} can produce ultrasound imaging signals with a four-fold enhancement in murine hind-limb inflammation models [79].

3.5 mRNA display

The mRNA display system was reported in 1997 by the Roberts group. The selection on the mRNA display platform is similar to that on a yeast display system. Specifically, the mRNA library comprises synthetic DNA templates (T7 promoter-start codon-ligand gene-deoxyadenosine spacer). mRNA is acquired after in vitro translation and then modified with an aminoacylated tRNA mimic puromycin to the mRNA C-terminal for gene–protein phenotype connection, and the mRNA-linker-protein mixture is then incubated with target molecules, followed by the regular screening process [80].

The puromycin linker is too small to influence ligand–target interactions, and the gene–protein phenotype linkage ensures stringent selection and decreases the diversity loss. Unnatural amino acids or chemical modification are also suitable for functionality improvement and structure reshaping [81]. Although the mRNA display platform has been rarely used for imaging applications, ligands selected from an mRNA display platform have been developed for biomedical applications [82].

By selecting a PD-L1 binding affibody, M1, with nanomole affinity, the Millward group recently showed that the M1-Cy5 fluorescent probe can localize to the tumor site after 1 h, with a signal that is nearly three times stronger than that of the HER2 affibody probe, in an EL4 tumor model. The M1 affibody can also inhibit the PD-L1 signaling pathway at micromolar concentrations, further demonstrating its therapeutic potential. Millward and coworkers also selected the HER2 affibody and autophagy protein LC3 targeting ligands [83].

Robert's team integrated an unnatural amino acid into an mRNA display library to improve the stability of the ligand scaffold. They selected a HER2-targeting cyclic peptide, SUPR4 (K_D = 76 ± 30 nM) with 500- and 3700-fold improvement in serum stability and protease resistance, respectively. The SUPR4-Cy5 probe gradually accumulates in the tumor from 4 to 24 h and has a low hepatic uptake at 48 h in the subcutaneous HER2-positve tumor model [84]. The group also used the same strategy to develop a PD-L1 targeting linear peptide, SPAM (K_D = 67 nM), with a binding site that overlaps with that of natural PD-1 and Atezolizumab, although the in vivo study remains to be conducted [85].

In addition to ligands targeting receptors, ligands active against other targets can be developed using mRNA display platforms. For example, Matsumoto's group recently designed a macrocyclic peptide mRNA library and selected HIP-8 to inhibit the hepatocyte growth factor. ⁶⁴Cu-labeled HiP-8-PEG11 rapidly localizes to the tumor within 20 min, with clearance from the liver, for tumor imaging at 90 min post injection in HGF knock-out tumor-bearing mice [86].