TSPO-targeted PET and Optical Probes for the Detection and Localization of Premalignant and Malignant Pancreatic Lesions

Pancreatic cancer is a leading cause of cancer-related deaths in the United States and abroad and is projected to be the second most common cause of death from cancer by 2030 (1–3). Despite growing awareness and targeted research efforts aimed at improved diagnosis and treatment, the overall 5-year survival rate for pancreatic cancer remains less than 10% (1). New ways to diagnose pancreatic cancer earlier are urgently needed and need to include predicting future cancer through identifying high-risk, premalignant precursor lesions. Identifying precursor lesions of various subtypes likely to progress to pancreatic cancer, such as high-grade pancreatic intraepithelial neoplasias (PanINs) or intraductal papillary mucinous neoplasms (IPMNs), could improve outcomes associated with this disease. The goal of this work was to develop a translational approach for precision pancreatic cancer care through noninvasive discrimination of high-risk indeterminate lesions and pancreatic cancer, as well as subsequent actionable surgery.

Cystic lesions of the pancreas are common incidental findings (incidentalomas) in routine cross-sectional CT and MRI (4–6). Though IPMNs are frequently nonmalignant at the time of diagnosis, over time, many are capable of progressing to invasive pancreatic ductal adenocarcinoma (PDAC; refs. 7, 8). Discovery of a pancreatic cyst results in a lifelong clinical management dilemma due to a lack of tools capable of determining its malignant potential. Conventional imaging tests rely on crude metrics such as cyst size, main pancreatic duct involvement, or solid components within the cyst, none of which provide molecular information or adequately predict malignant potential (9–12). As a result, preemptively, IPMNs account for a disproportionate number of pancreatic resections (∼30%), with attendant morbidity and mortality associated with aggressive surgery (13). Improved noninvasive imaging biomarkers of high-risk lesions that would benefit from surgical resection are needed.

Once a decision is made for pancreatic surgery, surgeons rely on visual and tactile information to differentiate tumor and surrounding nontumor tissue. However, challenges associated with this determination often lead to incomplete resection and recurrence. Image-guided surgery (IGS) is reported to improve resection margins in a number of cancer types (reviewed in refs. 14, 15), including pancreatic cancer (16–19). Given the current lack of imaging tools capable of selecting patients for surgery, coupled with the potential for improved outcomes stemming from IGS of pancreatic cancer, we envisioned a companion, multimodal approach centered around a single molecular target that combines noninvasive PET imaging with fluorescence-based IGS.

In this study, we evaluated translocator protein (TSPO) expression as a potential biomarker of high-risk premalignant lesions and pancreatic cancer. TSPO is an 18 kDa protein involved in cholesterol metabolism that is elevated in many human cancers, including glioma (20, 21), colon cancer (22, 23), and breast cancer (23–26). We evaluated TSPO expression in human pancreatic cancer through IHC staining of pancreatic tissue microarrays (TMAs) consisting of different grades of pancreatic cancer as well as premalignant cystic lesions (IPMNs). We also evaluated expression of TSPO in genetically engineered mouse (GEM) models of pancreatic cancer [Ptf1a-Cre;LSL-Kras^G12D/+ (KC), Ptf1a-Cre;LSL-Kras^G12D/+;Tgfbr2^+/− (KTC), and Ptf1a-Cre;LSL-Kras^G12D/+;Smad4^fl/fl (KSC)]. In both settings, TSPO levels were associated with high-grade premalignant lesions, including IPMNs and PanINs. We previously developed and utilized TSPO ligands for cancer imaging (20, 21, 25–38). Given the observed TSPO expression profiles, we evaluated one such PET imaging ligand, [¹⁸F]V-1008 (20, 29, 31, 33), for noninvasive imaging of pancreatic cancer within the context of two GEM models. Subsequently, to facilitate IGS, we developed a TSPO imaging ligand for near-IR (NIR) optical imaging based upon the V-1008 PET ligand pharmacophore. In a GEM model, the nanomolar affinity optical imaging ligand, V-1520, localized to premalignant pancreatic lesions and advanced tumors enabling preclinical, real-time IGS. Taken together, results presented here suggest that combined TSPO PET/IGS represents a novel, translational approach for precision pancreatic cancer care through discrimination of high-risk indeterminate lesions and actionable surgery.

A pancreatic TMA was generously provided by Dr. Michael Goggins (Johns Hopkins University, Baltimore, MD). A separate TMA was generated by the Vanderbilt Translational Pathology Shared Resource (TPSR) at Vanderbilt University Medical Center (VUMC, Nashville, TN). The TMAs contain cores from 40 normal pancreas samples, 39 chronic pancreatitis samples, 8 grade 1 pancreatic cancer samples, 33 grade 2 pancreatic cancer samples, and 13 grade 3 pancreatic cancer samples. All staining was performed by hand. Antigen retrieval was performed in pH 6.0 citrate buffer for 15 minutes at 105°C using a pressure cooker followed by cool down for 10 minutes at room temperature. Samples were quenched with 0.03% H₂O₂ with sodium azide for 5 minutes. The samples were then incubated with primary antibody against TSPO (PBR Goat Polyclonal Antibody, Novus Biologicals NB100–41398) at a dilution of 1:1,000 for 60 minutes. Detection was performed by incubating with Goat Probe (BioCare Medical) for 10 minutes and HRP Polymer (BioCare Medical, catalog no. GHP516) for 10 minutes followed by 3, 3′-diaminobenzidine (DAB+) chromogen for 5 minutes. All incubations were performed at room temperature. Positive tissue controls were included.

Slides were scored by three independent reviewers (J. Li, M.R. Hight., and H.C. Manning) and scores were confirmed by an expert gastrointestinal (GI) pathologist (C. Shi). TSPO IHC expression was assessed using an intensity score–derived methodology which defined no expression as zero, weak expression as 1, moderate expression as 2, and strong expression as 3.

Another TMA consisting of dysplastic IPMN samples from 65 patients was also utilized. This TMA was generated by the Vanderbilt TPSR at VUMC (Nashville, TN). Dysplasia scores ranged from 0 to 4. Patients with dysplasia scores of 0 or 1 were assigned to low-grade dysplasia, 2 to intermediate-grade dysplasia, and 3 or 4 to high-grade dysplasia. The highest dysplasia grade per patient was used. The TMA contained a total of 20 low-grade dysplasia samples, 23 intermediate-grade dysplasia samples, and 22 high-grade dysplasia samples. TSPO staining was performed as above and was scored as negative (0), weak (1 or 2), moderate (3 or 4), or strong (5 or 6). For statistical analyses, TSPO staining was categorized into three groups defined as weak (IHC score of 0, 1, or 2), moderate (IHC score of 3 or 4), and strong (IHC score of 5 or 6).

All animal procedures were in compliance with federal and institutional guidelines and were approved by the Vanderbilt Institutional Animal Care and Use Committee. Three validated genetically engineered mouse models which recapitulated genetic mutations relevant in human pancreatic cancer were generated and included: Ptf1a-Cre;LSL-Kras^G12D/+ (KC), Ptf1a-Cre;LSL-Kras^G12D/+;Tgfbr2^+/− (KTC), and Ptf1a-Cre;LSL-Kras^G12D/+;Smad4^fl/fl (KSC). Genetically engineered models more closely mimic the development and presentation of human disease in comparison to tumor xenografts (39–44). KC mice progress to higher grade PanIN lesions (42) and were chosen to model preinvasive disease. Advanced disease was modeled using KTC (43) and KSC (43, 44) mice. These mice are generated using the Cre-loxP system driven by the endogenous pancreas-specific locus Ptf1a (pancreatic transcription factor-1a) as described previously (42–44). Ptf1a plays an essential role in mammalian pancreatic development and differentiation (45). The stop codon between two loxP sites was excised using Cre-Lox recombination to activate the Kras mutation (Kras^G12D; ref. 43). Ptf1a-Cre, Ptf1a-Cre;Tgfbr2^fl/fl, and Ptf1a-Cre;Smad4^fl/fl mice were mated with LSL-Kras^G12D/+ mice to obtain the KC, KTC, and KSC mice, respectively, as well as littermate control mice. Animals possessing single vector (i.e., Kras, Cre, Tgfbr2, or Smad4 only) or wild genotypes were utilized as age-matched controls. Genotyping by PCR was performed 3 weeks after birth to identify the genetically engineered mice and controls needed for the experiments. TSPO PET and fluorescence imaging were evaluated at regular intervals that approached the median survival for each model. KTC and KSC mice were imaged up to 12 weeks of age to account for the rapid progression of tumors in these models (43, 44, 46). Animals were weighed and sacrificed for tissue collection shortly after in vivo imaging and white light photographs obtained during tissue resection to document relative location, size, and shape of the pancreas.

Pancreases were isolated from normal and genetically engineered mice at 4, 9, and ≥12 weeks of age. Tissues were fixed in cold 4% paraformaldehyde for 1–2 hours for normal tissues and 16–18 hours for tumor-bearing tissues. Duration of fixation was lengthened for tumor-bearing samples due to the increase in stromal density in these tissues. Following fixation, samples were embedded into paraffin blocks. Tissues were sectioned (5 μm thickness) and stained for TSPO. All staining was performed by hand. Antigen retrieval was performed on all samples in pH 6.0 Citrate Buffer for 15 minutes at 105°C using a pressure cooker followed by cool down for 10 minutes at room temperature. Samples were quenched with 0.03% H₂O₂ with sodium azide for 5 minutes. The samples were then incubated with primary antibody against TSPO (PBR Rabbit mAb, Novus Biologicals NBP1–95674) at a dilution of 1:400 for 60 minutes. Detection was performed by incubating with Dako K4003 EnVision+ System HRP labelled polymer Anti-Rabbit for 20 minutes followed by 3, 3′-diaminobenzidine (DAB+) chromogen for 5 minutes. All incubations were performed at room temperature. Additional tissue slices were stained using standard hematoxylin and eosin (H&E) and Masson's trichrome methods. Tissues were reviewed by an expert GI pathologist (C. Shi)

[¹⁸F]V-1008 was produced as published previously (20, 29). For small animal imaging studies, radiochemical purities were >99% and radiochemical specific activities were within the range of 170 ± 80 TBq/mmol. For blocking studies, V-8310, 7-chloro-N,N,5-trimethyl-4-oxo-3(6-fluoropyridin-2-yl)-3,5-dihydro-4H-pyridazino[4,5-b]indole-1-acetamide, was synthesized as reported previously (30).

NIR probe V-1520 was synthesized using a similar method to that previously described for an analogous tracer (38). Briefly, LI-COR IRDye 800CW NHS ester was purchased from LI-COR Biosciences. The dye (2 eq) and the unconjugated ligand were dissolved by stirring in dry DMSO in the presence of triethylamine (2 eq) in an argon-flushed vessel. The reaction was allowed to stir overnight, protected from room light, at room temperature. The product was purified by reverse phase high-performance liquid chromatography (HPLC) on a semiprep HPLC using a C18 column (Varian, 4 μm). The product was eluted with 50/50 acetonitrile/water as mobile phase. High-resolution mass spectrometry was performed to confirm synthesis of the desired product: C₇₆H₉₇N₇O₁₆S₄ m/z = 746.8010 (M+2H²⁺), found 746.8020; m/z = 1492.5947 (M+H⁺), found 1492.5958.

Animal handling and preparation for and during PET imaging studies were performed analogously to previously described methods (47). Upon study commencement, animals were anesthetized using 2% isofluorane anesthesia in 100% oxygen at 2 L/minute and kept warm throughout the duration of the PET scan. For PET imaging of KTC mice, animals were administered 10.4–11.8 MBq of PET imaging agent via intravenous injection and imaged using a dedicated Concorde Microsystems Focus 220 microPET scanner (Siemens Preclinical Solutions). PET images were acquired as 60-minute dynamic datasets initiated upon PET agent injection. Images were acquired at designated time points which corresponded with the age of the animal.

PET data were reconstructed using a three-dimensional (3D) ordered subset expectation maximization/maximum a posteriori (OSEM3D/MAP) algorithm. Dynamic data were binned into twelve 5-second (0–1 minutes) and fifty-nine 60-second (2–60 minutes) frames. The resulting 3D reconstructions had an x-y voxel size of 0.474 mm and interslice distance of 0.796 mm. For quantification of PET agent uptake, ASIPro software (Siemens Preclinical Solutions) was used to manually draw 3D regions of interest (ROIs) around the entire tumor volume and the measured counts converted to the percentage of the injected dose per gram of tissue (%ID/g). To aid in the localization of tumor-bearing pancreas, high-field anatomic MRI images were collected either immediately before or after PET imaging acquisitions. For initial evaluations, PET and MRI images were manually proportioned and overlaid using commercially available image processing software.

For PET imaging of KSC mice, animals were administered 6.6–9.7 MBq of PET imaging agent in 100 μL via retroorbital injection. The mice were then returned to their cages and fed ad libitum for 40 minutes. Mice were anesthetized with 2% isofluorane and imaged for 20 minutes in an Inveon microPET (Siemens Preclinical Solutions). All datasets were reconstructed using the MAP algorithm into 128 × 128 × 95 slices with a voxel size of 0.095 × 0.095 × 0.08 cm³ at a β value of 0.01. For anatomical CT images, immediately following the PET scans, the mice were imaged in a NanoSPECT/CT (Bioscan) at an x-ray beam intensity of 90 mAs with an x-ray energy of 45 kVp. The images were reconstructed at 170 × 170 × 114 with a voxel size of 0.4 × 0.4 × 0.4 mm³. To aid in the localization of tumor-bearing pancreas, high-field anatomic MRI were collected the day following PET imaging acquisitions. The MRI, PET, and CT images were loaded onto the image analysis tool Amide (www.souceforge.net). The PET and CT images were coregistered based on bed position. Then, the MR images were coregistered to the CT images visually based on biological fiducial markers such as kidneys. ROIs were drawn around the tumors in the PET scans.

For MRI acquisitions of KTC mice, animals were secured in a prone position in a 38-mm inner diameter radiofrequency coil and placed in a 7.0T (16 cm bore) Varian small animal imaging system. Throughout scanning, animals were warmed to 37°C using heated air flow and maintained under anesthesia using 2% isofluorane anesthesia in 100% oxygen at 2 L/minute. Following localization of the pancreas using multislice gradient echo images of all three imaging planes, T2-weighted fast-spin echo images were acquired over 22 slices in the axial and coronal planes using a 25.6 mm x 25.6 mm field of view, 1.0 mm slice thickness, and a 256 × 256 data matrix. Additional parameters included a 2-second repetition time, 36 ms effective echo time, 9 ms echo spacing, and an echo train length of 8. A pneumatic pillow was used to monitor the respiration cycle of the animals, as well as trigger the imaging acquisition to collect data at the same time point in the respiration cycle, to reduce motion-induced imaging artifacts.

For MRI acquisitions of KSC mice, mice were anesthetized via inhalation of 2%/98% isofluorane/oxygen. Animals were secured in the prone position in an in-house fabricated animal cradle with bite bar and placed in a 25-mm inner diameter radiofrequency coil. Animals were then placed in a 7T Bruker Avance III horizontal bore imaging system (Bruker BioSpin) for data collection. Respiration rate and internal body temperature were continuously monitored, and a constant body temperature of 37°C was maintained using heated air flow. For each animal, multislice gradient echo images were collected in all three imaging planes (axial, sagittal, and coronal) to localize the pancreas, with repetition time (TR) = 75 ms, echo time (TE) = 5 ms, slice thickness = 2 mm, flip angle = 35°, and an average of 4 acquisitions. Additional parameters include field of view (FOV) = 50 mm × 50 mm and data matrix = 128 × 128. Following localization of the pancreas, T2-weighted fast-spin echo images were acquired over 24 contiguous slices in the axial and coronal planes, with FOV = 25.6 mm × 25.6 mm, slice thickness = 1.0 mm, and data matrix = 256 × 256. Additional parameters include TR = 2 seconds, effective TE = 40.2 ms, echo train length = 8, echo spacing = 6.7 ms, and number of experiments = 16. A pneumatic pillow was used to monitor the respiration cycle of the mice, as well as trigger the imaging acquisition to collect data at the same time point in the respiration cycle, to reduce motion-induced imaging artifacts.

Genetically engineered KSC pancreatic cancer mice and age-matched littermate controls were used to evaluate uptake of [¹⁸F]V-1008 in normal pancreas versus preinvasive pancreatic lesions. Animals were administered 1.1–1.7 MBq of [¹⁸F]V-1008 in 100 μL of PBS via retroorbital injection. The mice were then returned to their cages and fed ad libitum for 60 minutes. Following 60 minutes of tracer uptake, the mice were euthanized and the pancreases were resected. Radioactivity and weights of the samples were determined using a Hidex automatic gamma counter (Lablogic). %ID/g values of each sample were calculated according to the weight and radioactivity of the samples.

The crystal structure of BcTSPO/PK11195 [PDB ID 4RYI (48)] was used as a target for ligand docking. Docking calculations were carried out using the Surflex-Dock method (49) from within Sybyl 2.1.1. Starting ligand coordinates were generated using CONCORD. The receptor was prepared and minimized using Sybyl protein structure preparation tools. The Surflex-Dock receptor protomol was generated on the basis of the PK11195 ligand atoms, using bloat and threshold parameters of 2.0 and 0.5, respectively. Docking was then carried out using the standard geomX parameters, with added starting conformers for each ligand fragment. The best pose we observed had a favorable total score of 13.2.

Binding affinity of V-1520 was determined using C6 glioma cell lysates as reported previously (20, 27, 32, 38). Lysates were incubated with N-(sec-butyl)-1-(2-chlorophenyl)-N-methyl-³H-isoquinoline-3-carboxamide ([³H]PK11195; final concentration = 6 nmol/L) and V-1520. The reaction was terminated by rapid filtration through a Brandel harvester and collection onto a filter presoaked with 0.3% polyethylenimine. Filters were then punched out into scintillation vials and bound radioactivity measured on a Beckman LS 6000 Scintillation Counter. All experiments were conducted in triplicate. Competition curves were analyzed with GraphPad Prism to obtain binding affinity (IC₅₀). K_i values were calculated using the equation K_i = IC₅₀/[1+(radioligand)/K_d], where IC₅₀ is determined from the competition curves, [radioligand] is the final concentration of [³H]PK11195 (6 nmol/L) and K_d is the dissociation constant for [³H]PK11195 (5 nmol/L).

NIR Probe V-1520 or free dye was dissolved in PBS to make a final concentration of 0.1 μmol/L. Mice were injected with 200 μL of solution via retroorbital injection. The mice were euthanized 18 hours postinjection and tissues were excised and washed in PBS. Excised tissues were imaged using Odyssey Imaging Systems (LI-COR Biosciences) under the 800 nm filter.

In vivo blocking studies were carried out in the KSC genetically engineered pancreatic cancer mice. Mice were switched to a low fluorescence diet [purified diet (Alfalfa-free), catalog no. 2020X, Envigo RMS Inc.] at least 3 days prior to imaging. V-8310 (7-chloro-N,N,5-trimethyl-4-oxo-3(6-fluoropyridin-2-yl)-3,5-dihydro-4H-pyridazino[4,5-b]indole-1-acetamide), a high-affinity pyridazinoindole ligand targeting TSPO, was synthesized as described previously (30). Stock solutions of the compounds were prepared in DMSO. The stock solutions were diluted in PBS to make the final concentrations for injection. Mice were administered 1 nmol of V-1520 in 100 μL via tail vein injection. For blocking studies, 100 nmol V-8310 (7-chloro-N,N,5-trimethyl-4-oxo-3(6-fluoropyridin-2-yl)-3,5-dihydro-4H-pyridazino[4,5-b]indole-1-acetamide; ref. 30) in 200 μL was injected 30 minutes prior to V-1520 via retroorbital injection. Mice were euthanized and the pancreases were collected 2 hours postinjection of V-1520. The pancreases were imaged on the Odyssey Imaging System using the 800 nm filter. Regions of interest were drawn around the pancreases and values are reported as fluorescence signal per area of the ROI.

In vivo imaging was carried out on the Pearl Small Animal Imaging System (LI-COR Biosciences). Mice were injected with 10 μmol/L V-1520 in 200 μL of PBS. Imaging was performed 18 hours postinjection. Animals were anesthetized using isofluorane and were positioned on the stage for imaging. An incision was made down the midline to expose the abdominal cavity. Fluorescence images were acquired using the 800 nm filter channel. Following imaging, the pancreases were isolated for histology as described above.

Genetically engineered KSC pancreatic cancer mice and age-matched littermate controls were used for experiments evaluating V-1520 in real-time intraoperative fluorescence guided surgeries. Mice were switched to a low fluorescence diet [purified diet (Alfalfa-free), catalog no. 2020X, Envigo RMS Inc.] at least 3 days prior to imaging. Mice were injected with 20 nmol of V-1520 in 100 μL of PBS intravenously and the surgery was performed 24 hours postinjection. A single, 20–30 mm long abdominal midline incision was made to expose pancreas areas. Tumor areas were exposed to the Curadel RP-1 OSN system. This system is combined with FLARE imaging systems to enable fluorescence imaging during surgery. Intraoperative fluorescence-guided surgery was visualized in real-time using both white light and fluorescence imaging with an 800 nm wavelength filter. Following imaging, pancreases were isolated, fixed, and embedded in paraffin blocks. Excised tissue blocks were imaged using Odyssey Imaging Systems (LI-COR Biosciences) under the 800 nm filter. Tissues were sectioned (5 μm thickness) and H&E was performed as above. Unstained slides were imaged using fluorescence microscopy as described below. Images of H&E slides were captured using a high throughput Leica SCN400 Slide Scanner automated digital image system from Leica Microsystems. Whole slides were imaged at 20× magnification to a resolution of 0.5 μm/pixel.

Zeiss Axio Observer Z1 Inverted Phase Contrast Fluorescence Microscope was used to capture fluorescence images. The microscope is equipped with a LED light source (Zeiss Colibri 7). The probe was visualized with a Cy7 filter (excitation: LP 710 nm, emission: LP 810 nm, beam splitter: 760 nm) and fluorescence signals were captured with a digital camera (Hamamatsu, Model: C11440–42U30) and are shown with a pseudocolor of red. 10× objective lens and an additional 10× eyepiece were used with total magnification of 100×. Images were further processed with the ZEN 2.3 software (Blue version, Zeiss).

Experimental replicates are reported as the mean ± SD. The association of TSPO expression among IPMN-positive and IPMN-negative patients as well as among different grades of dysplasia were summarized in frequency tables with statistical significance evaluated using the Pearson's χ² test. To determine subgroup differences in TSPO expression, a Kruskal–Wallis test was performed among all groups followed by pairwise comparisons using the Wilcoxon rank-sums test. Experiment-wise control of the type I error rate at 5% was maintained using the Bonferroni correction. Analysis of differences in expression for benign versus IPMN samples was performed using a Wilcoxon rank-sums test. The two-sample Student t test was applied for all other statistical comparisons using GraphPad Prism. For all tests, P values < 0.05 were considered statistically significant.

Logistic regression was used to model sensitivity, specificity, positive predictive values, negative predictive values, and accuracy for TSPO IHC staining of the TMAs to predict pancreatic cancer or IPMN positivity along the ROC curve. Sensitivity is defined as the true positive fraction. The false positive fraction is defined as 1-specificity. Positive predictive values are the probability that subjects with a positive screening test truly have the disease (either pancreatic cancer or IPMN). Negative predictive values are the probability that subjects with a negative screening test truly do not have the disease. The accuracy is the ratio of the sum of true positive and negative values divided by the sum of all positive and negative values. The ROC curve is a plot of the sensitivity (true positive fraction) versus 1-specificity (the false positive fraction) as the TSPO score changes. A 95% confidence interval for the AUC that excludes 0.5 suggests the marker has prognostic value. The points on ROC curves where sensitivity, specificity, and other indexes were presented are starred on their respective graphs.

TSPO expression was evaluated in PDAC in comparison with normal pancreas and chronic pancreatitis (CP) in a TMA of 133 patient samples. Overall, TSPO expression was found to be very low in normal pancreas tissue with modest staining in centroacinar cells and normal ducts (Fig. 1A). Low-level staining was also observed within the inflammatory infiltrate comprising chronic pancreatitis (CP). The mean IHC score of normal pancreas was 1.025 ± 0.660. Chronic pancreatitis (CP) exhibited slightly higher staining intensity with a mean IHC score of 1.385 ± 0.847 (Fig. 1B, not significant relative to normal pancreas), localized to inflammatory cells as expected (50–54). In contrast, pancreatic cancer exhibited robust TSPO expression, with mean scores of 2.625 ± 0.518 (P < 0.00001 relative to normal pancreas, P = 0.00022 relative to chronic pancreatitis), 2.576 ± 0.663 (P < 0.00001 relative to normal pancreas, P < 0.00001 relative to chronic pancreatitis), and 1.923 ± 0.760 (P = 0.00036 relative to normal pancreas) for grades 1–3, respectively (Fig. 1A and B). Using a cutoff for IHC score of 1, we obtained a sensitivity of 0.870, specificity of 0.658, positive predictive value of 0.635, negative predictive value of 0.881, and accuracy of 0.744. The area under the ROC curve is 0.85 with a 95% confidence interval of 0.79 to 0.92 (Fig. 1C).

In another group of patient samples, we evaluated TSPO expression in IPMN relative to benign lesions (n = 65). IPMN samples exhibited significantly higher TSPO expression compared with benign lesions (Wilcoxon rank-sum test, P = 0.00004; Fig. 2A and B). The mean IHC scores of benign lesions and IPMNs were 1.737 ± 1.661 and 3.717 ± 1.682, respectively. IPMN-positive patients had significantly higher TSPO expression than IPMN-negative patients (Supplementary Table S1; Pearson's χ² test, P < 0.0001). Seventy-eight percent (36/46) of IPMN-positive patients had moderate to strong TSPO expression compared with 21% (4/19) of IPMN-negative patients. Furthermore, TSPO expression was found to correlate directly with the grade of dysplasia across low-, intermediate-, and high-grade samples (Fig. 2; Supplementary Table S2; Pearson's χ² test, P < 0.0001). The mean TSPO IHC scores for low-, intermediate-, and high-grade dysplasia were 1.700 ± 1.625, 3.130 ± 1.456, and 4.455 ± 1.595, respectively. Fifty-five percent (12/22) of patients with high-grade dysplasia had strong TSPO expression compared with only 16% (7/43) of low- and intermediate-grade dysplasia samples. Using a cutoff for IHC score of 2, we obtained a sensitivity of 0.783, specificity of 0.789, positive predictive value of 0.900, negative predictive value of 0.600, and accuracy of 0.785. The area under the ROC curve is 0.80 with a 95% confidence interval of 0.68 to 0.92 (Fig. 2C).

GEM models of pancreatic cancer were used to further evaluate TSPO expression as a function of pancreatic cancer initiation and progression as shown in Supplementary Table S3. Each of the selected models [Ptf1a-Cre;LSL-Kras^G12D/+ (KC), Ptf1a-Cre;LSL-Kras^G12D/+;Tgfbr2^+/− (KTC), and Ptf1a-Cre;LSL-Kras^G12D/+;Smad4^fl/fl (KSC)] recapitulate relevant features of human pancreatic cancer as described previously (41–44). To model low-grade dysplasia, KC mice were utilized, which progress to higher grade PanIN lesions (42) and simulate preinvasive disease. Advanced disease was modeled using KTC (43) and KSC (43, 44) mice. Ptf1a-Cre mice were used as a control to evaluate normal pancreatic tissue. As shown in Fig. 3, TSPO expression was low in normal mouse pancreas (Ptf1a-Cre) and low-grade PanINs (KC). In contrast, high-grade PanINs and PDACs observed in similarly aged KTC mice demonstrated significantly higher TSPO staining. In KSC mice, both cystic and early invasive lesions exhibited higher TSPO levels compared with normal pancreas.

Noninvasive imaging studies using [¹⁸F]V-1008, a high-affinity TSPO PET ligand (20, 29, 31, 33), were conducted in KTC mice at two stages of disease progression. Focal, elevated uptake of [¹⁸F]V-1008 was observed in the pancreases of 7–8-week-old KTC mice (Supplementary Fig. S1A). Histologic evaluation of pancreas tissues collected from imaged 7–8-week-old mice confirmed the presence of abundant low- to moderate-grade dysplasia, which is known to progress to full PDAC by 20 weeks in this model (43). Mice of the same genotype that were 11–12 weeks old, which histologically exhibited more advanced disease, were also imaged. In these mice, [¹⁸F]V-1008 also showed increased uptake in tumor tissue (Supplementary Fig. S1B). We subsequently performed imaging studies in KSC mice. PET imaging indicated accumulation of the TSPO PET tracer in the pancreases of these mice with histologic evaluation of tissue confirming the presence of preinvasive cystic lesions (Fig. 4A). Uptake of [¹⁸F]V-1008 in preinvasive lesions was compared with uptake in normal pancreases by measuring radioactivity in excised tissues following injection of the tracer in mice. Preinvasive lesions had significantly higher uptake than normal pancreases (Fig. 4B, P < 0.01) indicating the potential of [¹⁸F]V-1008 to distinguish early disease.

Given the promising imaging results obtained with [¹⁸F]V-1008, we leveraged our prior experience with TSPO ligands for optical imaging (22, 25, 26, 34–38) and synthesized a novel TSPO ligand for NIR fluorescence imaging. To do so, we conjugated the V-1008 pyrazolopyrimidine pharmacophore to an FDA-approved NIR dye (55) through an 8-carbon linker, hypothesizing that such a tracer could facilitate IGS in this setting (Fig. 5A). Termed V-1520, the resulting probe exhibited high affinity for TSPO (K_i = 4.47 nmol/L; Fig. 5B). Similar to our previous observations with analogous tracers (38), computational modeling demonstrated that the ligand pharmacophore readily fit the binding domain formed by the five transmembrane α-helices, while the 8-carbon linker allowed the nonbinding dye unit to reside outside the TSPO-binding lobby (Fig. 5C).

Having demonstrated TSPO binding in vitro, we next evaluated the tissue biodistribution and accumulation of V-1520 ex vivo following in vivo administration. Genetically engineered mice (KSC) and age-matched littermate controls were intravenously administered V-1520. Following an uptake period, the mice were euthanized for collection and ex vivo imaging of various organs (Supplementary Figs. S2 and S3). Normal tissue accumulation of V-1520 agreed with that previously observed for TSPO ligands (21, 25, 29), including very modest accumulation in the healthy pancreas of control mice. Critically, accumulation of V-1520 was approximately 10-fold higher in pancreases collected from both 4-week-old (low-grade dysplasia) and 9-week-old (high-grade dysplasia) genetically engineered mice compared with healthy pancreas (Supplementary Fig. S2C).

We next explored the in vivo specificity of V-1520 and potential utility for in vivo imaging using KSC mice. Similar to the biodistribution studies, V-1520 accumulated in tumor tissue, while unconjugated free dye, IRDye800CW, did not (Supplementary Fig. S4). In addition, the free dye was unable to distinguish the malignant pancreas from normal pancreas as shown by comparing imaging data from genetically engineered and littermate control mice. Furthermore, tumor accumulation of V-1520 could be blocked to background levels by pretreating KSC mice with another high-affinity TSPO ligand (30; Supplementary Fig. S5).

To simulate IGS using V-1520, KSC mice were administered tracer and imaged in multiple formats. First, mice were surgically opened and imaged immediately following sacrifice using a preclinical mouse scanner (Supplementary Fig. S6A). Histologically confirmed, TSPO-positive, low-grade pancreatic dysplasia with early invasion (Supplementary Fig. S6B) was well discriminated from background tracer uptake (Supplementary Fig. S6A bottom, green arrow). We next carried out real-time fluorescence imaging of live KSC mice and age-matched littermate controls following V-1520 exposure during surgery using an intraoperative imaging system. Representative images taken during the surgery are shown in Fig. 6 and movies are shown in Supplementary Movies S1 and S2. Strong fluorescence signal was observed in lesions arising in KSC mice but not in the pancreases of healthy controls. Ex vivo imaging of the pancreases isolated from these mice confirmed the fluorescence observed intraoperatively and also demonstrated the heterogeneity of probe uptake which was similarly observed with TSPO IHC. Fluorescence microscopy indicated uptake of V-1520 in cancer cells as well as tumor-associated inflammatory cells present in the stroma of KSC mice whereas minimal uptake was seen in control pancreases (Supplementary Fig. S7).

New ways to diagnose pancreatic cancer at earlier stages are urgently needed. Prompting this study, we envisioned an opportunity to deploy multimodality molecular imaging to predict pancreatic cancer by prioritizing high-risk, premalignant precursor lesions in at-risk patients. While unlikely to flag suspicious PanINs due to spatial resolution limits coupled with their relative size (4, 56), cystic lesions are frequently discovered in routine cross-sectional CT and MRI (4–6) and, thus, represent potentially low-lying fruit for disambiguation. Here, we present preclinical data suggestive of a workflow for pancreatic cancer care initiated following identification of a suspicious indeterminate pancreatic lesion.

We identified TSPO as a biomarker of premalignant pancreatic lesions and pancreatic cancer. To take advantage of a dual-modality imaging approach, we utilized a TSPO-targeted ligand to develop two imaging agents: [¹⁸F]V-1008 for PET imaging and V-1520 for optical fluorescence imaging. We anticipate that [¹⁸F]V-1008 can be utilized to noninvasively image patients with indeterminate lesions identified by CT or MRI. Our data suggest that uptake of this tracer would indicate that the patient has high-grade lesions that are likely to progress to cancer and thus would enable selection of patients who should be candidates for surgery. V-1520 could subsequently be utilized as part of the surgery to enable visualization of tumors and other precursor lesions intraoperatively facilitating improved resection. This study described the preclinical evaluation of these two tracers as a step toward clinical translation. We performed in vivo imaging studies and uptake assays of excised tissues using [¹⁸F]V-1008 in genetically engineered mouse models of pancreatic cancer. We also evaluated V-1520 in vitro, ex vivo, and in vivo. [¹⁸F]V-1008 and V-1520 have high affinity for TSPO and showed utility in imaging of premalignant lesions and pancreatic cancer in genetically engineered mouse models. Importantly, V-1520 enabled visualization of pancreatic tumors intraoperatively.

TSPO expression in pancreatic cancer has not been previously reported despite the fact that it is known to be highly expressed in many human cancers, including glioma (20, 21), colon cancer (22, 23), and breast cancer (23–26). We first confirmed the relevance of TSPO in pancreatic cancer by evaluating the correlation between disease grade and TSPO expression in PDAC through IHC of TMAs consisting of human clinical samples from normal pancreas, precursor lesions, and advanced tumors. We observed high expression of TSPO in premalignant pancreatic lesions (IPMNs) as well as in all different grades of pancreatic cancer while there was minimal expression in normal pancreas and chronic pancreatitis. The low expression observed in chronic pancreatitis in comparison to tumor samples indicates that TSPO may be able to differentiate between inflammation, which is characteristic of chronic pancreatitis, and cancer. Using ROC analysis, we demonstrate the potential of TSPO to serve as a biomarker with high sensitivity, good negative predictive value, and high accuracy to distinguish pancreatic cancer from normal pancreas and chronic pancreatitis. On the basis of the limited tissue samples evaluated in this study, some potential for false positives cannot be ruled out in this setting given the moderate values (∼65%) obtained for specificity and positive predictive value. However, more the focus of this work, ROC analysis demonstrated that TSPO could discriminate IPMN from benign lesions with high sensitivity, specificity, positive predictive value, and accuracy. A few lesions classified as IPMN exhibited similar TSPO expression to benign lesions; given the subtleties of lesion discrimination in this space, we cannot exclude the possibility that a few lesions were incorrectly classified at pathology. In limited specimens, we observed a negative predictive value of 0.600, suggesting a possible sub-class of TSPO-negative IPMNs. Additional studies using more comprehensive tissue collections could further elaborate TSPO expression profiles; yet similarly, our study highlights the advantages of noninvasive imaging, which does not suffer from IHC's inherent limitations of sampling and heterogeneity.

Importantly, TSPO levels were not elevated in KC mice suggesting that TSPO is not merely associated with early initiation. Rather, elevated TSPO levels appear to reflect further genetic transformations beyond mutant Kras that lead to pancreatic cancer progression. As reported previously (42), KC mice develop PanINs around 1 year of age, which progress very slowly (and rarely) to PDAC. Combining the Kras mutation further with loss of Tgfbr2 (43) or Smad4 (43, 44), whose inactivation has been observed in invasive human PDAC (43), resulted in elevated TSPO levels that were detectable by imaging and, subsequently, rapid tumor formation. As aggressive pancreatic cancer models, KTC and KSC mice develop preinvasive lesions early in life that progress to pancreatic cancer, with nearly all mice succumbing to disease by 20 weeks of age. In both advanced GEM models, early, intermediate, and advanced lesions all demonstrated elevated expression levels of TSPO. Thus, as a biomarker, our data suggest that TSPO may be ideally suited to reflect early lesions on a rapid trajectory to invasive cancer, as well as identify cancer itself.

In summary, we have presented preclinical proof-of-principle studies illuminating a new multimodality strategy that enables noninvasive imaging of pancreatic cancer, prioritization of high-risk, premalignant pancreatic lesions, and subsequent IGS. Combining the subpharmacologic dosages of the radio- and optical imaging tracers for imaging efficacy with the proven safety profiles of PET and optical imaging suggests the feasibility of translating this PET/IGS approach to clinical evaluation.

A.S. Cohen reports grants from NIH, The Kleberg Foundation, and The Lustgarten Foundation during the conduct of the study. J. Li reports grants from NIH, The Kleberg Foundation, The Lustgarten Foundation, and Vanderbilt Center for Molecular Probes during the conduct of the study, and is listed as a coinventor on a provisional patent application on pyridazinoindole compounds for imaging that is owned by Vanderbilt University and licensed to Vanderbilt University. E.T. McKinley reports grants from NIH during the conduct of the study. A. Fu reports grants from NIH, The Kleberg Foundation, and The Lustgarten Foundation during the conduct of the study. G.D. Ayers reports grants from NIH during the conduct of the study. N.B. Merchant reports grants from SOBI Pharmaceuticals and other from SOBI Pharamceuticals (advisory board) outside the submitted work. H.C. Manning reports grants from NIH, Lustgarten Foundation, and Kleberg Foundation during the conduct of the study, and is listed as a coinventor on a provisional patent application on imaging Diagnostics that is owned by Vanderbilt University Medical Center and is currently unlicensed. No potential conflicts of interest were disclosed by the other authors.

A.S. Cohen: Formal analysis, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. J. Li: Formal analysis, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. M.R. Hight: Formal analysis, investigation, visualization, methodology, writing-original draft. E. McKinley: Formal analysis, investigation, methodology. A. Fu: Investigation, visualization, methodology. A. Payne: Writing-original draft. Y. Liu: Investigation, visualization, methodology, writing-original draft. D. Zhang: Investigation, methodology. Q. Xie: Investigation, visualization, methodology, writing-review and editing. M. Bai: Resources, supervision, methodology, writing-original draft, project administration, writing-review and editing. G.D. Ayers: Formal analysis, visualization, methodology, writing-review and editing. M.N. Tantawy: Formal analysis, investigation, visualization, writing-review and editing. J.A. Smith: Resources, formal analysis, investigation, visualization, methodology, writing-original draft. F. Revetta: Investigation. M.K. Washington: Supervision, methodology. C. Shi: Resources, formal analysis, supervision, methodology. N. Merchant: Conceptualization, supervision, writing-review and editing. H.C. Manning: Conceptualization, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

The authors acknowledge Fuxue Xin and Daniel Colvin, PhD for assistance with PET and MR imaging and data analysis, Dr. Robert Coffey and Anne Powell, PhD for assistance with breeding, and the VUMC Center for Small Animal Imaging and Radiochemistry Core facilities for assistance with PET imaging and synthesis of the PET radiotracer. The authors thank the Translational Pathology Shared Resource at Vanderbilt University Medical Center (Nashville, TN) for assistance with the IHC staining. The Translational Pathology Shared Resource is supported by NCI/NIH Cancer Center Support grant 2P30 CA068485-14 and the Vanderbilt Mouse Metabolic Phenotyping Center grant 5U24DK059637-13. Whole-slide imaging was performed in the Digital Histology Shared Resource at Vanderbilt University Medical Center (Nashville, TN, https://www.vumc.org/dhsr/). We also acknowledge Dr. Michael Goggins (Johns Hopkins University, Baltimore, MD) for providing the pancreatic cancer TMA. The authors acknowledge funding from K25 CA127349 (NIH NCI, to H.C. Manning), 1R01 CA163806 (NIH NCI, to J. Li, M.R. Hight, E. McKinley, H.C. Manning), the Vanderbilt Ingram Cancer Center (VICC) support grant (NIH NCI P30 CA068485, to A.S. Cohen, G.D. Ayers, M.N. Tantawy, F. Revetta, M.K. Washington, H.C. Manning), Vanderbilt University Medical Center's Digestive Disease Research Center [NIH National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) P30 DK058404, to A.S. Cohen, H.C. Manning], the VICC Specialized Program of Research Excellence (SPORE) in Gastrointestinal Cancer (NIH NCI P50 CA236733, to A.S. Cohen, A. Fu, M. Bai, G.D. Ayers, F. Revetta, M.K. Washington, H.C. Manning), The Kleberg Foundation (to H.C. Manning), The Lustgarten Foundation (to J. Li, M.R. Hight, E. McKinley, H.C. Manning), and the Vanderbilt Center for Molecular Probes (to H.C. Manning) This work was supported by NIH grants 1S10OD016245-01 for the Small-Animal PET Scanner, housed in the Vanderbilt Center for Small Animal Imaging, and 1S10OD019963-01A1 for the GE TRACERlab FX2 N and Comecer Hotcell, housed in the Vanderbilt Center for Molecular Probes—Radiochemistry Core.

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.