Exploitation of CD133 for the Targeted Imaging of Lethal Prostate Cancer

This article is featured in Highlights of This Issue, p. 979

The management of patients with metastatic prostate cancer initially relies on inhibiting the androgen receptor (AR) signaling axis by androgen deprivation therapy (ADT) through surgical castration or gonadotropin-releasing hormone agonists. The therapeutic benefit of ADT is transient and patients inevitably develop disease recurrence, also known as castration-resistant prostate cancer (CRPC), which can occur as early as 18 months after the initiation of ADT (1–3). Driven by aberrant AR signaling, second-generation antiandrogens have had a profound effect in extending the lifespan of patients with CRPC (3). Unfortunately, many patients present with de novo resistance to these therapies and those that receive an initial benefit often develop acquired resistance rapidly through mechanisms such as AR amplification, mutation, and splice variant expression. Similarly, many men with de novo or acquired resistance to AR signaling inhibitors may display a non-AR–driven form of disease referred to as aggressive variant prostate cancer (AVPC; refs. 2, 4, 5). AVPC broadly encompasses CRPC that is non-AR–driven and may or may not express neuroendocrine markers or possess small-cell morphology (6, 7). This lethal subset of prostate cancer is characterized by high metastatic burden in both the bone and viscera, minimal response to therapy, and poor overall prognosis (8, 9).

Effective treatment options for AVPC currently do not exist and novel therapies are urgently needed. Critical to the development of novel therapies for AVPC is the ability to accurately image this disease in patients. For decades now, bone scintigraphy using [^99mTc] methylene diphosphonate ([^99mTc]Tc-MDP) has been the standard for imaging metastatic prostate cancer despite its limited sensitivity and specificity for cancerous lesions (10, 11). Furthermore, recent studies using the PET tracer, [¹⁸F]sodium fluoride (Na[¹⁸F]F), have demonstrated superior sensitivity for detecting bone lesions in patients with prostate cancer (11, 12). While this is successful for patients who present with bone metastases, a striking number of patients with AVPC also present with visceral disease, rendering bone scintigraphy and Na[¹⁸F]F PET imaging inadequate (13, 14). [¹⁸F]fluorodeoxyglucose ([¹⁸F]FDG) is commonly used in the clinic to image types of cancer that are dependent on glucose metabolism (15). However, [¹⁸F]FDG has performed poorly for imaging prostate cancer due to its unique metabolic properties that result in metastatic lesions with little avidity for glucose (15–17). Other PET imaging agents, such as [¹¹C]acetate, [¹¹C]/[¹⁸F]choline, and [¹⁸F]fluciclovine, have been employed to image prostate cancer biochemical recurrence (12, 18–20); however, these agents have yet to be investigated for AVPC. Recently, there has been much success imaging prostate-specific membrane antigen (PSMA) in bone and visceral metastases of patients with prostate adenocarcinoma using small-molecule PET radioligand probes (21). Several PSMA imaging studies have documented a lack of probe uptake in AR-negative metastatic lesions suggesting that AVPC does not express PSMA (22–24). Recently, a ⁶⁸Ga-labeled gastrin-releasing peptide receptor (GRPR) antagonist demonstrated success at detecting prostate cancer in patients with biochemical recurrence by PET/MRI (25, 26). Given the overexpression of neuropeptides in neuroendocrine tumors, it is highly probable that such agents could be used to image AVPC with neuroendocrine-differentiation (27–29). Other imaging modalities such as MRI and CT have been useful in detecting metastatic disease, but tell nothing of the underlying biology of the cancer cell (30). Currently, however, there is no accurate imaging modality available for patients with AVPC.

The lack of targetable antigens specific to AVPC has complicated the development of an imaging agent for this disease subtype. The heavily glycosylated pentaspan transmembrane protein, CD133, has often been described as an antigen on the surface of both stem cells and cancer stem cells (31). In a previous study, we used human antibody phage display to identify a novel single-chain variable fragment (scFv) antibody for CD133, termed HA10 (32). Herein, we show that CD133 is highly overexpressed at the mRNA and protein level in a multitude of patients possessing AVPC with an AR-negative, neuroendocrine (NE) marker–positive (AR⁻/NE⁺) phenotype. By microarray analysis, we confirmed that CD133 and PSMA expression were inversely related and that AVPC is PSMA-negative. Moreover, we used a full-length human IgG version of HA10 to selectively identify CD133-positive cancer cells by near-infrared (NIR) optical and [⁸⁹Zr]-PET imaging in subcutaneous tumor and metastatic mouse models of AVPC. Our findings identified CD133 as a novel, previously unknown marker of AR⁻/NE⁺ AVPC that can be exploited as an imaging target to assess and monitor disease progression. In addition, CD133-targeted imaging agents could aid in the development of novel therapeutics for a subtype of prostate cancer that is currently incurable by monitoring patient response to therapy.

HEK293T cells were purchased from the ATCC and were maintained according to ATCC guidelines. CWR-R1 cells and luciferase-expressing CWR-R1-EnzR cells were obtained from Dr. Scott Dehm (Masonic Cancer Center, University of Minnesota, Minneapolis, MN) and Dr. Donald Vander Griend (Department of Pathology/Surgery, University of Illinois at Chicago, Chicago, IL), respectively. All parental cell lines were authenticated by short-tandem repeat profiling prior to manipulation. Parental CWR-R1 and CWR-R1-EnzR cells were lentivirally transduced to express CD133 as described previously (32). Expression of CD133 in transduced cell lines (CWR-R1^CD133 and CWR-R1-EnzR^CD133) was confirmed via qPCR and Western blot and compared with CD133-negative parental cell lines. All cells were grown in DMEM supplemented with 10% FBS, 1% antibiotic–antimycotic, and 1% glutamax and incubated at 37°C and 5% CO₂. In addition, CD133-expressing cells were continuously supplemented with 3 μg/mL puromycin to ensure stable levels of CD133 expression.

The protocols for biopanning, ELISA screening, scFv expression and purification, as well as affinity/specificity characterization of the isolated scFv clone, HA10, were followed as described previously (32). The heavy-chain (354 bp) and light-chain (318 bp) variable domains of the HA10 sequence were cloned separately into pFUSE2ss-derived human IgG expression vectors (InVivoGen) and cotransfected into HEK293T cells according to the manufacturer's guidelines. Following incubation, the serum was collected, filtered through a 0.45-μm filter, and purified using an iTrap Protein A HP column (GE Healthcare). The eluate was collected and concentrated using a 50 kDa centrifugal filter. A final buffer exchange was performed into 1× PBS using a Sephadex G25 PD10 desalting column (GE Healthcare). The purity of the final HA10 human IgG was analyzed by reduced and nonreduced SDS-PAGE, and the concentration was measured on the basis of absorbance at 280 nm using a NanoDrop One UV-Vis Spectrophotometer (Thermo Fisher Scientific).

Microarray data in Fig. 1 was extracted from previously published studies of a set of metastatic tumors from men with CRPC and patient-derived xenograft (PDX) models of prostate cancer (33, 34). Both datasets are available in the Gene Expression Omnibus under accessions GSE77930 and GSE93809. Whole-exome sequencing of CRPC samples published in Beltran and colleagues (23) used in Fig. 2 was acquired from the cBioPortal for Cancer Genomics (35).

Curated samples for analysis were obtained from the Prostate Cancer Biorepository Network (PCBN) and the University of Minnesota Masonic Cancer Center (Minneapolis, MN) using a University of Minnesota Human Subjects Division–approved IRB protocol for tissue acquisition (IRB#1604M86269) and with written patient consent. The tissue sections were stained for CD133 using the HA10 antibody in a rabbit scaffold as previously described (23). A pathologist reviewed the slides and assigned IHC scores as previously reported (36). The staining was assessed using a quasi-continuous scoring system by multiplying the optical density level (0– no stain, 1– faint stain, and 2- definitive stain) by the percentage of cells at each staining level. The sum of three multiplicands provided a final score for each sample ranging from 0 to 200. Intense staining ranged from 140 to 200, moderate from 70 to 140, weak from 1 to 70, and an absence of immunoreactivity was 0. The samples stained for CD133 were positive for chromogranin A (CHGA; scores ranging from 70 to 200; antibody – chromaganin A PB9097 Boster Biological Technology rabbit polyclonal) and absent for AR using an antibody that recognized the protein N-terminus (androgen receptor SP107 from Sigma. The AR⁻/CHGA⁺ (n = 25) soft-tissue metastases were liver (n = 14), lymph node (n = 3), lung (n = 2), retroperitoneal (n = 4), and pleural (n = 2) while the AR⁺/CHGA adenocarcinoma sections (n = 10) were liver (n = 6), lymph node (n = 2), and lung (n = 2).

All animal studies were performed in athymic nu/nu mice (Envigo) following Institutional Animal Care and Use Committee (IACUC) approval at the University of Minnesota (Minneapolis, MN). For subcutaneous tumor implantation, animals (n = 3–4/group) received unilateral injections on the right shoulder of 1 × 10⁶ CWR-R1 or CWR-R1^CD133 cells (in 100 μL) in a 1:1 dilution of Matrigel (Corning) to 1× PBS. Tumors were measured twice weekly with calipers and volumes were calculated as length × width × height. For the intracardiac dissemination model, animals (n = 4/group) received injections of 2 × 10⁵ CWR-R1-EnzR or CWR-R1-EnzR^CD133 cells (in 100 μL) in 1× PBS directly into the left ventricle of the heart, with a 75% accuracy rate. Following the intracardiac injections, mice were weighed twice/week to assess overall health and bioluminescent imaging was performed once/week to evaluate tumor formation and growth.

HA10 IgG was labeled with IRDye 800CW NHS Ester (LI-COR Biosciences) according to the manufacturer's instructions to develop a NIR imaging agent (NIR-HA10 IgG). Mice bearing subcutaneous tumors were imaged when all tumors in each group reached a threshold of 100 mm³. Each mouse was administered 1 nmol/L (155 μg) of NIR-HA10 IgG via tail vein and imaged at 1, 5, 24, 48, 72, 96, 120, and 144 hours postinjection. Following intracardiac dissemination, mice exhibiting sizeable metastatic lesions as indicated by bioluminescent imaging (BLI) were administered 1 nmol (155 μg) of NIR-HA10 IgG via tail vein and imaged at 24, 48, 72, and 144 hours. In both studies, animals were imaged using an IVIS Spectrum Scanner (Perkin Elmer) and euthanized at a fixed endpoint 6 days after the NIR-HA10 IgG injection or monitored for overall health and terminated when tumor volumes reached 1,000 mm³ or weight decreased more than 15%. To determine fluorescent intensity of subcutaneous xenografts, manually drawn regions-of-interest (ROI) were normalized to a background level of fluorescence on each mouse. To account for variability in size and luciferase expression of intracardiac tumors, a relative radiance unit was determined by dividing the total radiant efficiency of the NIR imaging by the total counts of the BLI signal and used to quantify differences in signal between CD133-positive and CD133-negative mice.

For nuclear imaging studies, HA10 IgG was conjugated to p-SCN-Bn-Deferoxamine (DFO, Macrocyclic) as described previously (37). Zirconium-89 (⁸⁹Zr) was purchased from the University of Wisconsin Medical Physics Department. Once received, [⁸⁹Zr]Zr-oxalate (277 MBq) in 1.0 mol/L oxalic acid (600 μL) was adjusted to pH 6.8–7.5 with 1.0 mol/L Na₂CO₃. To radiolabel the IgG, the DFO–HA10 IgG conjugate (400 μL, 3.49 mg/mL, 1.4 mg of mAb) in 0.5 mol/L HEPES (pH 7.5) was added to the neutralized [⁸⁹Zr]Zr-oxalate solution and incubated at room temperature with gentle agitation for 1 hour. The labeled product was purified using a size-exclusion PD-10 column preequilibrated with PBS buffer. Crude and purified samples were analyzed by radio-TLC using 50 mmol/L EDTA (pH 5.0) as the eluent. The specific activity of [⁸⁹Zr]Zr-HA10 IgG was calculated to be 85 MBq/mg and the radiochemical purity was >98%. The isotype control antibody was labeled using an identical procedure with a specific activity of 103 MBq/mg and radiochemical purity of 95%. Size-Exclusion HPLC analysis with a radioactive detector was not performed on any of the radiolabeled products.

The immunoreactivity of [⁸⁹Zr]Zr-HA10 IgG was assessed by using antigen-specific cellular binding assays using the CD133-transduced cell line (CWR-R1-EnzR^CD133) and the non–CD133-expressing parental cell line (CWR-R1-EnzR). CWR-R1-EnzR and CWR-R1-EnzR^CD133 cells were suspended in microcentrifuge tubes at concentrations of 0.5, 1, 2, 3, 4, and 5 × 10⁶ cells/mL in 500 μL of 1× PBS. Aliquots of [⁸⁹Zr]Zr-HA10 IgG (50 μL, 0.93 MBq) in 1% BSA were added to each cell suspension (final volume 550 μL) and incubated at room temperature with gentle agitation for 1 hour. Cells were resuspended and washed twice with ice-cold 1× PBS. The supernatant was removed and the [⁸⁹Zr] radioactivity of the cell pellets were counted using a HIDEX Automatic Gamma Counter (HIDEX). The count data was background-corrected and the immunoreactive fraction of [⁸⁹Zr]Zr-HA10 IgG was assessed by comparing the total number of counts in the cell suspensions by control samples.

The stability of [⁸⁹Zr]Zr-HA10 IgG with respect to change in radiochemical purity was evaluated at 0, 48, 96, and 144 hours following purification. For the stability studies, 1.3 MBq of [⁸⁹Zr]Zr-HA10 IgG was added to 500 μL of 1% BSA in 1× PBS and incubated at room temperature. The radiochemical purity of 5 μL of [⁸⁹Zr]Zr-HA10 IgG was assessed at each time point by radio-TLC.

Acute in vivo biodistribution studies were conducted to evaluate the uptake of [⁸⁹Zr]Zr-HA10 IgG in mice bearing subcutaneous CWR-R1-EnzR (100–500 mm³) or CWR-R1-EnzR^CD133 (100–500 mm³) tumors. Mice were randomized before the study and administered [⁸⁹Zr]Zr-HA10 IgG (0.37–0.55 MBq, 1–2 μg of mAb, in 100 μL of 1× PBS for injection) via tail vein injection. Mice (n = 3–4/group) were euthanized by cervical dislocation at 24 and 72 hours postinjection. Ten organs (including the tumor) were removed, rinsed in water, dried in air for 5 minutes, weighed, and counted on an automatic gamma-counter (Hidex) for accumulation of [⁸⁹Zr] radioactivity. The total number of counts (counts per minute, cpm) of each organ was compared with a standard syringe of known activity and mass. Count data were background- and decay-corrected, and the percentage injected dose per gram (%ID/g) for each tissue sample was calculated by normalization to the total amount of activity injected into each mouse.

PET imaging experiments were conducted on an Inveon μPET/CT Scanner (Siemens Medical Solutions). Mice were administered [⁸⁹Zr]Zr-HA10 IgG formulations (5.5–7.4 MBq, 25–30 μg of mAb, in 200 μL of 1× PBS) via tail vein injection. Mice were anesthetized by inhalation of 2% isoflurane and PET images were recorded at time points between 24 and 144 hours postinjection. PET list-mode data were acquired for 30 minutes using a gamma ray energy window of 350–650 keV and a coincidence timing window of 3.438 ns. CT acquisition was performed for 5 minutes at 80 kVp, 500 μA, 384 ms per step, and 340 steps covering 220 degrees. CT images were reconstructed using a Hu scaled Feldkamp algorithm resulting in 192 × 192 matrix and PET utilized Ordered Subset Expectation Maximization (OSEM-3D) with 18 subsets and 2 iterations resulting in a 128 × 128 matrix. Two-dimensional (2D) images were prepared in Inveon Research Workplace and quantified using AMIDE. An empirically determined system calibration factor was used to convert voxel count rates to activity concentrations and the resulting image data were normalized to the administered activity to parameterize images in terms of %ID/g. Drawn ellipsoid ROIs were used to determine the mean %ID/g in various tumors. The PET tracers Na[¹⁸F]F and [¹⁸F]2-fluorodeoxyglucose ([¹⁸F]FDG) were purchased from the Department of Radiology at the University of Minnesota (Minneapolis, MN). Both tracers were administered via tail vein (9 MBq per mouse) and imaged 1 hour postinjection.

All statistical analyses were performed using GraphPad 7.04. qRT-PCR and luciferase assay experiments prior to xenograft implantation were repeated three times with three technical replicates each time; all results were represented as mean ± SEM. IHC scores were compared using a Mann–Whitney t test to account for differences in sample size. Patient microarray data were analyzed using a one-way ANOVA, which was corrected for multiple comparisons using Sidak hypothesis testing and LuCaP microarray data were analyzed using a Welch t test. Pearson's correlation coefficient was used to study the relationships between the genes shown in scatterplots. Animal studies were performed with n = 3–4 and signals intensities were quantified as mean ± SEM. Statistical significance was determined using a two-way ANOVA and corrected for multiple comparisons using Sidak hypothesis testing. The symbols used to represent the P values were as follows: ns, P > 0.05; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.

Our initial studies on CD133 suggested an inverse relationship between the expression of AR and CD133 in a limited number of samples by IHC (32). To assess whether CD133 overexpression was associated with a particular prostate cancer phenotype, we investigated CD133 as well as gene signatures of AR status and NE differentiation in gene expression microarray data of 171 CRPC tumors from 63 rapid autopsy subjects at the University of Washington (Seattle, WA; Fig. 1A; refs. 6, 23, 33). The gene that encodes CD133, PROM1, displayed the highest level of expression in AR⁻/NE⁺ tissues (P < 0.0001, Fig. 1B). CD133 and AR expression exhibited a moderate inverse correlation (Pearson correlation = −0.5989, Fig. 1C). Similarly, expression of the prototypical AR-related genes, KLK3 and FOLH1, which encode for PSA and PSMA, respectively, were also inversely correlated to CD133 expression (Pearson correlation = −0.4789, Supplementary Fig. S1A; Pearson correlation = −0.5042; Fig. 1D).

Because early passages of PDX models display similar morphology to the original tumors from which they are derived (38), 24 early-passage LuCaP xenografts were investigated for expression of CD133. Only four PDX models displayed an AR⁻/NE⁺ gene signature and three (75%) of these models showed overexpression of PROM1 (Fig. 1E), which was statistically significant when compared with the AR⁺/NE⁻ PDX models (P < 0.0001, Fig. 1F). Complimentary to the gene expression in patient samples, CD133 exhibited a very strong inverse correlation to both AR and KLK3 in the PDX models (Pearson correlation = −0.886, Fig. 1G; Pearson correlation = −0.8565; Supplementary Fig. S1B).

Further highlighting the association of CD133 with NE-differentiated prostate cancer, whole-exome sequencing, published by Beltran and colleagues (23), of patients with CRPC with adenocarcinoma (n = 34) and neuroendocrine prostate cancer (n = 15) showed an upregulation of CD133 in NE-differentiated disease (Fig. 2A; P < 0.001). Having elucidated the relationship between AR status, NE differentiation, and CD133 at the mRNA level, we next used IHC with our HA10 scFv in a rabbit scaffold to investigate CD133 protein expression in soft-tissue metastases. Patient tissues (n = 35) were classified as either adenocarcinoma (n = 10) when sections were AR⁺/CHGA⁻ or neuroendocrine AVPC (n = 25) when sections were AR⁻/CHGA⁺. All of the adenocarcinoma tissue sections were negative when stained for CD133 (Fig. 2B and C). Contrarily, 92% (23/25) of the neuroendocrine AVPC tissue sections displayed intense staining and the remaining 8% (2/25) displayed moderate staining for CD133 (Fig. 2B). Scoring of the staining intensity in all samples revealed that CD133 expression was significantly different between patients with adenocarcinoma and neuroendocrine AVPC (P < 0.0001).

The HA10 scFv was expressed as a full-length bivalent human IgG (IgG1 scaffold) and labeled with the NIR fluorophore IRDye 800CW (NIR-HA10 IgG) for NIR optical imaging. NIR optical imaging was initially used as a proof-of-concept to document the specific localization of the antibody and acquire pharmacokinetic properties allowing for the informed selection of the appropriate radioisotope for PET imaging. Because no immortalized prostate cancer cell lines uniformly express CD133, model systems were developed. The human prostate cancer cell lines CWR-R1 and luciferase-expressing CWR-R1-EnzR were transduced to express CD133. CD133 expression in each cell line was quantified by qPCR (Supplementary Fig. S2A and S2B). Mice bearing either subcutaneous CWR-R1 (n = 4) or CWR-R1^CD133 (n = 4) tumors were administered 1 nmol of NIR-HA10 IgG via tail vein and imaged at 1, 5, 24, 48, 72, 96, 120, and 144 hours. As early as 24 hours postinjection of NIR-HA10 IgG, increased uptake in the CD133-positive CWR-R1^CD133 tumors was observed compared with the CD133-negative tumors (Fig. 3A). Localization of the antibody to CD133-positive tumor remained present up to 144 hours. A small amount of nonspecific localization due to the enhanced permeability retention (EPR) effect was observed in the CWR-R1 tumors; however, the antibody cleared by 72 hours postinjection. Similar EPR effects have been observed with the PSMA-targeted antibody and it is still considered specific enough for clinical translation as a theranostic agent (39). Average fluorescent intensity in CD133-positive tumors remained significantly higher than CD133-negative tumors from 24 to 72 hours (Fig. 3B). After the final 144-hour time point, the tumors and organs of the mice were excised and imaged by NIR (Fig. 3A). NIR imaging demonstrated a strong signal in the CWR-R1^CD133 tumors, which was not present in the CWR-R1 tumors, verifying that the NIR-HA10 IgG was retained by the tumor and able to selectively detect CD133 up to 144 hours postinjection. Nonspecific localization of NIR-HA10 IgG in secondary tissue was not observed in the ex vivo analysis. A nonspecific human IgG1 mAb was also labeled with IRDye-800CW and administered via tail vein to CWR-R1 (n = 4) and CWR-R1^CD133 (n = 4) subcutaneous xenograft mice as an isotype control (Fig. 3C and D). No localization of the isotype control antibody to the tumors was observed in the xenografts at the imaging time points and by ex vivo analysis at 144 hours postinjection. The CWR-R1^CD133 and CWR-R1 tumors used in this study were fixed, sectioned, and stained for CD133 with HA10 in a rabbit scaffold for IHC. CWR-R1^CD133 sections displayed high levels of CD133 expression across the entire tumor by IHC, indicating stable CD133 expression following xenograft implantation, while immunoreactivity was absent in the CWR-R1 tumor sections (Fig. 3E).

The ability of NIR-HA10 IgG to detect small, dispersed CD133-postive lesions in complex microenvironments was next tested in a metastasis model. To create an appropriate spontaneous metastasis model, mice received intracardiac injections of either luciferase-expressing CWR-R1-EnzR (n = 4) or luciferase-expressing CWR-R1-EnzR^CD133 (n = 3) cells. The luciferase activity of the cells was assessed 1–3 days prior to injection in all mouse models (Supplementary Fig. S2C). Bioluminescent imaging was performed on the mice once per week to assess lesion formation and size. Once mice in each group had at least one sizeable tumor (>3 × 10⁵ total counts), they received 1 nmol of NIR-HA10 IgG via tail vein. NIR optical imaging was then conducted at 24, 48, 72, and 144 hours postinjection (Fig. 3F). Comparable with the subcutaneous model, the signal was highly visible and specific for CD133-positive tumors at 24 hours, remained high for up to 72 hours, and was still detectable at 144 hours. The quantitative analysis supported this observation demonstrating a significant difference in relative radiance at 48 hours (P = 0.0383) and a noticeable decline in signal between 72 and 144 hours in CD133-positive tumors (Fig. 3G). Furthermore, the relative radiance signal remained low in the all of the CD133-negative tumors with minimal tumor or mouse variability, suggesting that the NIR-HA10 IgG is highly selective for CD133-positive tumors.

The results of the NIR optical imaging studies suggested that longitudinal PET-imaging studies could be performed using radiolabeled HA10 IgG. Therefore, we decided to use the long-lived positron-emitting radioisotope [⁸⁹Zr] (t_1/2 = 3.3 days) for imaging. In addition, PET imaging with [⁸⁹Zr]ZrDFO-conjugated antibodies, including studies with the PSMA-targeted antibody J591, are commonplace in the clinic (39–42). HA10 IgG was first conjugated to DFO and radiolabeled with [⁸⁹Zr]oxalate at room temperature under slightly alkaline conditions using modified methods from Zeglis and colleagues (37). Purity was assessed by radio-TLC and peaks were compared with a pure [⁸⁹Zr⁴⁺] standard (Supplementary Fig. S3A). Only [⁸⁹Zr]Zr-HA10 IgG sample preparations, which resulted in a purity of >98% and a specific activity of >74MBq/mg were used for future analyses.

The stability of [⁸⁹Zr]Zr-HA10 IgG was determined by radio-TLC after incubation in 1% BSA in PBS for up to 144 hours at room temperature (Supplementary Fig. S3B). The radiochemical purity remained above 96% at all the time points, indicating a <2% decrease at any given time. Thus, [⁸⁹Zr]Zr-HA10 IgG was expected to remain intact in vivo on the time scale described in our PET imaging studies. Similarly, immunoreactivity was measured by an antigen-specific in vitro cellular association assay using CWR-R1-EnzR and CWR-R1-EnzR^CD133 cells (Supplementary Fig. S3C). The immunoreactive fraction of [⁸⁹Zr]Zr-HA10 IgG was directly proportional to the number of CD133-positive cells in the sample and displayed a strong linear relationship in the CD133-positive cell line (R² = 0.9588). CD133-negative control cells showed no binding to [⁸⁹Zr]Zr-HA10 IgG (R² = 0.0003), further demonstrating the specificity of [⁸⁹Zr]Zr-HA10 IgG for CD133-expressing cells.

The ability of [⁸⁹Zr[Zr]-HA10 IgG to detect CD133-postive cancer cells in vivo was first tested in mice bearing subcutaneous CWR-R1^CD133 (n = 3) or parental CWR-R1 (n = 3) tumors. Mice were injected with 5.5 MBq (67 μg, 84 MBq/mg) via tail vein and imaged by μPET/CT at 24, 48, 72, and 144 hours (Fig. 4A). Transverse 2D images of the CWR-R1^CD133 xenografts showed the highest signals in the tumor compared with the CWR-R1 xenografts, which showed high liver uptake. In addition, three-dimensional (3D) reconstructions were generated to assess [⁸⁹Zr]Zr-HA10 IgG uptake across multiple planes. Tumor margins were well defined in CWR-R1^CD133 images between 24 and 72 hours, which was not observed in the CD133-negative xenografts. Both groups of mice displayed nearly complete clearance at 144 hours. Time activity curves were generated from the PET images to display the mean %ID/g of [⁸⁹Zr] uptake in the CD133-positive versus CD133-negative groups of tumor-bearing mice (Fig. 4B). A significant difference was observed at the 24- and 48-hour time points with mean %ID/g values averaging 28.96 ± 0.92 (CWR-R1^CD133) and 18.97 ± 1.56 (CWR-R1) at 24 hours (P = 0.0003) and 23.60 ± 2.04 (CWR-R1^CD133) and 15.94 ± 1.89 (CWR-R1) at 48 hours (P = 0.0041).

Ex vivo biodistribution was performed to evaluate overall distribution and potential off-target effects of [⁸⁹Zr]Zr-HA10 IgG (Fig. 4C; Supplementary Table S1). The data reveal that distribution of [⁸⁹Zr]Zr-HA10 IgG was comparable in all organs between the two groups, with the exception of the tumors. Uptake in the liver, the main clearance organ for full-length immunoglobulins, and the spleen where immunoglobulins are known to localize nonspecifically was observed in the models at the time points analyzed as anticipated (43, 44). At 24 hours, [⁸⁹Zr] uptake was higher in CD133-positive tumors at a level which was considered insignificant, but at 72 hours the average %ID/g of [⁸⁹Zr] uptake was approximately 3-fold higher in CD133-positive tumors (P < 0.0001). Similarly, minimal [⁸⁹Zr] uptake was observed in various other organs. These data correlate well with the PET imaging data and suggest that CD133 is a promising antigen that can be selectively targeted for the imaging of CD133-expressing tumors. The isotype control antibody was radiolabeled with [⁸⁹Zr] and injected into CWR-R1^CD133 (n = 3) and CWR-R1 (n = 3) subcutaneous xenografts [Fig. 4D; 5.5 MBq (53 μg, 103 MBq/mg)]. Imaging at 24, 48, and 72 hours postinjection documented insignificant localization to the tumors. In the 2D-reconstructed views, liver uptake was primarily observed. CWR-R1^CD133 (n = 2) and CWR-R1 (n = 2) mice were also injected with the commonly used PET tracer [¹⁸F]FDG (9 MBq). Poor uptake and tumor specificity of [¹⁸F]FDG was observed further highlighting the need for new imaging agents for prostate cancer.

The diagnostic potential of [⁸⁹Zr]Zr-HA10 IgG was further investigated by μPET/CT imaging in mice bearing spontaneous metastatic lesions. Following intracardiac injection with luciferase expressing CWR-R1-EnzR^CD133 or CWR-R1-EnzR cells, mice underwent bioluminescent imaging once per week to monitor spontaneous lesion development and growth. At approximately 3.5 weeks postinjection, the remaining healthy CD133-positive (n = 2) and CD133-negative (n = 2) mice had developed multiple metastatic lesions in or around the bone (Fig. 5A). Mice were injected with 7.4 MBq (22 μg, 350 MBq/mg) of [⁸⁹Zr]Zr-HA10 IgG via tail vein and imaged by μPET/CT at 24 and 72 hours. 2D and 3D images revealed a high level of [⁸⁹Zr] distribution at 24 hours in both groups of mice, while [⁸⁹Zr] uptake at 72 hours was exclusively observed in the tumors of CD133-positive mice. Quantification of the mean %ID/g of [⁸⁹Zr] uptake in the tumors confirmed these results by showing that there was no significant difference at 24 hours despite higher signals in the CD133-positive tumors, 27.41 ± 2.50 versus 19.35 ± 2.63 (Fig. 5B). At 72 hours, the mean %ID/g in the tumors was significantly different among the two groups, 24.30 ± 3.19 and 11.82 ± 0.57, respectively (P = 0.0069). Imaging of a CD133-postive dissemination model mice (representative image of n = 2) with Na[¹⁸F]F (9 MBq) documented localization to the bone and a lack of tumor specificity (Fig. 5C).

The emergence of non-AR–driven disease in patients with CRPC has steadily increased after the introduction of AR signaling inhibitors such as abiraterone and enzalutamide. At this stage of disease, survival is poor and therapies that dramatically prolong life do not exist. While new therapies are in demand, new imaging agents to monitor disease progression and response to therapy are also needed. PSMA-targeted radiotracers that have shown promise at detecting bone and visceral metastases in patients with adenocarcinoma are ineffective due to the lack of PSMA expression in non-AR–driven prostate cancer. As such, novel antigens and targeted imaging agents are needed to aid in the detection and monitoring of this subtype of disease. In this study, we have identified CD133 as a new targetable antigen that is overexpressed in one of these non–PSMA-expressing patient populations and developed a novel antibody-based imaging agent that can accurately detect CD133 in preclinical models of prostate cancer. To our knowledge, this is the first study demonstrating that CD133 is a marker for AR⁻/NE⁺ AVPC. Furthermore, despite the use of preclinical CD133-targeted PET imaging in other cancers (45), this is the first time that CD133 has been targeted for imaging of prostate cancer.

Our targeted agent, HA10 IgG, showed significant selectivity and sensitivity for CD133-expressing tumors using multiple imaging modalities and cancer models, verifying the specificity of the antibody–antigen interaction. Because of the fact that no prostate cancer cell lines express significant levels of endogenous CD133, we had to use models engineered to express CD133. Although artificial, the IHC staining intensities of CD133 in AVPC biopsies and our engineered models (Figs. 2C and 3E) were similar suggesting that CD133 expression in our models was representative of CD133 expression in patients with AVPC. Interestingly, subcutaneous xenograft models in both the NIR and PET imaging studies displayed better tumoral uptake at earlier time points (between 24 and 48 hours, Figs. 3A and 4A), whereas, metastatic xenografts displayed better uptake at slightly later time points (between 48 and 72 hours, Figs. 3D and 5A). It has been shown that different prostate cancer xenografts exhibit varying degrees of vasculature as determined by tumor–stromal interaction, size, and site of the tumor (46). Differences in these factors are likely to account for the slightly different uptake kinetics exhibited by the HA10 IgG. In addition, PET imaging revealed that [⁸⁹Zr]Zr-HA10 IgG tumoral uptake varied within the CD133-positive metastatic tumors. While all CD133-positive tumors were visible at 72 hours in the 2D images, only some tumors were visible in the 3D reconstruction after background normalization, suggesting there was less [⁸⁹Zr]Zr-HA10 IgG uptake in some of the metastatic lesions. Because of the high signals in the small spinal tumor and large mandibular tumor, we postulated that size and anatomic location were not the primary causes of this occurrence. We also determined that CD133 expression is stable following implantation of the xenografts, however, we did not rule out that the metastatic process and tumor microenvironment of the secondary site may have altered the tumor phenotype and resulted in less [⁸⁹Zr]Zr-HA10 IgG uptake. Multiple studies have suggested that preclinical metastatic models may lack the ability to faithfully mimic the tumor microenvironment during metastasis (47–49), which may explain the reduced PET signal in some of the CD133-positive metastatic lesions.

In conclusion, this study illustrates the importance of identifying new antigens for targeted imaging and treatment monitoring of patients with AVPC. Our data show that patients with AVPC with an AR⁻/NE⁺ phenotype possess high levels of the surface protein, CD133. Our previous studies show that the epitope our antibody recognizes on CD133 is not highly expressed in early stages of prostate cancer or healthy tissues (32), demonstrating its promise as targetable marker for patients with late-stage AVPC. Furthermore, our data show that CD133 can be exploited for improved imaging using a novel antibody developed by our lab. HA10 IgG was specific for CD133-positive tumors by various imaging modalities, including clinically relevant modality PET/CT imaging. These encouraging results indicate that [⁸⁹Zr]Zr-HA10 IgG displays high potential as a radiotracer for noninvasive immunoPET imaging of patients with AR⁻/NE⁺ AVPC.

No potential conflicts of interest were disclosed.

Conception and design: P.M. Glumac, J.P. Gallant, A.M. LeBeau

Development of methodology: P.M. Glumac, J.P. Gallant, Y. Li, A.M. LeBeau

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.M. Glumac, J.P. Gallant, P. Murugan, S. Gupta, I.M. Coleman, P.S. Nelson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.M. Glumac, J.P. Gallant, M. Shapovalova, P. Murugan, I.M. Coleman, A.M. LeBeau

Writing, review, and/or revision of the manuscript: P.M. Glumac, J.P. Gallant, Y. Li, S. Gupta, I.M. Coleman, P.S. Nelson, S.M. Dehm, A.M. LeBeau

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.M. Glumac, J.P. Gallant

Study supervision: S.M. Dehm, A.M. LeBeau

The authors acknowledge the University Imaging Centers at the University of Minnesota (Minneapolis, MN) for their technical advice and coordination of various imaging equipment. In particular, the authors thank Adrienne Sherman for her confocal microscopy assistance and Dr. Thomas Pengo for his advice and training regarding the medical imaging data analysis. They thank the University of Minnesota's Research Animal Resources Staff for assisting with research animal maintenance and include a special thanks to Dr. Nathan Koewler for his help in producing their intracardiac mouse model. They are also grateful to Colleen Forster and the rest of the Biorepository and Laboratory Services Division at the University of Minnesota (Minneapolis, MN) for their histology and pathology support. The authors are extremely grateful to Dr. Joshua M. Lang of the Carbone Cancer Center at the University of Wisconsin (Madison, WI) and Dr. John K. Lee of the Fred Hutchinson Cancer Research Center (Seattle, WA) for their assistance with this project. This work was supported by a Prostate Cancer Foundation Young Investigator Award (to A.M. LeBeau), Prostate Cancer Foundation Challenge Awards (to A.M. LeBeau and S.M. Dehm), the Minnesota Partnership for Biotechnology and Medical Genomics Infrastructure Award (MNP IF 16.05; to A.M. LeBeau), NIH/NCI CA090628 Paul Calabresi K12 Award (to A.M. LeBeau), NIH/NCI R01 CA237272 (to A.M LeBeau), NIH/NCI R01 CA233562 (to A.M LeBeau), and NIH/NCI R01 CA174777 (to S.M. Dehm). The Prostate Cancer Biorepository Network is funded by the Department of Defense Prostate Cancer Research Program Awards (W81XWH-14-2-0182, W81XWH-14-2-0183, W81XWH-14-2-0185, W81XWH-14-2-0186, and W81XWH-15-2-0062).

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.