Positron Emission Tomography Detects In Vivo Expression of Disialoganglioside GD2 in Mouse Models of Primary and Metastatic Osteosarcoma

The disialoganglioside GD2 is a cell-surface glycolipid involved in cell adhesion that is widely expressed by pediatric solid tumors (1) including osteosarcomas (2, 3) and neuroblastomas (4, 5). GD2 expression in normal tissue is limited to sympathetic neurons, skin melanocytes, and peripheral sensory nerve fibers (6–9). Tumor GD2 expression is linked to increased cell migration and invasion (10), and there is some evidence of GD2 upregulation in metastatic disease relative to the primary lesion (3). The intimate involvement of GD2 in disease progression and minimal expression in most non-neoplastic tissues make GD2 an attractive target for both immunotherapy (11–13) and imaging (1, 5, 14–16) that address the compelling clinical need to improve the outcomes of these patients.

Researchers have examined two murine IgG3 anti-GD2 antibodies for therapeutic efficacy in GD2-positive tumors (3): 3F8 and 14.18. Both antibodies as well as various analogues have been studied in multiple immunotherapy trials in patients with osteosarcoma, neuroblastoma, and melanoma (1, 12). A recent Children's Oncology Group (COG) trial that evaluated the chimeric antibody ch14.18 in patients with high-risk neuroblastoma demonstrated a 20% increase in two-year event-free survival (9). This is a substantial therapeutic improvement in a population with heretofore only 40%–50% two-year disease-free survival (4, 5). On the basis of the COG results and previous clinical experience, ch14.18 (dinutuximab, Unituxin) was recently approved for the treatment of high-risk neuroblastoma.

Osteosarcomas are the most common bone tumor in children and young adults (17), and GD2 has been detected in >88% of these tumors (2, 3). The overall survival rate for patients with recurrent osteosarcoma is less than 20% with no significant therapeutic improvements since the 1980s (17). Thus, there is a clear unmet need for new osteosarcoma therapies. This has led to an ongoing COG-sponsored trial investigating the utility of ch14.18 and cytokine therapy in recurrent osteosarcoma (17).

Although GD2 is nearly ubiquitous in osteosarcoma (2, 3) and neuroblastoma (6, 7), its expression level is highly variable (3, 12, 18–20). Terzic and colleagues demonstrated a correlation between the variability in GD2 expression and response to ch14.18 treatment in neuroblastoma (20), leading researchers to hypothesize that differences in GD2 expression, both between patients and within a given tumor, may contribute to variations in immunotherapy response (21).

One study reported higher GD2 expression in recurrent osteosarcomas compared with initial biopsy (3), whereas another observed no difference between primary and recurrent osteosarcoma sites (19). One reason for this discordance may be that current in vitro GD2 assays are highly dependent on tumor sampling (19, 20), tissue processing (the gangliosides are soluble in some solvents used to fix tissue; ref. 22), and operator interpretation. Detection of GD2 expression by IHC is widely used and is reported to be more sensitive and accurate than flow cytometry (23). There are, however, inconsistencies in the IHC-measured GD2 levels for various cell lines and tumor types, and reports vary as to whether samples analyzed as touch preparations, fresh-frozen tissue, or paraffin-embedding provide consistent results (7, 22). Likewise, flow cytometry can be challenging for clinical samples because of limited tissue availability. Detection of GD2 in blood (7) or bone marrow (24) samples obviates the limitations of biopsy, but does not provide information about the GD2 status of individual tumors. The challenging in vitro measurement of GD2 expression combined with the probable value of GD2 as a prognostic biomarker and the critical need to improve therapeutic outcomes creates an unmet clinical need for a method to measure GD2 expression in situ before treatment with anti-GD2 antibodies.

Positron emission tomography (PET) using a radiolabeled anti-GD2 antibody offers a potential noninvasive, quantitative method for initial staging, for informing treatment decisions, and for predicting therapeutic response for patients with osteosarcoma. It also offers a method for identifying GD2-positive primary and metastatic tumors, thus allowing evaluation of disease progression and surveillance. Imaging-based screening protocols would maximize the clinical impact of anti-GD2 immunotherapy and aid in monitoring response to the various combinations of drug, antibody, cell-cycle checkpoint inhibitors, or immunomodulatory agents currently being evaluated for the treatment of osteosarcoma (25).

The 3F8, 14.G2a, and ch14.18 anti-GD2 antibodies have been radiolabeled previously with ^99mTc and radioiodine for imaging (16, 26–30). These studies demonstrated the utility of qualitative GD2 imaging for earlier detection of skeletal and soft-tissue metastases. In addition, post-therapy images revealed that the GD2 binding was not saturated by therapeutic dosing with unlabeled antibody (16), indicating that imaging can be performed at multiple stages throughout therapy. More recently, studies demonstrated the feasibility of radiolabeling 14G.2a, ch14.18 (15), and hu14.18K322A (14) with ⁶⁴Cu as well as providing preliminary PET imaging results in strongly GD2-positive mouse models of neuroblastoma and melanoma. A subsequent study with ch14.18-ΔCH2, examining the effect of chelators on biodistribution, reported that S-2-benzyl-1,4,7-triazacyclononane-1,4,7-triacetic acid (Bn-NOTA) chelator conjugated to the antibody via a thiourea linkage provided the optimum signal-to-noise profile (31).

We report here the first preclinical assessment of the PET radiotracer [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A. The anti-GD2 antibody hu14.18K322A is the 98% humanized derivative of the murine 14.18 antibody, which has also shown great promise for immunotherapy of GD2-positive tumors (32, 33). This hu14.18 analogue contains a single point mutation in the CH2 domain (lysine-322 to alanine), resulting in fewer complement activation-dependent side-effects but with therapeutic potency comparable with the clinically approved ch14.18 (32, 34–36). The biphasic pharmacokinetics of hu14.18K322A are similar to those of ch14.18 and other clinically used full-length IgG1_κ mAbs (t_1/2α = 1.74 days; t_1/2β = 21.1 days; ref. 32).

Our results demonstrate an in vivo correlation between [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A accumulation and GD2 expression in patient-derived osteosarcoma xenografts. We also demonstrate the ability of this new radiotracer to detect small GD2-positive lesions in even a modestly GD2-expressing mouse model of osteosarcoma metastasis to the lung, a key criterion for staging patients with osteosarcoma (37). These preclinical data also demonstrate that heterogeneous GD2 expression in primary and metastatic osteosarcoma can be examined noninvasively using PET. Such quantitative imaging allows in situ interrogation of the entire volume of each tumor in a given patient, and surveillance of changes at all sites of disease throughout therapy. This could significantly improve patient selection for anti-GD2 immunotherapy and serve as an in vivo probe for investigating the correlation between target expression and therapeutic response.

All chemicals and reagents were purchased from Sigma Chemical Co. unless noted otherwise. The hu14.18K322A antibody was provided by Children's GMP, LLC, and is based on an expression construct licensed from EMD Sorono (a subsidiary of Merck KGaA). Copper-64 was purchased from The Mallinckrodt Institute of Radiology Cyclotron Facility, Washington University Medical Center (St. Louis, MO). Distilled, deionized water (Milli-Q Integral Water Purification System, Millipore; 18.2 MΩ cm resistivity) was used for all solutions. A CRC-15R dose calibrator (Capintec) or a Wizard² 3 automatic γ-counter (PerkinElmer) was used to measure radioactivity in samples. All animal experiments were performed in compliance with St. Jude Children's Research Hospital (St. Jude, Memphis, TN) Institutional Animal Care and Use Committee guidelines using approved protocols. The St. Jude animal program is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care Institutions (00429).

Conjugation of hu14.18K322A (4 mg/mL in 0.1 mol/L sodium bicarbonate, pH 9.0) with 2-S-(4-isothiocyanatobenzyl)-1,4,7-triazacyclononane-1,4,7-triacetic acid (p-SCN-Bn-NOTA; Macrocyclics, catalog no. B-605) was performed at a molar ratio of 10:1 (chelator:antibody) by the method of Vosjan and colleagues (31, 38). The conjugation solution was incubated at 37°C for 40 minutes with gentle mixing. Unconjugated p-SCN-Bn-NOTA was removed by filtration through a PD-10 column (GE Healthcare) pre-equilibrated with 0.25 mol/L sodium acetate, pH 5.5, followed by concentration by centrifugal filtration (Amicon Ultra-4, MWCO 30K, EMD Millipore, catalog no. Z648035) to a final antibody concentration of 4 mg/mL for storage at −80°C until radiolabeling. The number of chelates per antibody molecule available for ⁶⁴Cu complexation was estimated by the isotopic dilution method (39) using [⁶⁴Cu]CuCl₂ diluted with nonradioactive CuCl₂ (31). The methods described here gave >98% pure Bn-NOTA-hu14.18K322A conjugate at 2–3 chelates/antibody.

Radiolabeling with ⁶⁴Cu was performed using 5 μCi/μg (185 Bq/μg) of antibody in 0.25 mol/L sodium acetate, pH 5.5 (final antibody concentration 2 mg/mL) with gentle mixing for 40 minutes at 37°C. Labeling efficiency and radiochemical purity were determined using thin-layer chromatography as described previously (31). The final product had >95% radiochemical purity and 4–5 μCi/μg (145–185 Bq/μg) specific activity.

All cell lines were obtained through the ATCC and all cell culture reagents were purchased from Invitrogen. Saos-2 (catalog no. HTB-85, RRID:CVCL_0548) and U-2 OS (catalog no. HTB-96, RRID:CVCL_0042) cell lines were grown in McCoy's 5A medium, HOS cells (catalog no. CRL-1543, RRID:CVCL_0312) were grown in Eagle's Minimum Essential Medium and 143B cells (catalog no. CRL-8303, RRID:CVCL_2270) were grown in Eagle's Minimum Essential Medium in Earle's Balanced Salt Solution with 0.015 mg/mL 5-bromo-2′-deoxyuridine. All media contained 10% FBS, and 1 mmol/L l-glutamine. Cell lines were grown under humidified atmospheric conditions in 5% CO₂. All cell lines were verified Mycoplasma free (MycoAlert, Lonza, catalog no. LT07–418) and authenticated using short tandem repeat profiling (PowerPlex Fusion System, Promega, catalog no. DC2408) immediately prior to use (2–3 weeks from thawing).

Human tumor tissue was obtained from patients with osteosarcoma through the Childhood Solid Tumor Network (St. Jude, Memphis, TN) and directly implanted into NOD scid gamma mice. Tumors were then propagated from mouse to mouse by sterile orthotopic implant (40). Immunocompromised NCr-Foxn1^nu mice were anesthetized (1.5%–2.0% isoflurane) and the implant area was cleansed with alternating povidone iodine scrubs and 70% isopropyl alcohol wipes. A small horizontal incision was then made to insert a 100–125 mm³ donor tumor piece, subcutaneously, above the right rear flank. The incision site was then treated with penicillin/streptomycin solution (1–2 drops; 10,000 U/mL) and closed with tissue adhesive (1–2 drops; Vetbond, 3M). Tumors were allowed to grow to at least 150 mm³ before imaging.

Lentiviral-yellow fluorescent protein (YFP) luciferase-labeled human 143B osteosarcoma cells were suspended in Matrigel (BD Biosciences, catalog no. 354234) at 1 × 10⁵ cells/μL and placed on ice. NCr-Foxn1^nu nude mice were anesthetized (1.5% isoflurane) and placed in the supine position (41). Gentle rotation of a 25-gauge needle was used to pierce between the femoral condyles as the needle was advanced into the shaft of the femur, where 10 μL of cell suspension was deposited. Animals were allowed to recover and were monitored for tumor growth by physical examination of the implantation site and weekly whole-body bioluminescence imaging.

All antibodies were purchased from BD Biosciences unless otherwise stated. Osteosarcoma xenograft tissue was diced into small fragments (<1 mm³) in sterile RPMI 1640 medium and disaggregated by incubation at 37°C for 45 minutes in 0.26 Wünsch Units/mL of Liberase TM (Roche). The enzymatic reaction was terminated by adding EDTA to a final concentration of 5 mmol/L. Cells were filtered (70 μm) and washed three times with cold PBS containing 0.1% (w/v) PBS/BSA. For antibody staining, live cells (10⁶) were transferred to tubes containing 50 μL PBS/BSA and 5 μL nonspecific serum. An antibody cocktail containing CD133-APC (Miltenyi Biotec, catalog no. 130-090-826, RRID:AB_244340) and CD117-BV421 (catalog no. 562434, RRID:AB_11154222) tumor/stem cell markers, CD45-Alexa Fluor 700 leukocyte marker (BioLegend, catalog no. 304024, RRID:AB_493761), β2-microglobulin (catalog no. 551338, RRID:AB_394153) or HLA-ABC-FITC (catalog no. 555552, RRID:AB_395935) human-specific antigens, and either 14.G2a-PE anti-GD2 murine antibody (catalog no. 562100, RRID:AB_11154036) or isotype-matched IgG-PE control (catalog no. 555574, RRID:AB_395953) was added to each tube. After 10 minutes at ambient temperature, the cells were washed twice and resuspended in 400 μL PBS/BSA. A live/dead marker (Draq 7; Abcam, catalog no. ab109202) was added before data collection on an LSRFortessa cytometer (BD Biosciences). Cells were gated for single, viable cells that expressed human cell surface markers (HLA-ABC or β2M) and were negative for CD45.

Osteosarcoma patient-derived xenograft (PDX) tumors were resected, embedded in optimal cutting temperature (OCT) compound (Sakura Tissue-Tek, catalog no. 4583), snap-frozen in liquid nitrogen, and stored at −20°C. For the 143B osteosarcoma lung metastasis model, the liver, lungs, and the 143B cell-engrafted hind limb were collected from each mouse and either snap-frozen or processed for formalin-fixed paraffin embedding (FFPE; 10% neutral buffered formalin) and sectioning. Snap-frozen tissues were embedded in OCT media, serial sections (8 μm) were cut and mounted onto positively charged glass slides (Superfrost Plus; Thermo Fisher Scientific), and stained with hematoxylin & eosin (H&E) using an automated stainer (Dako). FFPE tissues were sectioned (4 μm), dried at 60°C for 20 minutes before dewaxing and staining with H&E. All H&E-stained tissues were evaluated for the extent of neoplastic engraftment by a board-certified veterinary pathologist to confirm the presence and morphologic phenotype of the primary xenograft and for evaluation of any metastatic disease in the sampled tissues.

Serial sections taken from both FFPE and snap-frozen specimens were labeled with 14.G2a anti-GD2 disialoganglioside antibody (1:500 dilution; BD Biosciences, catalog no. 554272, RRID:AB_395336), and counterstained with hematoxylin (IntelliPath FLX; Biocare Medical). These IHC sections were analyzed by two pathologists for the presence or absence of GD2 expression using an Eclipse Ni microscope (Nikon).

In vivo bioluminescence imaging of luciferase-expressing 143B cells was performed using an IVIS-200 imaging system (PerkinElmer). Mice were injected intraperitoneally with d-luciferin (PerkinElmer) at 100 mg/kg body weight. After 3–5 minutes, the animals were anesthetized (2% isoflurane), and ventral and dorsal images were acquired using 1-minute exposures with small binning and exposure adjustment when the signal was saturated. Data were analyzed using Living Image 4.5 software (PerkinElmer).

MRI was performed using a 7-Tesla Brüker ClinScan small-animal MRI scanner (Brüker BioSpin MRI, GmbH) equipped with a Brüker 12S gradient and a two-channel phased-array surface coil. Animals were anesthetized (2%–3% isoflurane) and double-gated (for respiration and cardiac motion) gradient-echo protocols (TR 411 ms; TE 3 ms) were used to produce lung images (coronal and axial) using a 176 × 256 matrix, 20 × 30 mm field of view (FOV), and a slice thickness of 0.6 mm. All images were analyzed on a Siemens workstation using Syngo MR B15 software (Siemens).

Small-animal PET, with CT for attenuation correction, was performed on an Inveon system (Siemens Medical Solutions) using proprietary Inveon Research Workspace data analysis software with OSEM 3D reconstruction. Tumor-bearing mice were anesthetized (1.5%–2% isoflurane), injected retro-orbitally (r.o.) with [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A [65–80 μCi (2.4–3.0 MBq) in 100 μL] and allowed to recover. At 48 hours post-injection, mice were anesthetized and single-position, whole-body, 20-minute static PET images were acquired followed by a low resolution CT scan (250 ms exposure with two bed positions and 36% overlap, 120 projections with FOV 54 × 133 mm) for anatomic coregistration and attenuation correction of the PET data. Tumor and nontarget tissue regions of interest were drawn on the CT images and transferred to the PET images to calculate radioactivity concentration (percent of the injected dose/cubic centimeter, %ID/cc).

Osteosarcoma tumor-bearing NCr-Foxn1^nu nude mice (n = 5–6 per tumor type) were anesthetized (1.5%–2% isoflurane) and injected r.o. with 75 μCi (2.4–3.0 MBq, 100μL) of [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A formulated in isotonic saline. At 48 hours post-injection, animals were anesthetized and euthanized by decapitation. Tissues were dissected, weighed, and counted for radioactivity. Data were calculated as the percentage of injected dose per gram of wet tissue (%ID/g).

We and others have demonstrated highly selective PET imaging of GD2 in neuroblastoma and melanoma (14, 15). However, osteosarcoma imaging is rendered more difficult due to the inherent heterogeneity, calcification, and variable incorporation of stromal cells in osteosarcoma tumors (42). The goal of this study was not only to demonstrate detection of these heterogeneous GD2-positive tumors, but also to characterize GD2 expression in vivo. We imaged immunocompromised (NCr-Foxn1^nu) mice bearing osteosarcoma PDXs using [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A and PET. At 48 hours post-injection, most of the unbound [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A had cleared from nontarget tissue and only the strongly GD2-positive SJOS072 PDX tumor (>99% GD2-positive cells by flow cytometry, Table 1) and diffuse thoracic radioactivity were visible on the images (Fig. 1A). Mean tumor radioactivity accumulation in the SJOS072 tumor was 11.4% ID/cc. A second, modestly GD2-positive PDX (SJOS051, 69% GD2-positive) also showed specific accumulation of radiotracer (Fig. 1B) to 9.1% ID/cc.

For a more detailed characterization of the radiotracer distribution, we injected additional SJOS072 and SJOS051 PDX tumor-bearing mice with [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A and euthanized them at 48 hours post-injection for dissection and tissue radioactivity assay. As observed in the PET data, the average tumor concentration of the tracer in SJOS072 (21 ± 5.0% ID/g) and SJOS051 (8.3 ±1.9% ID/g) was proportional to tumor GD2 expression (Fig. 1C), demonstrating selective accumulation of radiotracer in the GD2-positive osteosarcoma versus other tissues. The ⁶⁴Cu concentration in the blood at 48 hours was still relatively high (8.7% ± 4.5% and 7.3% ±1.6% ID/g, respectively) due to the 1.74-day biological half-life of hu14.18K322A (32) possibly combined with specific binding to circulating GD2 shed from tumors (7, 24). The concentration of ⁶⁴Cu in liver (the primary metabolic organ) was ≤5% ID/g, indicating minimal loss of ⁶⁴Cu from the conjugate and similar to the ⁶⁴Cu concentration in all other nontarget organs. As expected for a full-length IgG, we did not observe significant brain accumulation of radioactivity (<0.3% ID/g). These data indicate that [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A exhibits highly selective accumulation in GD2-positive osteosarcomas and can easily detect tumors in vivo even at modest (69%) antigen expression levels.

The different concentrations of [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A in the SJOS051 and SJOS072 PDX osteosarcomas (Fig. 1) suggested that we could use this radiotracer to measure tumor GD2 expression levels in vivo. To test this hypothesis, we determined the GD2 expression in six osteosarcoma PDX tumor types using IHC and flow cytometry and compared these data with the [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A concentration in the same tumors measured by PET. Five of the six osteosarcoma PDXs exhibited varying degrees of GD2 expression by IHC (Fig. 2A–E). GD2 was not detected in the SJOS845 PDX sample (Fig. 2F).

Further analysis by flow cytometry demonstrated a broad range of GD2 expression in these PDX tissues (Fig. 2G–L) and in four osteosarcoma cell lines, U-2 OS, Saos-2, HOS, and 143B, (Fig. 2M–P, respectively; Table 1). The SJOS071 (Fig. 2G) and SJOS072 (Fig. 2H) PDX samples each exhibited >99% GD2-positive cells; however, their mean fluorescence intensity differed more than 10-fold (Table 1) reflecting a marked variance in the amount of GD2 on the cell surface. The percent GD2-positive cells in SJOS051, SJOS081, and SJOS121 (Fig. 2I–K) ranged from 32% to 69%. Consistent with the IHC results, SJOS845 (Fig. 2L) was essentially GD2-negative, displaying <1% GD2-positive cells (Table 1). U-2 OS cells (Fig. 2M) strongly express GD2 (10), whereas Saos-2 cells (Fig. 2N) have minimal GD2 expression, and thus served as positive and negative controls, respectively, for the flow cytometry analyses.

We grew all six PDXs as flank tumors and imaged them by PET-CT using [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A. Because of the heterogeneous nature of osteosarcoma tumors, we calculated the radioactivity concentration as both the mean concentration across the whole tumor volume (PET_(mean)) and the maximum concentration area within the tumor (PET_(max)), reporting the values as %ID/cc. As shown in Fig. 3A, we observed a strong correlation (Pearson r = 0.88, P = 0.01) between the tumor radioactivity concentration in each of the six osteosarcoma PDXs measured by PET and the percentage of GD2-positive cells measured by flow cytometry, supporting the utility of PET imaging to evaluate GD2 in osteosarcoma. A higher linear correlation would not be expected, as osteosarcomas contain varying proportions of bone, cartilage, osteoid and fibrous tissue, in addition to other stromal and inflammatory cells (42).

The SJOS071 and SJOS072 PDXs had the highest uptake of [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A, consistent with the >99% GD2 expression determined by flow cytometry. Conversely, SJOS845 (<1% GD2-positive) showed minimal [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A accumulation and, as expected, was not discernable above normal background by PET. However, SJOS845 was a single sample whose propagation was unreliable and required 7–8 months, thus limiting its availability during the time frame of this study. Because additional GD2-negative PDXs were not available, the low GD2-expressing HOS osteosarcoma cell line (6.5% GD2-positive cells, Fig. 2O) was examined for comparison (10). PET imaging of HOS flank tumors showed ⁶⁴Cu concentrations (PET_(mean) = 7.2 ± 1.2 %ID/cc) comparable with blood levels but slightly higher than expected for nonspecific uptake via the enhanced permeability and retention (EPR) affect [approximately 5% ID/cc (14, 43)], and higher than predicted by a strictly linear correlation (Fig. 3A). Regardless, inclusion of the HOS tumor data in the PDX data analysis had minimal effect on the strength of the correlation between PET and GD2 expression (Pearson r = 0.86, P < 0.01).

PET could not distinguish between PDXs with GD2-positive cells in the 32%–69% range. As the sample sizes were small, we grouped the PET data into high (SJOS071 and SJOS072), moderate (SJOS051, SJOS081, and SJOS121), and low (SJOS845) GD2-expressing tumors. We then used a two-tailed exact Wilcoxon rank-sum test to compare the median of the PET_(mean) (Fig. 3B) and PET_(max) (Fig. 3C) radioactivity concentrations in the moderate expression tumor group (15 individual data points) with those of the highest expressers (7 data points). For PET_(mean), there was no significant difference between the moderate and high expression groups (P = 0.34). However, the highest GD2 expressers displayed PET_(max) [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A concentration values that were significantly higher (P < 0.05) than those of the moderate GD2-expressing group.

The low GD2-expressing HOS tumors were easily distinguished (Fig. 3B and C, P < 0.001) from other GD2-expressing tumors. Likewise, there was a striking 7-fold difference in radiotracer accumulation between the GD2-negative SJOS845 PDX and the next highest GD2-expressing PDX (SJOS121, 44% GD2-positive). However, as SJOS845 was a single data point, we did not include it in the statistical analysis.

We confirmed this correlation observed between radiotracer accumulation and GD2 expression using dissection and counting methods (Fig. 1C; Supplementary Fig. S1). As expected, ⁶⁴Cu concentrations measured in resected tumors showed a linear correlation with the PET data (Pearson r = 0.93, P = 0.01). As was seen in the PET images, the high GD2-expressing tumors SJOS071 and SJOS072 were easily distinguishable from those with moderate GD2 expression (SJOS051, SJOS081, and SJOS121; P < 0.001; Fig. 3D).

Thus, [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A can distinguish between low, moderate, and high GD2-expressing osteosarcomas across a wide gradient of GD2 expression on the cell surface. Analysis of the maximum radioactivity accumulation, PET_(max), within these heterogeneous tumors would provide a more useful clinical assessment of GD2 activity.

The lungs are a major site of metastatic disease, and the primary source of mortality, in osteosarcoma (37). Therefore, we investigated the efficacy of PET using [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A to detect lung lesions. We generated a murine model of metastatic osteosarcoma by injecting luciferase-expressing 143B cells into the femur marrow of mice (n = 8; refs. 41, 44–46), which produces pulmonary metastases in 70% to 80% of the animals (45, 47). This cell line exhibits intermediate GD2 expression (34% GD2-positive cells; Fig. 2P; Table 1), which provides a more representative model for the detection limitations of PET. We then subjected tumor-bearing mice to weekly imaging sessions (see Supplementary Fig. S2), starting with bioluminescence imaging (BLI) of YFP-luciferase-labeled 143B cells to monitor primary tumor engraftment as well as the development of metastatic disease. Animals that were positive for lung metastases on BLI were scanned using MRI, and then injected r.o. with [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A and imaged 48 hours post-injection. These results are summarized in Supplementary Table S1.

A representative mouse that was BLI-positive for lung metastasis 18 days post-implant is shown in Fig. 4A (yellow arrow). MRI of this mouse (Fig. 4B) revealed sites of high contrast in both lungs that were suspicious for osteosarcoma (white arrows). The primary tumor from the injection of the 143B cells into the femoral marrow was clearly visible in the PET-CT image of this animal (Fig. 4C, white arrow). Interestingly, radioactivity accumulation was also apparent in the area of the lungs and thoracic cavity (orange arrow). To verify that the signal in the lungs was metastatic disease rather than nonspecific accumulation or residual blood radioactivity, we harvested, sectioned, and examined the lungs using histology (H&E staining) and IHC staining for GD2. All lung lesions identified by PET (Fig. 4D) coregistered with areas exhibiting morphology consistent with the primary tumor phenotype (Fig. 4E) and stained positive for GD2 (Fig. 4F). Notably, the smallest of these metastatic lesions clearly identified by PET represents a volume of only 29 mm³.

At the conclusion of imaging or at 7 weeks post-implant of tumors, by which time the primary tumor size had progressed to the humane endpoint, we imaged and then euthanized all animals and confirmed suspected sites of disease by histology and IHC staining for GD2 (Fig. 5). We confirmed engraftment at the primary injection site in seven of the eight animals by H&E staining. Figure 5A and B show an example of bone marrow engraftment and expansion of the tumor into the extramedullary muscle. GD2 staining of the primary tumor (Fig. 5C) corresponded with the 34% GD2 expression determined by flow cytometry (Fig. 2P; Table 1). We also confirmed disease in the lung in 6 mice, with morphology consistent with the primary tumor phenotype (Fig. 5D). These metastatic sites also stained positive for GD2 expression (Fig. 5E and F). MRI identified histology-confirmed lung lesions as small as approximately 1 mm³. A histology-confirmed 12-mm³ lung lesion was visible on both MRI and PET; however, the radiotracer accumulation in this lesion was not significantly above the blood levels in the pulmonary vessels. Those histology-confirmed lesions not seen on imaging consisted of neoplastic cells that individualized or formed small clusters in capillaries, but lacked clear establishment within the lungs. These data confirm that the metastatic lesions identified by PET imaging were histologically identical to the primary osteosarcoma and were positive for GD2 by IHC staining. Thus, PET provides a noninvasive method to monitor extremely small osteosarcoma lesions in a clinically relevant lung metastasis model.

In this study, we present the first preclinical assessment of a ⁶⁴Cu-labeled GD2 antibody in a metastatic osteosarcoma model and demonstrate in vivo correlation between radiotracer accumulation and GD2 expression in osteosarcoma PDXs. This novel quantitative imaging method bridges a critical gap in our ability to examine the frequency and intensity of GD2 expression in osteosarcoma, for which patients currently have limited access to the proven clinical benefits of anti-GD2 immunotherapy.

Osteosarcoma and other solid tumors are known to express cell-surface disialogangliosides, particularly GD2 and GD3 (10, 11, 19, 48, 49). Building on the therapeutic success of recent clinical trials of GD2-targeted immunotherapy in neuroblastoma (11, 49), there is strong clinical interest in treatment protocols that include anti-GD2 immunotherapy for other malignancies, such as osteosarcoma, where GD2 expression is lower and/or less uniform than in neuroblastoma (17). Current in vitro methods for measuring GD2 expression (IHC, quantitative PCR and flow cytometry) are highly dependent on factors such as tumor sampling, tissue processing, and operator interpretation, which could explain inconsistencies in reported GD2 levels (19, 20, 22). A PET-based assay of GD2 expression enables quantitative in situ interrogation of the entire tumor volume and allows surveillance of changes in all disease sites throughout the body.

We have shown that GD2-positive primary and metastatic osteosarcomas are readily detected using the novel PET radiotracer [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A, which clearly visualized GD2 expression ranging from 32% to 100% of cells. Pearson analysis of a group of six subcutaneously implanted PDX tumors (Fig. 3A) showed a strong correlation (r = 0.88, P = 0.01) between the radiotracer concentration and the flow cytometry data for these tumors.

Consistent with known osteosarcoma tumor heterogeneity, we observed larger than expected variations in the accumulation of [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A in tumors within some PDX groups but not in the corresponding normal tissues. This variability could be due to histologic heterogeneity of osteosarcomas or actual variation in the amount of GD2 expressed on cell surfaces within a given PDX and was particularly pronounced in the high GD2-expressing PDXs, SJOS071 and SJOS072 (Fig. 3A; Supplementary Fig. S1). Despite this variability, [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A PET could clearly distinguish between high GD2-expressing (>99% GD2-positive; SJOS072 and SJOS071) and moderate GD2-expressing (32%–69% GD2-positive; SJOS121, SJOS051 and SJOS081) osteosarcomas (P < 0.05) using the maximum radiotracer accumulation values (PET_(max)), which are less affected by tumor heterogeneity. We did not observe a difference by PET among the moderately GD2-expressing PDX group.

The low GD2-expressing HOS cell (6.5% GD2 expression) and SJOS0845 PDX (<1% GD2 expression) tumors had minimal radiotracer accumulation, and were easily distinguished (P < 0.001, Fig. 3C) from other GD2-expressing tumors, despite the fact that the actual GD2 expression of the HOS cells is lower than predicted by PET (Fig. 3A). This discrepancy could arise because the immortalized cell line produces more homogeneous tumors exhibiting a higher tumor cell density per gram than the PDX models. Likewise, the exceptionally low radiotracer accumulation in SJOS0845 is consistent with the high level of calcification observed in this PDX or could also be indicative of particularly low vascularization. However, the high blood radioactivity concentration and EPR effect limit quantification in this low GD2 expression range.

These data suggest that the PET_(max) values are a better indicator of actual GD2 status than the PET_(mean) accumulation values, and support using PET imaging to establish a clinical GD2 threshold below which patients are unlikely to respond to anti-GD2 immunotherapy.

Our primary goal is to develop a PET radiotracer for GD2 that can be rapidly translated to human use. Thus, we used the full-length hu14.18K322A antibody (already used in clinical studies) to prepare the corresponding ⁶⁴Cu-labeled radiotracer. A key advantage of ⁶⁴Cu in preclinical investigation is that it allows weekly imaging to evaluate therapy response and to examine the potential effect of pharmacologic doses of GD2 antibodies and other therapeutic agents on the images. However, owing to the slow clearance of the full-length antibody [hu14.18K322A t_1/2α = 1.74 d (32)], possibly combined with specific binding of antibody to circulating ganglioside shed from the GD2-expressing tumors (7), the blood levels of [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A observed at 48 hours still exceed those in the major organs (Fig. 1C). This residual radioactivity in the blood resulted in visualization of the major vessels on the images, which could confound the detection of metastatic lung lesions as well as potentially adding variability when quantifying images of well-vascularized tumors. Co-injection of [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A with excess unmodified hu14.18K322A did not decrease the blood radioactivity levels at 48 hours (P = 0.57; Supplementary Fig. S3). Thus, anti-GD2 PET imaging demonstrates no saturable specific binding to GD2 in the blood and will not be adversely affected by therapeutic levels of anti-GD2 antibodies in patients undergoing immunotherapy.

In an effort to improve on the radiotracer pharmacokinetics of full-length antibodies, various groups have used protein engineering of antibody fragments to improve blood clearance (31, 50). Alternatively, imaging challenges due to the slow antibody clearance could be addressed by using a radionuclide with a longer physical half-life. For example, several full-length antibodies labeled with ⁸⁹Zr (t_1/2 = 78 hours) rather than ⁶⁴Cu are currently used in clinical trials (38). The longer half-life of ⁸⁹Zr allows imaging 5–7 days (3–4 biological half-lives) after radiotracer injection, giving additional time for unbound radiotracer to clear. The disadvantage of ⁸⁹Zr is that images are not available until 5–7 days post-injection and sequential imaging is not possible for 14–20 days, to allow for clearance and decay of the ⁸⁹Zr from the initial injection. Also, development of ⁶⁴Cu-labeled antibodies such as [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A for diagnostic imaging allows direct translation to use in radioimmunotherapy as the ⁶⁴Cu/⁶⁷Cu theranostic pair.

Our results show that despite this limitation, [⁶⁴Cu]Cu-Bn-NOTA-hu14.18K322A can distinguish lesions as small as 29 mm³ from background even in a modestly GD2-positive (34% GD2-positive cells) model of metastatic osteosarcoma of the lung (Fig. 4D). Interestingly, metastatic lesions in this model exhibited a higher ⁶⁴Cu concentration than did the primary lesion in the same animal (e.g., 10.5% vs. 8.5% ID/cc for the lung metastasis and primary tumor, respectively). Dearling and colleagues showed that the concentration of radiolabeled anti-GD2 antibody accumulating in tumors is inversely proportional to tumor size (51), and that lesions smaller than 100 mm³ showed 2–3-fold higher accumulation of radiotracer compared with larger tumors. Roth and colleagues observed that metastatic sites often express levels of GD2 that are higher than those in the primary lesions (3). We found no discernable difference in GD2 expression of primary vs. metastatic lesions by IHC (Fig. 5C and F); however, and the priority for histopathologic confirmation to validate the imaging results precluded cytometric analysis of GD2 expression in the resected tissues. Further examination will be required to determine which of these phenomena, dependence of the PET signal on tumor size or increased GD2 expression (or a combination of both), is responsible for the higher PET signal at the metastatic sites.

Our results demonstrate that PET imaging of GD2 expression can serve as a companion diagnostic method for anti-GD2 immunotherapy. PET imaging allows in situ assessment of GD2 expression in all sites of disease throughout therapy and could serve as a noninvasive, in vivo probe for investigating the relationship between target expression and therapeutic response. These studies also provide in vivo antibody distribution data that inform further development of anti-GD2 radioimmunotherapy, which has potential for treating minimal residual disease before these hidden clusters of cancer cells present as metastatic disease.

J.L.J. Dearling reports receiving a commercial research grant from Clarity Pharmaceuticals. A.B. Packard reports receiving a commercial research grant from Clarity Pharmaceuticals and is a consultant/advisory board member for Iotron Industries. S.E. Snyder reports receiving other commercial research support from United Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E.R. Butch, J.L.J. Dearling, A.B. Packard, B.L. Shulkin, S.E. Snyder

Development of methodology: P.E. Mead, H. Tillman, J.K. Mishra, J.L.J. Dearling, A.L. Vāvere

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.R. Butch, P.E. Mead, V.A. Diaz, H. Tillman, E. Stewart, J.K. Mishra, J. Kim, A. Bahrami, S.V. Stoddard, S.E. Snyder

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.R. Butch, P.E. Mead, V.A. Diaz, H. Tillman, J.K. Mishra, J. Kim, A. Bahrami, Y. Han, S.E. Snyder

Writing, review, and/or revision of the manuscript: E.R. Butch, P.E. Mead, H. Tillman, J.K. Mishra, A. Bahrami, J.L.J. Dearling, A.B. Packard, S.V. Stoddard, A.L. Vāvere, B.L. Shulkin, S.E. Snyder

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.R. Butch, V.A. Diaz, J.K. Mishra, S.E. Snyder

Study supervision: E.R. Butch, B.L. Shulkin, S.E. Snyder

The authors thank Kimberly Guy and Adam Petraglia for animal assistance, Dr. Yimei Li for biostatistics consulting, and the Childhood Solid Tumor Network (St. Jude) and Dr. Chris Morton for supplying osteosarcoma xenograft mice and tissues. Special thanks are given to Drs. Raymond Barfield, Michael Bishop, Michael Dyer, Amos Loh, and Fariba Navid for advice and multiple discussions on immunotherapy, and to the following St. Jude Shared Resources: the Center for In Vivo Imaging and Therapeutics (CIVIT), Biostatistics, and the Hartwell Center. Financial support for this work was provided by Cancer Center Support Grant P30 CA021765 (Developmental Project pilot funds to S. Snyder) and by American Lebanese Syrian Associated Charities (St. Jude, Memphis, TN).

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.