Alternative Non-Antibody Protein Scaffolds for Molecular Imaging of Cancer

Abstract

The development of improved methods for early detection and characterization of cancer presents a major clinical challenge. One approach that has shown excellent potential in preclinical and clinical evaluation is molecular imaging with small-scaffold, non-antibody based, engineered proteins. These novel diagnostic agents produce high contrast images due to their fast clearance from the bloodstream and healthy tissues, can be evolved to bind a multitude of cancer biomarkers, and are easily functionalized by site-specific bioconjugation methods. Several small protein scaffolds have been verified for in vivo molecular imaging including affibodies and their two-helix variants, knottins, fibronectins, DARPins, and several natural ligands. Further, the biodistribution of these engineered ligands can be optimized through rational mutation of the conserved regions, careful selection and placement of chelator, and modification of molecular size.

Introduction

Molecular imaging can provide critical clinical information regarding the presence, concentration, and localization of cancer biomarkers in vivo. The data, which can be obtained dynamically or longitudinally if desired, empowers early detection, patient stratification, and treatment monitoring. A major challenge in this field is the development of high affinity, specific ligands for the multitude of important biomarkers. Directed evolution and other strategies position engineered proteins to offer a robust, high throughput means for ligand generation (Figure 1). Many of these engineered proteins have shown promising results in preclinical evaluation as well as early clinical results. While antibodies and their fragments have been explored [1], their large size and slow clearance are not ideal due to the functional requirement of reduced background signal necessary for imaging contrast; moreover, for targeting of poorly vascularized tumors, such as those in early formation, large size slows extravasation and delivery [2]. For these reasons, as well as benefits in stability, production, and chemical conjugation, alternative protein topologies have been studied as scaffolds for molecular imaging. Short linear and cyclic peptides have exhibited significant success as molecular imaging agents [3], although their robustness in engineering high affinity toward novel targets is limited. Thus, this review will focus on advances in molecular imaging of cancer, particularly positron emission tomography (PET), single-photon emission computed tomography (SPECT), and gamma-camera imaging using engineered, folded, non-antibody proteins (Figure 2, Table 1). The majority of the research has been performed using subcutaneous xenografted tumors in mice. There are exceptions to these experimental systems, including clinical data, which will be noted.

Figure 1.

Schematic for molecular imaging agent development. Several scaffolds have been validated for in vivo molecular imaging. Selection of an appropriate scaffold, followed by directed evolution – for binding and delivery characteristics – and radiolabeling yield an imaging agent with potent delivery and retention properties capable of producing high contrast molecular images.

Figure 2.

Cartoon structure with surface representation of protein scaffolds used in molecular imaging applications. Variable regions are highlighted in red. Conserved portions are shown in grey. A: affibody (PDB: 2B88); B: knottin (1HYK); C: fibronectin (1TTF); D: two-helix affibody (modified from three-helix affibody 2KZJ); E: DARPin (2JAB); F: natural ligand (EGF) (2KV4).

Table 1. Characteristics of non-antibody molecular imaging agents.

Name	Topology	Amino Acids	Typical Paratope
Affibody	triple α-helical bundle	58	surface of helices 1 and 2 (13 amino acids)
Knottin	knotted 3-disulfide core	30 - 50	loop 1
Fibronectin	β sandwich	94	1-3 loops (10-24 amino acids)
DARPin	4-6 repeats of: β turn + two α helices	130 - 200	β turn and first α helix (7 amino acids)
Two-Helix Affibody	2 α-helices	35	surface of helices 1 and 2 (13 amino acids)
Natural Ligands	e.g. VEGF, EGF, annexin V	variable	natural interface

EGF(R): epidermal growth factor (receptor)

EpCAM: epithelial cell adhesion molecule

HER2: human epidermal growth factor receptor type II

IGF1R: insulin-like growth factor type I receptor

VEGF(R): vascular endothelial growth factor (receptor)

Protein-Based Molecular Imaging Agents

Affibody

The affibody is a 58 amino acid, three helical bundle [4]. Typically, randomization of 13 amino acids on the surface of helices 1 and 2 is used to generate novel binding ligands. The most extensively studied class of affibodies is those targeting human epidermal growth factor receptor 2 (HER2). The second generation HER2-binding affibody, Z_HER2:342, was engineered to bind with 22 pM affinity [5], and has successfully imaged HER2-expressing tumor xenografts in mice when labeled with ¹²⁵I [5], ¹¹¹In [6], ^99mTc [7], ¹⁸F [8], ^114mIn [9], ¹²⁴I [10], ⁶⁸Ga [11], and ¹¹C [12]. Tumor targeting is effective across many mouse models with a median tumor uptake of 9 %ID/g at 4 h (range: 1 – 26 %ID/g), tumor-to-blood ratio median of 31 (range: 1.5 – 187), and tumor-to-muscle ratio median of 61 (range; 3 – 650). Kidney uptake is generally high with a median of 111 %ID/g (range 2 – 324 %ID/g). Hepatic retention is minimal with a median of 1.5 %ID/g (range 0.2 – 19 %ID/g). Z_HER2:342 has been extensively modified to maximize imaging contrast and decrease signal in clearance organs. Clinical translation has been initiated as one derivative was labeled with ¹¹¹In and ⁶⁸Ga for SPECT and PET imaging of metastatic breast cancer patients [13]. The tracers effectively imaged metastases and were well tolerated.

Three different anti-epidermal growth factor receptor (EGFR) affibodies (0.9 – 50 nM affinity) labeled with ¹¹¹In identified tumors using a gamma camera, exhibiting 2.4-3.4 %ID/g tumor with tumor-to-blood and tumor-to-muscle ratios of 7-15 and 18-27 [14-16]. One of these affibodies was labeled at a unique Cys residue with ⁶⁴Cu via a 1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid (DOTA) chelator. PET imaging in xenografted mice exhibited 12±2 %ID/g tumor but only 1.04±0.01 tumor-to-blood at 4 h. Pre-injection of 50 μg cold affibody elevated tumor uptake to 17±4 %ID/g with 10±3 tumor-to-blood and 49±11 tumor-to-muscle [17]. Separately, this affibody was labeled with an ¹⁸F precursor. PET imaging exhibited 8±1 %ID/g tumor with 2.6±0.5 tumor-to-blood and 37±13 tumor-to-muscle at 3 h [18].

An affibody engineered to 0.5 nM affinity for insulin-like growth factor type I receptor (IGF1R) labeled with ¹¹¹In yielded 1.3±0.1 %ID/g tumor with 2.5±0.2 tumor-to-blood and 4.3±1.0 tumor-to-muscle at 4 h [19]. However, no significant contrast was observed for tumor relative to pancreas, spleen, stomach, and colon. Contrast was modestly improved using ^99mTc labeling via an HEHEHE purification tag [20].

Two-helix affibody

A 36 amino acid two-helix derivative has been explored as a smaller alternative to the affibody via removal of the third helix. The resulting destabilization was partially compensated by stabilizing mutations and disulfide bonding [21,22]. The two-helix affibody has only been able to achieve low nanomolar affinity to date [21]. The two-helix affibody has been applied to molecular imaging in HER2-expressing tumor xenograft models successfully when labeled with ⁶⁸Ga [23], ¹⁸F [24], and ¹¹¹In [25]. In a comparative study, the two-helix affibody (2 nM affinity) showed higher tumor-to-blood ratio than the parental affibody (78 pM affinity) [25]. However, the tumor uptake of the two-helix derivative was 40% lower than the parental clone. Further stabilization and affinity maturation could make the two-helix affibody an advantageous option for high-contrast molecular imaging.

Knottin

Knottins are 30-50 amino acid polypeptides containing three disulfide bonds that form a knotted structure [26]. Binding motifs have been grafted into one to two loops to provide binding functionality. Randomizations of loop lengths, motif position, and surrounding amino acids have been used for affinity maturation. An integrin-binding motif was grafted into Ecballium elaterium trypsin inhibitor II and evolved for mid-nanomolar affinity to integrin α_vβ₃ with cross-reactivity for integrins α₅β₁ and α_vβ₅. ⁶⁴Cu radiolabeling via DOTA enabled PET imaging with 4.5±1.2 %ID/g tumor, 6±1 tumor-to-blood, and 17±8 tumor-to-muscle at 1 h and 4.0±0.7 %ID/g tumor, 42±9 tumor-to-blood, and 21±13 tumor-to-muscle at 4 h [27]. Renal retention was 4±1 %ID/g at 1 h while hepatic signal was 2.2±0.2 %ID/g. Minimal degradation was observed in blood, minor breakdown in tumor, and >50% degradation in kidneys. A knottin variant with six loop mutations, 2.5D, was then labeled with ¹⁸F so that radioisotope kinetics would better match pharmacokinetics [28]. Effective tumor imaging was observed with 2.6±0.7 %ID/g tumor at 0.5 h. Kidney and liver uptake were similarly decreased relative to ⁶⁴Cu-knottin values. In a comparative study in a transgenic mouse model with nascent lung tumors, the ⁶⁴Cu-knottin tracer exhibited superior tumor-to-lung ratios (6.0±0.6 versus 4.4±0.7) relative to ¹⁸F-fluorodeoxyglucose [29]. The 2.5D knottin was also dually labeled with ⁶⁴Cu and near infrared fluorophore as a multimodality probe, which improved tumor retention but elevated renal and hepatic signal [30].

Integrin-binding motifs were also grafted and evolved to mid-nanomolar affinity in the Agouti-related protein knottin to yield clone 7C. ⁶⁴Cu labeling enabled effective PET imaging of xenografted tumors with rapid uptake including 2.7±0.9 %ID/g tumor at 1 h and 6.6±1.2 tumor-to-blood and 17±5 tumor-muscle by 2 h [31]. Kidney retention was higher (60±18 %ID/g at 2 h) than with the Ecballium elaterium trypsin inhibitor II scaffold. ¹¹¹In labeling of this knottin yielded 5.7±1.6 %ID/g tumor at 0.5 h but rapid reduction in tumor signal (2.5±0.4 %ID/g tumor at 2 h) [32]. Renal retention was improved relative to the ⁶⁴Cu probe (34±8 %ID/g). ¹⁸F radiolabeling via ¹⁸F-fluoropropionate yields effective PET imaging with a more translatable isotope: 2.5±0.2 %ID/g tumor, 20±2 %ID/g kidney, 7±1 tumor-to-muscle, and 4±1 tumor-to-blood at 1 h [33].

The breadth of knottin scaffolds and their targets was expanded as Momordica cochinchinensis trypsin inhibitor II was grafted with integrin α_vβ₆-binding motif and evolved to 3-6 nM affinity [34]. ⁶⁴Cu labeling yielded 4.3±0.7 %ID/g tumor at 1 h with 9±2 tumor-to-muscle but 75±5 %ID/g kidney. Grafting this loop onto a serine-rich Lens culinaris trypsin inhibitor knottin reduced renal retention to 18±4 %ID/g, although tumor uptake was reduced two-fold. This tracer also effectively imaged orthotopic pancreatic tumors. Radiolabeling with ¹⁸F yielded effective PET imaging as early as 0.5 h with significant decline in renal retention [35]. In further evidence of the breadth of this topology, fluorophore-conjugated agatoxin was successfully able to image integrin α_vβ₃-expressing U87MG glioblastoma xenografts in mice [36].

Fibronectin Domain

The tenth type III domain of human fibronectin (also known as monobody or Adnectin) is a 94 amino acid β-sandwich protein. Diversification of 10-24 amino acids in one to three solvent-exposed loops imparts novel binding activity [37,38]. Recently, the fibronectin scaffold was validated for in vivo molecular PET imaging in mouse tumor xenograft models [39]. PET imaging with a ⁶⁴Cu-DOTA conjugated EGFR-binding fibronectin showed good tumor uptake (3.4±1.0 %ID/g at 1 h), retention (2.7±0.6 %ID/g at 24 h), and contrast (8.6±3.0 tumor-to-muscle ratio, 8.9±4.7 tumor-to-blood ratio). Dynamic PET indicated effective contrast was present as early at 15 minutes post-injection. High renal retention was partially alleviated by ¹⁸F radiolabeling. Biodistribution was further modified by mutation of hydrophobic and charged residues [40]. A ⁶⁴Cu-DOTA-fibronectin specific to CD20 has also been developed and tested in a humanized transgenic mouse model for B-cell imaging (Arut Natarajan, BJH, Sanjiv Gambhir, unpublished). Resulting PET images showed encouraging results for the application of this fibronectin in imaging non-Hodgkin's lymphoma. Both spleen uptake and spleen-to-blood ratio for the ⁶⁴Cu-DOTA-fibronectin were significantly higher than ⁶⁴Cu-DOTA-rituximab, showing that the biodistribution properties and imaging ability of the fibronectin domain are superior to monoclonal antibodies.

Designed Ankyrin Repeat Protein

The designed ankyrin repeat protein (DARPin) is composed of 4-6 repeat units of 33 amino acids making one β turn followed by 2 α-helices. Diversification of 7 amino acids per repeat unit yields novel binding activity [41]. ^99mTc(CO)₃-labeled DARPins with affinities between 90 pM and 270 nM have been applied to imaging HER2-expressing tumor xenografts in mice using SPECT/CT [42]. The tumor uptake ranged from 1.5±0.2 %ID/g for the weakest affinity to 9.1±1.8 %ID/g for the strongest affinity DARPin at 1 h. The tumor-to-blood ratio for the 90 pM agent is 27±15 at 4 h, however, kidney uptake is 239±33 %ID/g. Poly(ethylene glycol) (PEG) conjugation increased tumor uptake at late time points but reduced tumor-to-blood ratios; ratios above 1 were not observed until the 24 h post injection time point. Biodistribution for a fusion of Pseudomonas aeruginosa exotoxin A and an EpCAM-specific DARPin was determined in vivo with a Cy5.5-conjugated DARPin fusion construct [43]. Ex vivo analysis 48 h post injection shows the highest localization in the EpCAM-expressing tumor, providing greater signal than the clearance organs.

Natural Ligands

Targeting agents can also be derived from natural ligands as these molecules often have high specificity and affinity for their respective receptors. However, the major disadvantage of using these ligands for molecular imaging is their potential agonistic effects. Despite this, several natural ligands have been used. Vascular endothelial growth factor (VEGF) has been used to image angiogenesis in tumors using several different radioisotopes including ¹²³I [44], ⁶⁴Cu [45], and ⁶⁸Ga [46]. A single chain analogue of VEGF has also been used to image VEGF receptor-expressing tumor xenografts [47,48]. Radiolabeled EGF has been applied for imaging of EGFR-expressing tumors in mouse models, although high liver signal is a noted problem in multiple studies [49,50]. Radiolabeled annexin V has been utilized for treatment monitoring applications. Annexin V, which binds the apoptosis marker phosphatidylserine, has been used to assess the effectiveness of radiation therapy [51] and doxorubicin-based therapy [52]. Annexin V effectively predicted response to chemotherapy with 94% accuracy in separating responders from non-responders [53].

Designing Delivery

It is evident that disparate protein topologies can be engineered for in vivo molecular recognition and delivery of radioisotopes to tumors. As the field continues to mature, we should strive to identify the ideal molecule for the application of interest, to maximize tumor uptake, minimize off-target retention, and tune kinetics for efficacy, logistics, and patient safety. Quantitative design rules would be tremendously useful to optimize performance. Select studies have been performed that initiate this effort.

Molecular Size

As previously stated, small size aids permeability across the vascular wall [2,54], which speeds tumor uptake. It also speeds clearance, which is beneficial for low background but reduces the input function available for tumor targeting. A mathematical compartmental model predicts uptake and specificity are maximized by small proteins [2], which is in agreement with experimental data. A HER2-targeted affibody exhibited ≥10-fold greater tumor-to-blood relative to the antibody trastuzumab at 6, 24, and 72 h [10] (Figure 3). 7 kDa affibody monomers yielded a median of 3.4-fold higher tumor uptake than 15 kDa affibody dimers despite blood clearance that was comparable to the monomer (median 1.0-fold) [5,14,55-57]. A two-helix affibody derivative yielded 2.7-fold higher tumor-to-blood and 1.8-fold higher tumor-to-muscle at 1 h relative to the larger three-helix affibody albeit with a 40% reduction in tumor uptake [25]. A 15 kDa DARPin also exhibited 2.3-fold greater tumor signal at 1 h and similar tumor signal at 4 h relative to a 35 kDa conjugate with PEG despite an increased blood concentration for the PEGylated domain [42].

Figure 3.

Female BALB/c nu/nu mice with HER2-expressing NCI-N87 xenografts were imaged via PET using ¹²⁴I-PIB-Z_HER2:342 affibody (A-C) and ¹²⁴I-PIB-trastuzumab (D-F). Mice were sacrificed and imaged 6 (A and D), 24 (B and E), and 72 h (C and F) post injection. Urinary bladders were removed before scanning. Figure reproduced.[10]

Affinity

The affinity of the engineered probe for its target affects retention in the tumor site. ¹¹¹In-DOTA-affibodies engineered to 117 pM, 157 pM, and 3.8 nM affinities for HER2 were compared in SKOV-3 (HER2^high) and LS174T (HER2^low) tumor xenograft models [58]. In the SKOV-3 model, tumor uptake was indistinguishable at 4 h (13±3, 16±7, and 14±1 %ID/g respectively), but at 24 h, the tumor signal was 2-fold higher for the picomolar affinity affibodies. In the LS174T model, the tumor uptake for the 3.8 nM affinity was only 1.7±0.2 %ID/g at 4 h compared to 7.4±1.6 and 10.8±1.3 %ID/g for the picomolar affibodies, respectively. ^99mTc-DOTA-DARPins with 90 pM and 1.5 nM affinities for HER2 in SKOV-3 model showed similar results, with tumor uptakes being indistinguishable at 4 h, but the 90 pM affinity showing better retention at longer time points [42]. The tumor uptake of these DARPins became distinguishable when a 10 nM DARPin is compared to the 90 pM DARPin over all time points.

Hydrophilicity

Four fibronectin domains were engineered with a range of hydrophilicities to study the impact on biodistribution. Renal retention was observed to positively correlate with hydrophilicity whereas hepatic retention was inversely correlated [40]. Moreover, using phylogenetic sequence frequency, a metric was identified to select appropriate mutations to introduce hydrophilicity. Thus, protein hydrophilicity is an addressable modulator of physiological distribution. The decision between renal and hepatic retention can be made based on the application; else, hydrophilicity can be introduced to minimize hepatic retention and renal retention can be reduced by other means.

Molecular Charge

Modification of molecular charge has impacted renal retention in numerous studies, although the impacts of total and net charge are unclear. Charge removal from cystine knots reduced renal retention [34]. Removal of two or three lysines from an affibody chelator eliminated 74 and 86 % of renal retention [59]; removal of two or three basic residues eliminated 91 and 84 %ID/g tumor [7]. Removal of six acidic and five basic residues from fibronectin decreased renal retention by 34% [40]. In a separate fibronectin clone, removal of four acidic and two basic residues also reduced renal retention by 34%; however, removal of one or two more acidic residues re-elevated renal retention.

Chelator

Both chelator structure and location of ligand conjugation can be impactful on in vivo biodistribution. The comparison of three ¹¹¹In-labeled macrocyclic chelators, DOTA, 1,4,7-triazacyclononane-N,N′,N″-triacetic acid (NOTA), and 1-(1,3-carboxypropyl)-4,7-carboxymethyl-1,4,7-triazacyclononane (NODAGA), using anti-HER2 affibodies showed NOTA to have a two-fold high liver signal than the others at 4 h [60]. NODAGA demonstrated the fastest clearance kinetics providing the highest tumor-to-blood imaging contrast whereas DOTA had a slightly higher tumor uptake due to longer plasma residence. In addition to ring structures, amino acid based chelators (3 residues in length) provide high labeling efficiency and may be chosen based on imaging background requirements. In particular, the kidneys and intestines are commonly sources for off-target uptake and a tradeoff may be made between the two with chelator choice. Positively or negatively charged chelators (Asp-Ser-Asp and Lys-Lys-Lys) showed high kidney signal (33 and 120 %ID/g) [59,61] whereas neutral chelators (Ser-Ser-Ser and Gly-Gly-Gly) reduced kidney uptake (19 and 6 %ID/g) [59,62]. The location of chelator conjugation is also impactful on biodistribution; C-terminal DOTA outperformed those placed N-terminal and mid-sequence with a two-fold improvement in tumor-to-blood signal [63].

In summary, several protein topologies provide intriguing preclinical results, and the initial clinical results with the affibody are encouraging. The field should aim to continue to broaden the target portfolio, elucidate generalizable means to optimize specific targeting – both in terms of tumor uptake and off-target retention – and translate agents into the clinic.

Highlights.

• Molecular imaging is a valuable clinical tool for cancer detection and treatment.

• Non-antibody protein scaffolds are an effective source of molecular imaging agents.

• Affibodies, knottins, fibronectins, and DARPins have imaged many cancer biomarkers.

• Ligand biophysical properties can be modulated to optimize in vivo performance.