A Novel Engineered Small Protein for Positron Emission Tomography Imaging of Human Programmed Death Ligand-1: Validation in Mouse Models and Human Cancer Tissues

Despite multiple mechanisms employed by cancer cells to evade and inhibit the host tissue's immunologic surveillance (1), many treatment strategies based on antagonizing coinhibitory, or activating costimulatory, pathways of immune cells have emerged (2). Clinical studies based on targeting these pathways clearly indicate prolonged patient survival by 6–12 months (3, 4). In particular, the targeting of immune checkpoints (IC) such as programmed death receptor-1 (PD-1, CD279) and its ligand PD-L1 (B7-H1, CD274), and anti-CTLA-4 therapy is rapidly advancing the field of cancer immunotherapy.

In the case of PD-1/PD-L1 immune checkpoint inhibitor therapy (5), PD-1 is expressed on activated T, B, and natural killer lymphocytes, whereas PD-L1 is expressed in a wide variety of tumors including melanoma, non–small cell lung cancer, Merkel cell carcinoma, breast cancer, and squamous cell carcinoma (4, 6). PD-L1 is also expressed on the surface of macrophages, endothelial cells, and other nonmalignant tissues (e.g., pancreatic), which aids in the prevention of autoimmune disease (7, 8).

In the tumor microenvironment, PD-L1 expression is mainly promoted by the adaptive immune response mechanisms to IFNγ secreted by antigen-bound tumor-infiltrating T cells (8, 9). KRAS and PTEN gene mutations (9p24.1) in certain tumors also promote PD-L1 expression (10, 11). The overall role of PD-L1 expression by tumors is to inhibit activation of infiltrating PD-1–expressing cytotoxic T cells, thereby evading the host immune system and suppressing its response (12).

In current clinical practice, the decision to proceed with IC inhibitor therapy with anti-PD-L1 immunotherapy is based on anti-PD-L1 IHC staining in biopsied tumor tissue sections (13, 14). The amount of PD-L1 expression is also used as a surrogate for determining whether a patient might respond to anti-PD-1 immunotherapy (15). Although IHC is a well-established technique that undergoes rigorous validation prior to clinical utilization, numerous factors may contribute to inconsistent results: (i) under- or mis-sampling of the tumor region with the highest PD-L1 expression at the time of biopsy, (ii) difficulty in obtaining adequate specimens in patients with metastatic disease, and (iii) variability in baseline biomarker expression due to previous anticancer therapies. Furthermore, IHC lacks the spatial and temporal resolution of the biomarker expression profile within the entire tumor. On the other hand, noninvasive PET imaging allows repetitive visualization of PD-L1-expressing tumors throughout the body that enables improved lesion localization and characterization (16, 17).

Many preclinical studies have shown that PD-L1 expression in tumors can be assessed by PET imaging (18–23). However, antibody-based PD-L1 tracer imaging studies suffer from slow clearance of the antibodies from the nontarget tissues and limit the use of shorter half-life radioisotopes for labeling. For diagnostic imaging, tracers with faster clearance profiles from nontarget tissues would enable quantitation of the signals from the target tissues at earlier time points (e.g., 1–4 hours postinjection). This can be achieved using small proteins (∼25–50 kDa) or nanobodies (∼10 kDa), a class of molecules that can be imaged in target tissue within 4 hours of administration. Such an approach would be more patient-friendly (for example the patient would not have to return the day after radiotracer injection for PET imaging) and expedite the initiation of the appropriate IC therapy. We have previously developed a class of protein scaffolds, for example, the tenth type III domain of human fibronectin (FN3, ∼10 kDa) as a platform that can provide faster clearance (<12 hours vs. ∼72 hours in the case of antibodies) and demonstrates a specific in vivo targeting ability to yield excellent tumor-to-background contrast (24).

FN3 has been engineered for many targets (22, 23) with picomolar to nanomolar affinity, including molecular imaging of cancer using PET in murine models (24) and in phase II clinical studies in therapeutic oncology (25). FN3 demonstrates a high stability scaffold, contains three solvent-exposed loops that can be mutated to introduce new high-affinity variants, and a single lysine that provides rapid amine conjugation of chelators (23, 26, 27). Similar kinds of molecules that have already been validated for clinical use include affibodies (21, 28), knottins (29), nanobodies (30, 31), peptides (32, 33), and antibody fragments (34–36). These small molecules are designed to enhance vascular extravasation (37) and tissue penetration for delivery into solid tumors (38, 39).

Anti-hPD-L1 antibody (atezolizumab)-based imaging studies revealed that hPD-L1 can be detected in the tumor microenvironment using various isotopes (¹¹¹In, ⁶⁴Cu, and ⁸⁹Zr) in mouse tumor xenograft models (18). In this report, we present the development of a novel small protein molecule PET tracer, ⁶⁴Cu-FN3_hPD-L1, for imaging of hPD-L1 expression in an hPD-L1–expressing tumor mouse model at time points as early as 1-hour postinjection. In human cancer tissue specimens, we further show a similarity between our FN3_hPD-L1 binder and two validated hPD-L1 antibodies in regard to detecting hPD-L1 expression status.

All reagents were obtained from Sigma-Aldrich unless otherwise stated. N-succinimidyl-DOTA (NHS-DOTA) was purchased from Macrocyclics. The CT26 mouse colon carcinoma cells engineered to express human PD-L1 were authenticated by short tandem repeat profiling in Dr. Irving Weissman's laboratory, and kindly provided for this study. Quantitative analysis of indirect immunofluorescence staining in flow cytometry was used to calculate the number of hPD-L1 molecules per CT26/hPD-L1 cell (QIFKIT, Code K0078, Dako). The Raji human Burkitt lymphoma hPD-L1–negative cells were obtained from ATCC (catalog number CCL-86). The MDA-MB-231 human triple-negative breast adenocarcinoma (TNBC) cells naturally expressing hPD-L1 were purchased (ATCC, catalog number HTB-26). All three cell lines were maintained according to standard techniques and used within three passages. They were maintained in DMEM supplemented with 10% FCS, 2 mmol/L glutamine, 100 U/mL penicillin, 100 μg/mL streptomycin, and 0.25 μg/mL fungizone. Media and supplements were obtained from Life Technologies.

High-performance liquid chromatography (HPLC) was performed on HPLC-Ultimate (Thermo Fisher Scientific) with an ultraviolet detector and an inline radioactivity detector. The system used a SEC-2000 LC column (300 × 7.8 mm) with 5 μm hydrophilic bonded silica support and 400 Å pore size (Phenomenex). Matrix-assisted laser desorption ionization (MALDI) mass spectrometry was performed with a TOF/TOF 5800 (SCIEX) operated in linear mode with sinapinic acid as matrix.

The human PD-L1 (hPD-L1, or B7-H1, CD274) protein was purchased (catalog no. 10084-H02H-100, Sino Biologics), biotinylated, characterized by MALDI, and immobilized on streptavidin magnetic beads for screening of FN3 binders. The yeast surface displayed FN3 G4 library with diversified loops and was sorted and matured as described previously (26, 40). Briefly, yeast displaying 2.5 × 10⁸ FN3 mutants were sorted for binding to magnetic beads with immobilized hPD-L1 protein, followed by FACS for full-length proteins using the C-terminal c-myc epitope. Plasmid DNA from selected clones was recovered, mutated by error-prone PCR of either the entire FN3 gene or the paratope loops, and reintroduced into yeast by electroporation with homologous recombination. As binder enrichment progressed in later evolutionary cycles, FACS for binding to soluble hPD-L1 protein was also used. Five cycles of selection and mutation were performed. Plasmid DNA was recovered, transformed into bacteria, and individual clones were sequenced by standard DNA-sequencing methods. The best binder (FN3_hPD-L1, Fig. 1A) was expressed in bacterial culture with a His₆ tag, purified by nickel column chromatography, and reverse-phase HPLC (41).

Bacterial expression plasmids were constructed to express FN3_hPD-L1. The plasmids also encoded for a C-terminal His₆ epitope tag for purification. Plasmids were transformed into BL21 (DE3) Escherichia coli. Cells were grown in 1 L of lysogeny broth medium for 4 hours and induced with 0.5 mmol/L isopropyl β-d-1-thiogalactopyranoside for 1 hour. Cells were pelleted, resuspended in 10 mL of lysis buffer (50 mmol/L sodium phosphate, pH 8.0, 500 mmol/L sodium chloride, 5% glycerol, 5 mmol/L CHAPS detergent, 25 mmol/L imidazole, and complete ethylenediaminetetraacetic acid–free protease inhibitor cocktail), frozen and thawed, and sonicated. The sample was centrifuged at 12,000 × g for 10 minutes. Fibronectin was purified from the soluble fraction by immobilized metal affinity chromatography and reverse-phase HPLC with a C18 column (Phenomenex). Protein mass was verified by mass spectrometry.

Octet experiments were conducted at 25°C in a buffer of PBS pH 7.4, 0.01% (v/v) Tween-20 and 1% BSA, and sample plates were agitated at 1,000 rpm. Biotin-FN3_hPD-L1 was coupled onto streptavidin tips (Thermo Fisher Scientific). hPD-L1 protein was titrated into 5–500 nmol/L binding sites using a 1.2-fold dilution series. The FN3_hPD-L1 binder was immobilized on a streptavidin-coated tip. These mixtures were allowed to bind the sensor tip–coupled FN3 binder for 10 minutes. We confirmed that neither hPD-L1 protein nor binder bound nonspecifically to the unmodified tips. Samples were analyzed on duplicate tips to verify that the assay was reproducible between tips. Octet data were exported into Scrubber v.2.0a (BioLogic Software Pvt Ltd) for data processing and analysis. hPD-L1 biomarker protein was analyzed in 10 mmol/L HEPES, pH 7.4, 150 mmol/L NaCl, 0.005% (v/v) Tween-20 at five concentrations between 0 and 500 nmol/L. Assays were performed in duplicate and the response from an empty flow cell and from buffer injections was subtracted from each dataset. The data were analyzed using with a global fitting to the 1:1 binding model.

Cells were incubated at 37°C in humidified air with 5% CO₂. For affinity measurement, 1 × 10⁵ CT26/hPD-L1 or Raji (hPD-L1-negative) cells were washed with 0.1% BSA (w/v) in PBS and incubated with various concentrations of FN3_hPD-L1. Cells were pelleted, washed with 0.1% BSA (w/v) in PBS, and incubated with 100 μL of 0.05 μg/μL Alexa Fluor 488–conjugated mouse anti-His₆ antibody (clone AD1.1.10, Bio-Rad) in 0.1% BSA (w/v) in PBS. Cells were washed and analyzed using flow cytometry. The minimum and maximum fluorescence and the affinity values were determined by minimizing the sum of squared errors assuming a 1:1 binding interaction. Experiments were performed in triplicate.

The DOTA-NHS ligand has already shown good biological performance when used in protein conjugation of various radionuclides such as ⁶⁸Ga and ⁶⁴Cu (19, 42). DOTA-FN3_hPD-L1 (Do-FN3_hPD-L1) tracer was prepared by conjugating DOTA-NHS to FN3_hPD-L1 according to a published procedure (41). Briefly, lyophilized FN3_hPD-L1 protein was resuspended in dimethylformamide with 2% triethylamine and reacted at room temperature for 1 hour with 20 equivalents of DOTA-NHS. DOTA-FN3_hPD-L1 was purified by HPLC and lyophilized for ⁶⁴Cu labeling. The number of DOTA chelators conjugated to each FN3_hPD-L1 molecule was calculated by mass spectrometry by comparing the mass of FN3_hPD-L1 and Do-FN3_hPD-L1 (see Fig. 1A; ref. 41).

⁶⁴CuCl₂ was received from the University of Wisconsin (Madison, WI) and its specific activity at the time of shipment (24 hours prior to labeling) was 1.6 ± 0.2 Ci/μmol. Radiolabeling of Do-FN3_hPD-L1 was performed using ⁶⁴CuCl₂ as follows: Do-FN3_hPD-L1, 25–50 μg in 100 μL of 0.25 mol/L ammonium acetate buffer (pH 5.5) was reacted with 92.5–185 MBq of neutralized ⁶⁴CuCl₂ solution at 37°C for 1 hour. After incubation, 0.1 mol/L diethylenetriamine pentaacetic acid (pH 7.0) was added to a final concentration of 5 mmol/L and incubated at room temperature for 15 minutes to scavenge unchelated ⁶⁴CuCl₂ in the reaction mixture. Purification of the ⁶⁴Cu-FN3_hPD-L1 was achieved using SEC-2000 HPLC with a flow rate of 1.0 mL/minute in PBS [0.1 mol/L NaCl, 0.05 mol/L sodium phosphate (pH 7.4)]. The final radioconjugate of ⁶⁴Cu-FN3_hPD-L1 was filtered through a 0.2μm filter into a sterile vial.

Immunoreactivity of the ⁶⁴Cu-FN3_hPD-L1 tracer was tested by cell-binding assays as previously described (41). Two hundred microliters of CT26/hPD-L1 cells were suspended in microcentrifuge tubes at concentrations of 5.0, 0.6, 0.3, 0.16, and 0.08 × 10⁶ cells/mL in PBS (pH 7.4) with 1% BSA (PBSA). Thereafter, each tube received aliquots of 50 μL of ⁶⁴Cu-FN3_hPD-L1 (from a stock solution of 10 μCi in 10 mL PBSA). The triplicate tubes containing the tracer (n = 15; final volume 250 μL each) were gently vortexed and incubated at 37°C. Two hours later, the solutions were centrifuged (300 × g for 3 minutes), resuspended, and washed twice with ice-cold PBS before removing the supernatant. ⁶⁴Cu activity associated with the cell pellet was measured with a gamma counter (1470 WIZARD Automatic Gamma Counter; Perkin Elmer). Competition assays were also performed by the same procedure but with the Raji cells. Linear regression analysis of a plot of total/bound activity versus 1/(normalized cell concentration) was performed, and the immunoreactive fraction was calculated as 1/y-intercept.

Animal studies were approved by the Administrative Panel on Laboratory Animal Care (APLAC) at Stanford University (Stanford, CA). NSG (NOD.Cg-PrkdcscidIl2rgtm1wjl/SzJ) mice were purchased from the Jackson Laboratory (Bar Harbor, ME) and maintained in-house in an AAALAC-accredited facility. The average weight of the NSG mice was 23.0 ± 2.0 g. Six- to 8-week-old NSG mice were implanted subcutaneously with 5 × 10⁶ CT26/hPD-L1 (left shoulder), 5 × 10⁶ Raji (hPD-L1-negative, right shoulder) cells, or 2 × 10⁶ MDA-MB-231 (naturally hPD-L1-expressing, left axilla) cells in 50 μL of PBS mixed with 50 μL of Matrigel Matrix (catalog no. 356234, Corning, Inc.). Mice with tumors ranging in size from 350–450 mm³ (for CT26/hPD-L1 and Raji tumors) or 200–250 mm³ (for MDA-MB-231 tumors) were chosen for the studies. Four groups of NSG mice (n = 3–5/group) received ⁶⁴Cu-FN3_hPD-L1: (i) CT26/hPD-L1-nonblocking (nblk, no blocking of hPD-L1 with nonradioactive FN3_hPD-L1 prior to radiotracer injection), (ii) CT26/hPD-L1-blocking (blk), (iii) Raji (hPD-L1-negative)-nblk, and (iv) MDA-MB-231-nblk. Each mouse received 200 μL of ⁶⁴Cu-FN3_hPD-L1 diluted in PBS, corresponding to 3.7 ± 0.4 MBq (8–10 μg of FN3_hPD-L1), via tail vein injection. The blk group received a blocking dose (100-fold excess) of nonradioactive FN3_hPD-L1 in 200 μL of PBS 2 hours before radiotracer injection. After radiotracer administration, the animals were imaged under 2% isofluorane delivered with 100% oxygen at 0.5, 1, 4, 18, and 24 hours postinjection using a Siemens Inveon small-animal multimodality PET/CT system (Preclinical Solutions, Siemens Healthcare Molecular Imaging). Results are reported as percent injected dose per gram of tissue (%ID/g). Image files were assessed with region of interest (ROI)-based analyses using an Inveon Research Workspace (IRW, Siemens Healthcare Molecular Imaging). For each small-animal PET scan, three-dimensional ROIs were drawn around the tumor, heart, liver, kidneys, and muscles on decay-corrected whole-body images. The tumor ROI analysis was performed in the same manner as that of the other organs (i.e., based on the anatomic information of the CT image in which organs and the tumor are well-delineated to enable drawing of the ROIs). The average radioactivity concentration in the ROI was obtained from the mean pixel values within the ROI volume. These data were converted to counts/mL/minute by using a predetermined conversion factor (41). The results were then divided by the injected dose to obtain an image ROI-derived %ID/g.

Deidentified tissue sections from biopsies and corresponding subsequent resections of 6 cancer patients were obtained through a Stanford institutional review board (IRB)-approved protocol (IRB-44051). The cases consisted of primary lung adenocarcinoma (n = 2), metastatic lung adenocarcinoma to brain (n = 3), and oncocytic thyroid carcinoma (n = 1). Supplementary Table S1 online summarizes the clinical characteristics of the patients from whom the tissue sections were obtained.

The human cancer tissues and implanted mouse xenografts (MDA-MB-231 and Raji) were obtained in paraffin-embedded blocks and sectioned into 4–5 μm–thick sections for IHC and hematoxylin and eosin (H&E) staining, and 10-μm thick sections for immunofluorescence (IF). All tissue sections were deparaffinized prior to staining, which was performed in a blinded manner. A pathologist (S.R. Long) reviewed the IHC and IF staining results to determine the hPD-L1 expression in the H&E-confirmed tumor areas after reviewing multiple fields (4–20 ×) in each specimen. During scoring, the pathologist focused on areas of tumor and excluded areas of necrosis or tumor-infiltrating inflammatory cells such as lymphocytes, which are known to stain for hPD-L1.

For IHC, two rabbit mAbs against human-PD-L1 that had been validated in placental tissue were used: clone E1L3N (catalog no. 13684, Cell Signaling Technology) and clone SP263 (catalog no. 790–4905, Ventana Medical Systems, Inc.). The IHC methods were performed according to the manufacturers' protocols. Briefly, clone E1L3N was diluted 1:500 on the Bond Polymer Refine Detection platform (catalog no. DS9800, Leica Biosystems) and clone SP263 was prediluted (1.61 μg/mL) and applied with the BenchMark ULTRA system (Ventana Medical Systems, Inc.) using the OptiView DAB IHC Detection Kit (catalog no. 760–700, Ventana Medical Systems, Inc.). After validation, the two hPD-L1 antibody clones were separately applied to the human cancer resection tissues from the six cases in this study (there was insufficient corresponding prior biopsy tissue on which to study the validated antibodies) and the implanted mouse xenografts (MDA-MB-231 and Raji). The slides were scanned on an UltraFast Scanner Digital (Philips) digital pathology slide scanner.

For IF, the microscope slides were rinsed twice with Dulbecco PBS (DPBS) at room temperature. A PAP pen was used to outline a hydrophobic barrier around each tissue section. The sections were permeabilized with 0.1% Triton X-100 (v/v) in DPBS for 5 minutes at room temperature, followed by 3 washes (5 minutes each) with DPBS. The tissues were blocked for 1 hour in 10% (v/v) FBS in DPBS. The blocking solution was then removed, and the sections were washed with wash buffer (0.1% BSA (w/v) in DPBS) twice (5 minutes each). The FN3_hPD-L1 binder was diluted 1:200 in 0.01% Tween-20 (v/v) in 1% BSA (w/v) in DBPS, from the stock concentration of 5 mg/mL in DMSO to a final concentration of 25 μg/mL FN3_hPD-L1. The diluted binder was then applied to the tissue sections overnight at 4°C. Afterwards, the binder was removed with three DPBS washes (5 minutes each). The DyLight 650–conjugated anti-6X His tag secondary antibody (ab117504, Abcam) was diluted 1:800 in 1% BSA (w/v) in DPBS and applied to the tissues in the dark for 1 hour at room temperature. The secondary antibody was removed with three DPBS washes (5 minutes each). Coverslips were mounted and their edges sealed with clear nail polish. Images were acquired using a NanoZoomer 2.0-RS whole slide imager (Hamamatsu) and saved as TIFF files using the NanoZoomer Digital Pathology (NDP) Scan version 2.5 software.

Data were analyzed and plotted using Prism version 5.01 (GraphPad Software Inc.). Data were tested for normality using the Shapiro–Wilk test. An unpaired two-tailed Student t test was used to compare the means of groups of normally distributed data. For non-normally distributed data, the Wilcoxon rank-sum analysis of medians was performed. A P value less than 0.05 was considered to be statistically significant. Bonferroni correction was used to adjust alpha for multiple comparisons. Data are presented as mean ± SD.

On the basis of the hPD-L1 cell binding study, the best unique clone against hPD-L1 protein was selected and called FN3_hPD-L1. We performed the sequence analyses for confirmation of the full framework and uniqueness of the loop region. The soluble protein yield was >5 mg/L with >95% purity by HPLC and demonstrated 11,826 kDa molecular weight when measured by mass spectrometry (expected 11,824 kDa). This soluble FN3_hPD-L1 binder was biotinylated to determine the binding affinity (one biotin per FN3_hPD-L1 molecule, as determined by MALDI spectrometry). The purified soluble FN3_hPD-L1 binder displayed the dissociation constant (K_d) of 0.61 ± 0.02 nmol/L for purified hPD-L1 (Fig. 1B). A flow cytometry assay with live CT26/hPD-L1 cells indicated the binding affinity of FN3_hPD-L1 for cell surface hPD-L1 was 1.4 ± 0.3 nmol/L (Fig. 1C).

The characterization of ⁶⁴Cu-FN3_hPD-L1 radiotracer is summarized in Table 1. DOTA-NHS (Do) was conjugated to lysine groups of FN3_hPD-L1 and yielded Do-FN3_hPD-L1 with 1-2 chelates per binder, as confirmed by MALDI-TOF mass spectrometry (Supplementary Fig. S2). This DOTA-conjugated FN3_hPD-L1 binder (Do-FN3_hPD-L1) was further purified using SEC2000-HPLC column and radiolabeled with ⁶⁴Cu. The final yield and purity of ⁶⁴Cu-FN3_hPD-L1 were greater than 70% and 95%, respectively (Fig. 2A). SEC2000-HPLC showed that the specific activity of FN3_hPD-L1 was 5.3 ± 0.5 GBq/μmol (Table 1). Immunoreactivity of the ⁶⁴Cu-FN3_hPD-L1 was tested against the CT26/hPD-L1 cells (mean ± SD = 3 × 10⁵ ± 1.5 × 10⁴ hPD-L1 molecules per CT26/hPD-L1 cell) and hPD-L1–negative Raji cells and found to be 83.6% ± 8.8% and 4.9% ± 0.3% (P = 0.0001), respectively (Fig. 2B, results from Raji cells were not shown due to low immunoreactivity). Thus, the ⁶⁴Cu-FN3_hPD-L1 tracer was specific for the hPD-L1 antigen.

Figure 3A shows representative coronal ⁶⁴Cu-FN3_hPD-L1 small-animal PET/CT images (at 24 hours postinjection) of NSG mice bearing the CT26/hPD-L1 tumors in the left shoulder (nblk and blk) and NSG mice bearing the hPD-L1–negative Raji tumors in the right shoulder (Raji-nblk). These PET images clearly show that the ⁶⁴Cu-FN3_hPD-L1 tracer was able to detect CT26 tumors that express hPD-L1. Figure 3B shows the uptake of the ⁶⁴Cu-FN3_hPD-L1 radiotracer in kidney at various time points from 0.5 to 24 hours postinjection (mean %ID/g ± SD). The ROI analyses of kidney uptake in the three different NSG mouse groups (CT26/hPD-L1-nblk, CT26/hPD-L1-blk, and Raji-nblk) showed the radiotracer clearance pattern (mean ± SD %ID/g) to be comparable at 24 hours postinjection (25.4 ± 3.1, 23.4 ± 0.6, and 22.7 ± 0.6, respectively, at 24 hours postinjection). Figure 3C shows higher tumor uptake in the CT26/hPD-L1-nblk group compared with the CT26/hPD-L1-blk and Raji-nblk groups at each time point. In particular, in mice bearing CT26/hPD-L1 tumors, the difference between nblk and blk cohorts was observed as early as 1-hour postinjection (2.6 ± 0.5 vs. 0.7 ± 0.02 %ID/g, P = 0.004) and was maintained at 4 hours postinjection (3.6 ± 0.7 vs. 0.7 ± 0.1 %ID/g, P = 0.006). The 24-hour postinjection uptake of ⁶⁴Cu-FN3_hPD-L1 (mean ± SD %ID/g in the CT26/hPD-L1-nblk group) was 1.2 ± 0.18 and 4.9 ± 0.36 in the heart and liver, respectively, compared with 5.0 ± 0.8 in the tumor. Small-animal PET/CT 3D visualization of ⁶⁴Cu-FN3_hPD-L1 radiotracer uptake at 1-hour postinjection in the CT26/hPD-L1-blk and CT26/hPD-L1-nblk groups is shown in Supplementary Video S3.

Supplementary Fig. S5A shows representative coronal ⁶⁴Cu-FN3_hPD-L1 small-animal PET/CT images (at 24 hours postinjection) of NSG mice bearing the MDA-MB-231 xenograft in the left shoulder (nblk). These images clearly show that the ⁶⁴Cu-FN3_hPD-L1 tracer was able to detect hPD-L1 protein, which is naturally expressed in this cell line. Supplementary Figure S5B shows higher ⁶⁴Cu-FN3_hPD-L1 ROI-based tumor uptake (mean ± SD %ID/g) in the MDA-MB-231-nblk group compared with the MDA-MB-231-blk group at 4 hours (1.8 ± 0.1 vs. 1.2 ± 0.1 %ID/g, respectively, P = 0.004) and 24 hours (3.6 ± 0.4 vs. 1.6 ± 0.2 %ID/g, respectively, P = 0.004) postinjection. The difference in tracer uptake between the nblk and blk groups was apparent as early as 1 hour postinjection (1.4 ± 0.09 vs. 0.8 ± 0.2 %ID/g, respectively, P = 0.012). The radiotracer clearance from kidney at 24 hours postinjection (mean ± SD %ID/g) in the two different MDA-MB-231 xenograft cohorts (nblk and blk) was 17.2 ± 0.7 and 12.6 ± 1.4, respectively (P = 0.02).

We further calculated the tumor-to-muscle ratios of the PET signal at 1 and 4 hours postinjection in the CT26/hPD-L1-nblk, CT26/hPD-L1-blk, and Raji-nblk groups as %ID/g, after decay correction (Fig. 4A). The tumor-to-muscle ratio of ⁶⁴Cu-FN3_hPD-L1 was significantly greater in the CT26/hPD-L1-nblk group compared with the corresponding blk group at both 1 hour postinjection (5.6 ± 0.9 vs. 2.1 ± 0.2, P = 0.0014) and 4 hours postinjection (13.1 ± 2.3 vs. 1.5 ± 0.1, P = 0.0005). Figure 4B shows the tracer uptake at 24 hours postinjection in a panel of ex vivo organs from the three groups of mice. Ex vivo tumor uptake in the CT26/hPD-L1-nblk, CT26/hPD-L1-blk, and Raji-nblk groups was 4.2 ± 0.5, 1.0 ± 0.3, and 1.0 ± 0.2 %ID/g, respectively (P = 0.001 for CT26/hPD-L1 nblk vs. blk, and P = 0.003 for CT26/hPD-L1-nblk vs Raji-nblk groups, two separate t tests). Tumor-to-muscle ratio from ex vivo data in the respective groups was 23.8 ± 4.8, 6.3 ± 2.9, and 4.7 ± 1.4 (P = 0.01 for CT26/hPD-L1-nblk compared with each of the other groups, two separate t tests). Ex vivo kidney uptake (%ID/g) in the CT26/hPD-L1-nblk and -blk mice was 32.4 ± 3.1 and 34.9 ± 2.7, respectively (P = 0.04). Ex vivo liver uptake (%ID/g) in the CT26/hPD-L1-nblk and -blk mice was 3.4 ± 0.2 and 4.3 ± 0.1, respectively (P = 0.003). Tumor radiotracer uptake (%ID/g) in the CT26/hPD-L1-nblk group at 24 hours postinjection as measured by the ROI from the imaging study was 5.0 ± 0.8, compared with 4.2 ± 0.5 based on the ex vivo tissue uptake cpm count results (P = 0.15).

In the mice bearing the MDA-MB-231 xenografts, the ROI-based ⁶⁴Cu-FN3_hPD-L1 tumor-to-muscle ratio was greater in the nblk group compared with the corresponding blk group at both 1 hour (2.2 ± 0.2 vs. 1.3 ± 0.1, respectively, P = 0.012) and 4 hours (3.4 ± 0.4 vs. 2.0 ± 0.5, respectively, P = 0.022) postinjection. The ex vivo tumor uptake at 24 hours postinjection in the nblk and blk groups was 3.6 ± 0.5 and 1.8 ± 0.3 %ID/g, respectively (P = 0.013). Tumor uptake (%ID/g mean ± SD) in the nblk group at 24 hours postinjection was 3.6 ± 0.4 (ROI-based) compared with 3.6 ± 0.5 (ex vivo), P = 0.97.

To assess the translational potential of the FN3 _hPD-L1 scaffold's ability to correctly identify the presence of hPD-L1 expression in human tissue, FN3_hPD-L1 binder immunofluorescence was compared with IHC of the two validated hPD-L1 antibodies in tissue sections from six human cancer cases (Supplementary Table S1). Supplementary Figure S4 displays the hPD-L1 expression status at the time of biopsy (based on the results from an FDA-approved send-out lab) and subsequent resection (using the two validated hPD-L1 antibodies and the FN3_hPD-L1 binder). Supplementary Figure S4 shows the heterogeneity of hPD-L1 expression not only from one patient to another, but also from biopsy to resection within the same patient. In the very low (<5%) hPD-L1–expressing cases (cases 1–3, the two lung primary tumors and one of the metastatic lung to brain tumors) at the time of resection, there was 100% agreement among the two validated hPD-L1 antibodies and the FN3_hPD-L1 binder. In the remaining three cases with moderate-high (30%–80%) hPD-L1 expression (cases 4–6), the two validated antibodies resulted in comparable (i.e., within 5%–15% of each other) hPD-L1 expression results. In these same three cases, the magnitude of hPD-L1 expression based on the FN3_hPD-L1 binder was lower but still ≥10% (Supplementary Fig. S4). Figure 5 shows H&E, IHC (using two validated hPD-L1 antibodies), and immunofluorescence (using the FN3_hPD-L1 binder) in representative low and high hPD-L1–expressing human cancer tissues (cases 3 and 6, respectively). At these extrema, the FN3_hPD-L1 staining pattern demonstrated a similar trend compared with that of the two validated hPD-L1 antibodies.

In the MDA-MB-231 xenograft that received ⁶⁴Cu-FN3_hPD-L1 (100 μCi/10 μg of FN3_hPD-L1 binder), hPD-L1 staining with the validated antibodies was scored as 55% using clone E1L3N and 85% using clone SP263 (Supplementary Fig. S5C). We also performed validated hPD-L1 antibody staining in MDA-MB-231 (hPD-L1 expressing) and Raji (hPD-L1 nonexpressing) xenografts that were not exposed to ⁶⁴Cu-FN3_hPD-L1 (Supplementary Fig. S6). The MDA-MB-231 and Raji hPD-L1 expression was scored as 60% and 0%, respectively (clone E1L3N), and 90% and 5%, respectively (clone SP263).

In this report, we present the development of a novel radiotracer that targets hPD-L1, an immune checkpoint protein expressed in most tumors. The resulting FN3-based small protein binder of hPD-L1 (FN3_hPD-L1), which was approximately one-tenth the size of an antibody, bound purified hPD-L1 protein and cells engineered to express hPD-L1 (19, 24, 41). Whereas antibody-based tracers can take a few days to attain optimal tumor uptake and then be cleared from the background tissues (43), small protein binders can do so within 24 hours (29). We made three molecular components (FN3_hPD-L1 binder, Do-FN3_hPD-L1 conjugate, and ⁶⁴Cu-FN3_hPD-L1 radiotracer), each of which was carefully assayed for its binding property against purified hPD-L1 protein (Figs. 1 and 2).

The preclinical in vivo imaging study revealed that the ⁶⁴Cu-FN3_hPD-L1 radiotracer rapidly targeted the tumors expressing hPD-L1 (Fig. 3). In fact, this tracer provided a tumor-to-muscle ratio in the CT26/hPD-L1-nblk group that was 3 and 9 times greater than that in the group preinjected with nonradioactive FN3_hPD-L1 (blk), as early as 1 and 4 hours postinjection, respectively (Figs. 3-4). Furthermore, we evaluated the specificity of the tracer in vivo under preblocking with nonradioactive FN3_hPD-L1 (blk) and pre-nonblocking (nblk) conditions in mice bearing hPD-L1–positive (CT26/hPD-L1) and hPD-L1–negative (Raji) tumors. The results indicated that the PET signals at 4 hours postinjection from the CT26/hPD-L1 tumors were substantially lower in the blk group compared with the nblk group by 6-fold (Fig. 3).

Previously, we developed the hPD-L1-targeting ⁶⁴Cu-DOTA-HAC tracer and tested it in a NSG mouse model bearing CT26/hPD-L1 cells engineered to expresses hPD-L1 protein, which resulted in a moderate uptake of ∼2 %ID/g at 1 hour postinjection (19). At 1 hour postinjection, the current ⁶⁴Cu-FN3_hPD-L1 tracer and the ⁶⁴Cu-DOTA-HAC tracer that used a different scaffold (19) each had a tumor-to-muscle ratio of approximately 6. A protein binder's binding affinity is influenced by its binding domain, not its size (24). As reported in refs. 24, 40, and 44 and this study, the smaller the protein binder, the faster its clearance. For example, based on ROI analysis in a study by Olafsen and colleagues, the in vivo tumor-specific targeting of their ⁶⁴Cu-anti-CD20 minibody with two binding domains in CD20-positive tumors versus CD20-negative tumors was 2.3-fold higher at 4 hours postinjection and 1.9-fold higher at 19 hours postinjection (44). In contrast, based on ROI analysis the in vivo tumor-specific targeting of our single-binding domain ⁶⁴Cu-FN3_hPD-L1 radiotracer in hPD-L1–positive versus hPD-L1–negative tumors was 5.6-fold higher at 4 hours postinjection and 8.1-fold higher at 24 hours postinjection. The apparent increased tumor retention of our small protein binder, both within a few hours postinjection as well as over time, is promising. It may be a good candidate tracer for noninvasive imaging-based determination of hPD-L1 expression status in tumors at early time points postinjection, which could help to determine likelihood of patient response to immune checkpoint blockade therapies.

To confirm the applicability of our tracer to human tumors with heterogeneous expression levels of hPD-L1, we performed ⁶⁴Cu-FN3_hPD-L1 PET imaging in a mouse model of CT26/hPD-L1 murine colon carcinoma cell line and hPD-L1–negative Raji Burkitt lymphoma cell line. We found that at 2 hours postinjection, the ⁶⁴Cu-FN3_hPD-L1 uptake in the CT26/hPD-L1 tumors was 4.7-fold higher compared with the Raji tumors. Recently, Chatterjee and colleagues reported a peptide-based radiotracer, [⁶⁴Cu]WL12, targeting hPD-L1 in mouse models of Chinese hamster ovary (CHO) tumors and CHO tumors engineered to stably express hPD-L1 (33). They reported a 3.1-fold higher uptake of their tracer in the CHO-hPD-L1 group compared with the CHO group at 2 hours postinjection (33). Donnelly and colleagues reported the results of 2-hour dynamic PET imaging (23). At 2 hours postinjection, the ratio of ¹⁸F-BMS-986192 uptake in the higher hPD-L1–expressing L2987 to the lower hPD-L1–expressing HT-29 tumors was 3.4 (23).

Although the ⁶⁴Cu-FN3_hPD-L1 tracer showed excellent tumor uptake at early time points and rapid clearance from most nontarget tissues, its elevated signals in the liver and kidney at 24 hours postinjection were of less concern for two reasons. The liver activity may be due to dissociation of ⁶⁴Cu from DOTA (45) or charge effects of the engineered protein and the DOTA chelator. Kidney retention is a common problem for small proteins (44) because they pass through the glomerulus and can be reabsorbed in the renal tubules. Although we observed increased uptake of ⁶⁴Cu-FN3_hPD-L1 in the kidneys compared with that in the tumors, such uptake is similar to that of many other comparable-sized ⁶⁴Cu-DOTA-labeled molecules (e.g., ⁶⁴Cu-anti-CD20 minibody; ref. 44). Over the first 4 hours after administration, the tracer uptake increased 38% in the tumor, whereas it decreased by 16% in the kidney, suggesting renal clearance of the radiotracer (Fig. 3). On the other hand, in our previous study of a different class of binder derived from hPD-1 protein [high-affinity consensus (HAC); ref. 19], the ⁶⁴Cu-DOTA-HAC radiotracer uptake decreased in both the tumor and kidney over the first 4 hours after administration, by 14% and 27%, respectively. Over the course of hPD-L1 tracer development in our laboratory, the ⁶⁴Cu-FN3_hPD-L1 radiotracer appears to have achieved improved tumor retention by 1.3-fold over the other binder class, HAC (19).

The results of our study and others (18, 44, 46) demonstrate that smaller binders can provide improved imaging results compared with high molecular weight antibodies both in terms of tumor-to-background ratio and absolute tumor uptake at early time points after tracer injection (e.g., 1–4 hours). For example, antibody-based tracers can take 3–10 days to attain optimal tumor uptake and be cleared from the background tissues (47, 48). An ex vivo biodistribution study by Lesniak and colleagues found that the antibody-based radiotracer [⁶⁴Cu]atezolizumab had a 1.3-fold higher uptake in mice bearing MDA-MB-231 TNBC xenografts compared with those bearing hPD-L1-low–expressing SUM149 TNBC xenografts 24 hours postinjection (49). At the same postinjection time point, we found a 3.5-fold higher uptake of ⁶⁴Cu-FN3_hPD-L1 in mice bearing MDA-MB-231 xenografts compared with those bearing non-hPD-L1–expressing Raji Burkitt lymphoma xenografts. Hence, the FN3-based small protein binder strategy has considerable advantages as an imaging agent at the earlier post-tracer injection time points of in vivo tumor imaging in delineating the tumor-to-background tissue (e.g., muscle and blood). From the patient perspective, the development of a PET tracer to visualize hPD-L1–expressing tumor cells at early time points would enable tracer injection and PET imaging on the same day, facilitating earlier treatment decisions.

This novel tracer might be used to take advantage of the sensitivity of PET in cases of tumors with moderate to high hPD-L1 expression, as demonstrated by the results of the comparative histology analysis we performed in multiple human cancer tissue specimens (Fig. 5; Supplementary Fig. S4). It should be noted that hPD-L1 is also expressed on lymphocytes, macrophages, and histiocytes, as well as in areas of tumor necrosis (50). In addition to the cell membrane, hPD-L1 is also expressed variably within cancer cells. The two hPD-L1 clones evaluated by validated antibodies used in this study resulted in comparable overall hPD-L1 expression in adjacent tissue sections despite not overlapping completely. This confirms the previously known heterogeneity of hPD-L1 expression patterns based on the antibody used and the respective cutoffs for positivity (13, 14, 16). Our FN3_hPD-L1 binder identified similar hPD-L1 expression to that of the two validated hPD-L1 antibodies in the very low (<5%) hPD-L1–expressing tumors, suggesting high specificity of the FN3_hPD-L1 binder for correctly identifying tumors that will likely not be responsive to IC blockade therapy, according to current clinical standards. In the remaining moderate–high (30%–80%) hPD-L1–expressing tumors based on the validated antibodies, the FN3_hPD-L1 binder staining was also elevated, but its magnitude was not as high (10%–20%). Taken together, these results indicate that a FN3_hPD-L1 binder–based value of ≤5% could serve as the cutoff to rule-out the patient subgroup of low-hPD-L1–expressing tumors. Again, based on the FN3_hPD-L1 binder immunofluorescence results, a cut-off value of ≥10% may identify the moderate- to high- hPD-L1–expressing tumors. It should be reiterated that these FN3_hPD-L1 binder-based cut-off values differ from the antibody-based cut-off values for low- and moderate-/high-hPD-L1 expression. Given the heterogeneity of staining patterns within and between various classes of hPD-L1 binders (e.g., antibodies, fibronectin-based), a one-to-one correlation between hPD-L1 antibody staining pattern versus FN3_hPD-L1 binder staining pattern at the cellular level was not anticipated. However, Fig. 5 shows that at the tissue level (10 × magnification), there was overlap between the areas stained by the hPD-L1 antibodies and by the FN3_hPD-L1 binder. The immunofluorescence staining confirmed one of the goals of this study, that FN3_hPD-L1 bound hPD-L1 in human cancer tissues.

Overall, PET imaging with ⁶⁴Cu-FN3_hPD-L1 in the CT26/hPD-L1 syngeneic tumor model indicated favorable hPD-L1-expressing tumor visualization at the earliest time points (1–4 hours postinjection, Supplementary Video S3), and the PET signal was specific for hPD-L1 when compared against the preblocked CT26/hPD-L1 (hPD-L1–positive) group and the nonblocked Raji (hPD-L1–negative) group. Furthermore, ⁶⁴Cu-FN3_hPD-L1 uptake in the tumors increased while in the background tissues it decreased at 24 hours postinjection. Furthermore, we established that FN3_hPD-L1 bound hPD-L1 in human cancer tissues. On the basis of these findings, the ⁶⁴Cu-FN3_hPD-L1 radiotracer has the potential for human translation to prospectively identify likely responders to immune checkpoint blockade therapy.

In conclusion, we developed and radiolabeled a novel 12-kDa FN3-based anti-hPD-L1 tracer (⁶⁴Cu-FN3_hPD-L1) for PET imaging of human PD-L1 expressed in a mouse colorectal carcinoma syngeneic tumor model. In addition, FN3_hPD-L1 was confirmed to bind hPD-L1 in human cancer tissue specimens with known hPD-L1 expression status based on validated hPD-L1 antibodies. This indicates the potential for clinical translation of this radiotracer.

S.S. Gambhir is an employee of and reports other remuneration from Cellsight, holds ownership interest (including patents) in CellSight Technologies and ImaginAb, and is a consultant/advisory board member for ImaginAb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Natarajan, S.S. Gambhir

Development of methodology: A. Natarajan, C.B. Patel, S.S. Gambhir

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Natarajan, C.B. Patel, S. Ramakrishnan, P.S. Panesar, S.R. Long, S.S. Gambhir

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Natarajan, C.B. Patel, S. Ramakrishnan, P.S. Panesar, S.R. Long, S.S. Gambhir

Writing, review, and/or revision of the manuscript: A. Natarajan, C.B. Patel, S. Ramakrishnan, P.S. Panesar, S.R. Long, S.S. Gambhir

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Natarajan, C.B. Patel, S. Ramakrishnan, S.S. Gambhir

Study supervision: A. Natarajan, C.B. Patel, S.S. Gambhir

We would like to thank the Ben and Catherine Ivy Foundation (to S.S. Gambhir), National Cancer Institute (R01 CA201719, to S.S. Gambhir), Stanford Cancer Institute Fellowship for Cancer Research (to C.B. Patel), and American Brain Tumor Association Basic Research Fellowship supported by the Ryan J. Hanrahan Memorial (to C.B. Patel) for their support in funding this research. We acknowledge the support of Dr. Timothy Doyle for the small-animal PET/CT imaging performed at the Stanford Center for Innovative In Vivo Imaging (SCi³), Ms. Pauline Chu for tissue processing and H&E staining, Ms. Ellen Gomulia for IHC, and Ms. Karen Eliahu for protein purification.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.