Site-specifically labeled CA19.9-targeted immunoconjugates for the PET, NIRF, and multimodal PET/NIRF imaging of pancreatic cancer

Significance

Pancreatic cancer will soon be the second leading cause of cancer deaths annually, yet no adequate molecular imaging tools exist to aid in the staging, monitoring, and treatment of the disease. Here we describe the development and preclinical evaluation of three unique immunoconjugates for positron emission tomography, near-infrared fluorescent optical imaging, and multimodal imaging of pancreatic ductal adenocarcinoma (PDAC). The probes were developed using a site-specific, chemoenzymatic methodology that is robust, reproducible, and modular. By targeting CA19.9, the most abundant antigen in >90% of PDAC tumors, we were able to obtain high-quality images in multiple murine models of PDAC, suggesting these constructs could be the core of a molecular imaging toolkit aimed at improving outcomes for patients with PDAC.

Abstract

Molecular imaging agents for preoperative positron emission tomography (PET) and near-infrared fluorescent (NIRF)-guided delineation of surgical margins could greatly enhance the diagnosis, staging, and resection of pancreatic cancer. PET and NIRF optical imaging offer complementary clinical applications, enabling the noninvasive whole-body imaging to localize disease and identification of tumor margins during surgery, respectively. We report the development of PET, NIRF, and dual-modal (PET/NIRF) imaging agents, using 5B1, a fully human monoclonal antibody that targets CA19.9, a well-established pancreatic cancer biomarker. Desferrioxamine (DFO) and/or a NIRF dye (FL) were conjugated to the heavy-chain glycans of 5B1, using a robust and reproducible site-specific (ss) labeling methodology to generate three constructs (^ssDFO-5B1, ^ssFL-5B1, and ^ssdual-5B1) in which the immunoreactivity was not affected by the conjugation of either label. Each construct was evaluated in a s.c. xenograft model, using CA19.9-positive (BxPC3) and -negative (MIAPaCa-2) human pancreatic cancer cell lines. Each construct showed exceptional uptake and contrast in antigen-positive tumors with negligible nonspecific uptake in antigen-negative tumors. Additionally, the dual-modal construct was evaluated in an orthotopic murine pancreatic cancer model, using the human pancreatic cancer cell line, Suit-2. The ^ssdual-5B1 demonstrated a remarkable capacity to delineate metastases and to map the sentinel lymph nodes via tandem PET-computed tomography (PET/CT) and NIRF imaging. Fluorescence microscopy, histopathology, and autoradiography were performed on representative sections of excised tumors to visualize the distribution of the constructs within the tumors. These imaging tools have tremendous potential for further preclinical research and for clinical translation.

Pancreatic ductal adenocarcinoma (PDAC) is currently the fourth leading cause of cancer mortality and is expected to surpass both colorectal and breast cancer in total annual deaths by 2030 (1, 2). Surgical resection of the pancreas is the only curative treatment, but the presence of metastases precludes over 80% of patients from resection ab initio (3). The overall 5-y survival rate is ∼5%, and for those who qualify for surgical resection, the 5-y survival rate is only 25% due to the high incidence of undiscovered metastases (4, 5). Further complicating this dire situation, patients with PDAC are regularly misdiagnosed or understaged, confounding treatment strategies and preventing proper enrollment in clinical trials. Many of these problems could be avoided and outcomes improved if adequate clinical tools for diagnosing, staging, and treating PDAC were available.

Positron emission tomography (PET) is a promising technological platform for detecting, staging, and monitoring the progression or regression of many solid tumors, including PDAC. Optical imaging is a complementary platform that makes possible the accurate identification of tumor tissue in an intraoperative setting, which was recently demonstrated in human patients with ovarian cancer (6). Currently, the only Food and Drug Administration approved imaging agent for PDAC is 2-deoxy-2-[¹⁸F]-fluoro-D-glucose (FDG). FDG PET imaging relies on increased tumor metabolism relative to nonmalignant cells (Warburg effect) (7). However, FDG has numerous shortcomings when it comes to PDAC, including unreliable detection of small primary lesions (<7 mm) (8) or liver metastases (<1 cm) (9), an inherent inability to discriminate between benign disease (i.e., pancreatitis) and malignancy (10), and decreased tumor avidity for FDG upon chemo- or radiation therapy (11). The development of an arsenal of imaging tools, particularly a dual-modal imaging agent that seamlessly incorporates the advantages of both PET and optical imaging, could definitively improve the outcomes in patients with PDAC.

Monoclonal antibodies (mAbs) can provide the necessary specificity, sensitivity, and flexibility for the development of such tools. PET imaging with a radiolabeled mAb (immunoPET) would enhance our ability to noninvasively detect small lesions and slowly growing epithelial cancers (12–14). Near-infrared fluorescent (NIRF) dyes are particularly attractive in intraoperative applications because they show good tissue penetration (up to 1 cm) and low background from autofluorescence. Optical imaging with a NIRF-labeled mAb would allow surgeons to precisely identify tumor margins during resection, ensuring minimal healthy tissue is removed and that no residual tumor tissue is overlooked. Technological advancements in the clinic are now at a stage that allows clinical translation into humans, providing renewed impetus for the preclinical development of tools for NIRF imaging (15, 16).

CA19.9 (also known as sialyl Lewis^a) is a ligand for epithelial leukocyte adhesion molecules, and its overexpression is a key event in invasion and metastasis of many cancers, including PDAC (17). CA19.9 is an attractive target for imaging of PDAC because it is the most highly expressed tumor antigen (18, 19) and is minimally expressed in healthy pancreas tissue (20). In fact, the diagnosis of PDAC is often aided by the detection of elevated levels of circulating CA19.9 (21, 22). The promise of CA19.9 as a biomarker of PDAC led to the initiation of several antibody discovery programs (14, 20, 23) and the development of the fully human mAb 5B1, which binds an extracellular epitope of CA19.9 with low nanomolar affinity (23). Recently, we demonstrated that CA19.9 could serve as a target for immunoPET imaging of PDAC, even in the context of circulating antigen (24). Based on those results, we set out to improve and expand upon the usefulness of 5B1 in the context of PDAC imaging.

The canonical methodology for the development of mAbs for PET and/or optical imaging suffers from several shortcomings that are a consequence of the indiscriminate conjugation of chelators or dyes to nucleophilic amino acids. Those shortcomings include the loss of immunoreactivity due to conjugation at the antigen-binding region, random conjugation that leads to poorly defined constructs, and an intrinsic lack of reproducibility, as well as laborious, costly optimization of each novel construct. The combination of glycan engineering and bioorthogonal “click” chemistry has proved a successful strategy for conjugating molecules distal to the antigen-binding region of mAbs in a manner that is highly specific and reproducible, circumventing the aforementioned problems (25, 26). Specifically, using a site-specific conjugation strategy for affixing chelator and/or dye molecules via the heavy chain glycans leads to well-defined, robust immunoconjugates in a highly reproducible manner that requires minimal optimization and results in a minimal loss of immunoreactivity. Furthermore, conjugation via the heavy chain glycans offers an exceptional opportunity to construct site-specific, dual-modal immunoconjugates, which are otherwise challenging to develop using traditional conjugation methodology.

Herein, we describe the development of three distinct immunoconjugates that were site-specifically conjugated with DFO for radiolabeling with ⁸⁹Zr and PET imaging, a NIRF dye for optical imaging, or a dual-labeled construct with both DFO and NIRF dye combining the advantages of PET and NIRF into a single construct. Using the site-specific, bioorthogonal conjugation strategy produced well-defined constructs that retained high levels of immunoreactivity compared with their nonspecifically labeled counterparts. The dual-labeled 5B1 construct, in particular, showed excellent uptake in murine models of PDAC, including delineation of small metastases and dissemination of antigen to sentinel lymph nodes in an orthotopic model.

Results and Discussion

Site-Specific Modification of 5B1.

A recently developed chemoenzymatic strategy for the site-specific modification of antibodies was used in the creation of all of the 5B1 immunoconjugates (Fig. 1A). This methodology harnesses enzymatic transformations and the strain-promoted azide–alkyne cycloaddition reaction (Fig. S1) to specifically modify the biantennary N-linked oligosaccharide chains on the heavy chain of the antibody. By specifically targeting the heavy chain glycans for bioconjugation, this methodology prevents the inadvertent conjugation of payloads to the antigen-binding domains of the antibody—and the consequent reduction in immunoreactivity—that can occur when using non-site-specific, amine-reactive bifunctional molecules. This site-specific modification approach is composed of three steps: (i) the removal of the terminal galactose residues of the heavy chain glycans using β-1,4-galactosidase; (ii) the attachment of an azide-bearing galactose sugar (GalNAz) to the heavy chain glycans, using a promiscuous galactosyltransferase, GalT(Y289L); and (iii) the strain-promoted click conjugation of payload-bearing dibenzocyclooctyne (DIBO) moieties to the azide-bearing galactose residues. In some cases, when a chelator has been appended to the antibody to create a radiolabeling precursor, an additional step is required: (iv) the radiometalation of the immunoconjugate.

Fig. 1.

Strategies for assembling the 5B1 immunoconjugates and their subsequent characterization. (A) The enzyme-mediated, site-specific modification of the heavy chain glycans followed by the click chemistry-mediated installation of fluorophores and/or chelators. (B) The biochemical analysis of the three reported constructs was carried out via SDS/PAGE gel. Unmodified 5B1 (lanes 1 and 5), ^ssDFO-5B1 (lanes 2 and 6), ^ssdual-5B1 (lanes 3 and 7), and ^ssFL-5B1 (lanes 4 and 8) are shown after treatment with PNGaseF (lanes 1–4) or without treatment (lanes 5–8). The residual PNGaseF (lanes 4–8, 36.5 kDa) is indicated with a black arrow and molecular weight standards are in the first and last lanes (MW).

Fig. S1.

The highly specific, orthogonal chemistry of the click reaction between the UDP-GalNAz and DIBO/DBCO-chelator/dye conjugates is illustrated.

For the investigation at hand, ⁸⁹Zr-labeled, fluorophore-labeled, and dual-labeled 5B1 immunoconjugates were synthesized using DIBOs bearing either the ⁸⁹Zr⁴⁺ chelator desferrioxamine (DIBO-DFO) or a fluorophore with a tissue-penetrating, near-infrared 800-nm emission (DIBO-FL). Over the course of the procedure, the multistep yields for site-specific DFO-5B1 (^ssDFO-5B1), ^ssFL-5B1, and ^ssDual-5B1 were 75 ± 15%, 78 ± 14%, and 84 ± 16%, respectively (n = 3). The degree of labeling (DOL) of the ^ssFL-5B1 and ^ssDual-5B1 were 1.5 ± 0.1 and 1.1 ± 0.1 fluorophores per mAb, respectively. Notably, this value is well below the degree of labeling previously obtained using this methodology with other dyes and payloads (∼3.5 moieties per mAb) and may be the result of the large and hydrophobic nature of the near-infrared fluorophore. However, these DOL values are ideal as it has been shown that as the ratio of fluorophore to mAb increases toward or beyond 2:1, the uptake of the tracer in nontarget tissues, particularly the liver, increases, leading to decreased tumor uptake and lower tumor to background ratio (27). Additionally, when the ratio of fluorophore to mAb increases beyond 2:1, there is a risk of self-quenching that can also lower the efficiency of tracer.

The site-specific nature of the bioconjugation was confirmed using SDS/PAGE experiments (Fig. 1B). A distinct increase in the molecular weight (M_r) of the heavy chain of the site-specifically modified immunoconjugates can be seen relative to that of the unmodified 5B1 (Fig. 1B, lanes 2–5). Importantly, a similar shift is not observed in the M_r of the light chain, showing that the modification occurs only on the heavy chain of the IgG. Further evidence for the glycan-specific nature of the bioconjugation is provided by treatment of the immunoconjugates with PNGaseF, an amidase that specifically cleaves between the asparagine residue of the Fc domain and the innermost sugar of the heavy chain glycans. As expected, SDS/PAGE experiments reveal that PNGaseF treatment has no effect on the molecular weight of the light chains of any of the 5B1 constructs. However, PNGaseF treatment produces marked downward shifts in the molecular weight of the heavy chains of the site-specifically labeled immunoconjugates. Importantly, the heavy chains of all of the immunoconjugates—unmodified 5B1, ^ssDFO-5B1, ^ssFL-5B1, and ^ssDual-5B—all shift to the same molecular weight (Fig. 1B, lanes 6–8). Taken together, these two sets of experiments clearly illustrate that the modification of the 5B1 antibody occurs site specifically on the heavy chain glycans.

Radiolabeling and in Vitro Evaluation of DFO-Modified Constructs.

After confirming that the site-specific modification was successful, the next step was to determine whether the modifications offered an improvement over the traditional nonspecific labeling strategy. To do so, the first step was radiolabeling the two DFO-conjugated constructs, ^ssDFO-5B1 and ^ssdual-5B1, with ⁸⁹Zr and to then assess those radiolabeled constructs in vitro. To that end, we used well-established methods for radiolabeling biomolecules with ⁸⁹Zr in a neutralized oxalate solution to generate ⁸⁹Zr-^ssDFO-5B1 and ⁸⁹Zr-^ssdual-5B1 (13, 28, 29).

For a direct comparison with the previously reported, nonspecifically modified ⁸⁹Zr-DFO-5B1, we considered the specific activity, averaged over multiple radiolabeling experiments, of the purified bioconjugates as the primary benchmark. The average specific activity of ⁸⁹Zr-^ssDFO-5B1 (n = 5) was 5.1 ± 1.1 mCi/mg and for ⁸⁹Zr-^ssdual-5B1 (n = 5) it was 1.9 ± 0.7 mCi/mg. In every radiolabeling experiment, the radiochemical purity of the purified constructs was >98%. Although both of the site-specifically labeled constructs had a lower specific activity than the nonspecifically labeled construct, which touted a rather impressive 12.1 ± 1.14 mCi/mg, the specific activities were in a suitable range for in vivo experiments. In fact, previous studies with 5B1 had shown that lowering the effective specific activity of 5B1 was beneficial, and in fact necessary, in the context of murine models that shed CA19.9 from the site of the primary tumor into the bloodstream. In such cases, it is important to inject enough of the radiotracer to ensure that a sufficient amount can reach and bind its target at the tumor even if some is sequestered in the blood. Taking this into consideration, the lower specific activity of the site-specifically modified 5B1 constructs is expected to circumvent the need to add “cold” antibody in the context of shed antigen.

A second and perhaps more important benchmark for comparing the site-specific and nonspecific conjugation strategies is the effect on the constructs’ antigen binding. The site-specific strategy ensures that modification of 5B1 occurs distal to the antigen-binding site whereas nonspecific conjugation runs the risk of appending a chelator or fluorophore proximal to the antigen-binding site, thereby disrupting the ability of 5B1 to bind CA19.9. So, it makes sense to expect that the immunoreactivity would be improved using the site-specific methodology. To assess the immunoreactivity of ⁸⁹Zr-^ssDFO-5B1 and ⁸⁹Zr-^ssdual-5B1, a well-established in vitro assay (30) was performed using both BxPC3 and MIAPaCa-2 cells, which are CA19.9 positive and CA19.9 negative, respectively. Our analysis showed that the immunoreactivity of the ⁸⁹Zr-^ssDFO-5B1 to BxPC3 was in excess of 98% whereas the immunoreactivity of ⁸⁹Zr-^ssdual-5B1 was in excess of 90%. These results confirmed that the site-specific labeling strategy yielded constructs with improved immunoreactivity relative to the nonspecifically modified ⁸⁹Zr-DFO-5B1, which had an immunoreactivity of 72.4 ± 1.1% (24). By improving the immunoreactivity by more than 20%, we expect that the site-specifically modified constructs will offer enhanced in vivo behavior compared with nonspecifically modified ⁸⁹Zr-DFO-5B1. We complemented these studies with a cell-based 96-well plate binding study of the ⁸⁹Zr-^ssdual-5B1 and ^ssFL-5B1 constructs, which showed that both had low nanomolar affinity for CA19.9 (Fig. S2). Ultimately, the site-specific bioconjugation strategy produced reliable and robust immunoconjugates that retain their binding properties and are structurally well defined.

Fig. S2.

Binding curves for ^ssdual-5B1 and ^ssFL-5B1 determined via a cell-based (BxPC3) 96-well plate assay.

Acute Biodistribution of ^ssDual-5B1.

To directly quantify the uptake of the radiolabeled ⁸⁹Zr-^ssdual-5B1, the acute biodistribution of the radiotracers was determined at 48 h and 120 h in athymic, nude mice (n = 4) bearing a single BxPC3 s.c. xenograft on the right flank (Fig. 2A and Table S1). Although the bilateral model showed negligible uptake in the imaging experiments described later, it was not used for the acute biodistribution study. Instead, we chose to use a single xenograft model to allow for a direct comparison with the previous experiments using ⁸⁹Zr-DFO-5B1 that was not site-specifically modified.

Fig. 2.

In vivo distributions of the radiolabeled constructs. (A) A graph of the uptake of ⁸⁹Zr-^ssdual-5B1 at 48 h and 120 h in mice bearing a single BxPC3 (CA19.9-positive) tumor determined by acute biodistribution. Error bars represent the SD (n = 4). Serial PET imaging of ⁸⁹Zr-^ssDFO-5B1 (B) and ⁸⁹Zr-^ssdual-5B1 (C) is also shown. Coronal (B and C, Top) and transverse (B and C, Bottom) slices of a representative mouse from each imaging cohort (n = 3) are shown. The uptake is reported in percentage of the injected dose per gram of tissue (% ID/g).

Table S1.

Values from the acute biodistribution study of ⁸⁹Zr-^ssdual-5B1 and comparison with previously reported values with ⁸⁹Zr-DFO-5B1 that was not site-specifically labeled (5)

	89Zr-ssdual-5B1		89Zr-DFO-5B19:
Tissue	48 h	120 h	120 h
Blood	9.6 ± 2.9	4.8 ± 1.2	9.3 ± 1.2
Tumor	83.5 ± 9.4	102.9 ± 26.0	114.1 ± 23.1
Heart	3.1 ± 0.9	2.1 ± 0.4	4.0 ± 0.9
Lungs	6.2 ± 2.3	3.5 ± 0.7	8.6 ± 3.6
Liver	18.8 ± 4.5	9.7 ± 1.3	6.7 ± 3.7
Spleen	7.0 ± 1.6	5.7 ± 0.5	19.8 ± 7.9
Pancreas	1.2 ± 0.6	1.0 ± 0.5	1.0 ± 0.4
Stomach	1.1 ± 0.2	0.5 ± 0.1	1.5 ± 0.4
Small intestine	1.1 ± 0.3	0.6 ± 0.1	1.8 ± 0.3
Large intestine	0.6 ± 0.1	0.4 ± 0.1	1.2 ± 0.6
Kidneys	6.1 ± 0.5	4.9 ± 0.8	7.0 ± 2.8
Bone	4.6 ± 1.9	4.5 ± 2.2	9.5 ± 1.4
Muscle	1.4 ± 0.6	0.6 ± 0.1	0.8 ± 0.4

The acute biodistribution data indicated that the uptake of ⁸⁹Zr-^ssdual-5B1 in the BxPC3 xenografts at 48 h [83.5 ± 9.4% injected dose (ID)/g] and 120 h (102.9 ± 26.0% ID/g) was exceptional. However, comparison with the previously reported acute biodistribution data with ⁸⁹Zr-DFO-5B1 (Table S1) shows that the overall uptake in the tumors is approximately the same at both time points, begging the question of whether the improvement in immunoreactivity resulting from the site-specific labeling of the ⁸⁹Zr-^ssdual-5B1 is reflected in the biodistribution data. When considering PET or NIRF imaging of PDAC in a clinical setting, it will be most important to achieve contrast between the tumor and four organs in particular: the pancreas, the spleen, the kidneys, and the liver. This is simply due to the pancreas residing in direct contact with the liver and spleen in the human body.

Of the nontarget tissues, the accumulation of ⁸⁹Zr-^ssdual-5B1 was highest in the liver at 48 h (18.8 ± 4.5% ID/g), but that value was significantly reduced at 120 h (9.7 ± 1.3% ID/g) to give a tumor to liver ratio of greater than 10:1. In fact, a comparison of ^ssdual-5B1 to ⁸⁹Zr-DFO-5B1 reveals that ⁸⁹Zr-^ssdual-5B1 displayed lower retention in every nontarget tissue at 120 h with the exception of the liver, which was a modest 3% ID/g higher with ⁸⁹Zr-^ssdual-5B1. Considering the drastic improvement in immunoreactivity, the increased uptake in the liver relative to ⁸⁹Zr-DFO-5B1 is likely the result of appending a fluorescent dye to the antibody, which is known to increase liver uptake in dual-labeled constructs (27).

A better reflection of the effects of site-specific modification is the retention of ^ssdual-5B1 in the spleen at 120 h. Rapid uptake and retention of molecules in the spleen are commonly associated with large particles (i.e., nanoparticles and liposomes), and in the case of mAb radiotracers, such uptake is often the result of aggregation. Compared with ⁸⁹Zr-DFO-5B1, the retention in the spleen was greatly reduced (7.0% ID/g vs. 19.8% ID/g) with ^ssdual-5B1. The immunoreactivity of ⁸⁹Zr-^ssdual-5B1 was ∼20% higher, and comparison of the previously reported data suggests this was likely—at least to some extent—due to aggregation of the nonspecifically modified ⁸⁹Zr-DFO-5B1. The question becomes whether this nearly threefold decrease in spleen retention is offset by the increased uptake of ⁸⁹Zr-^ssdual-5B1 in the liver. Due to the close proximity of the human pancreas to the liver and spleen, it is important that the uptake in any neoplastic tissue is high enough to make it apparent in a PET scan, especially in the context of metastases in those tissues. So, the overall increase in contrast between the tumor and the liver/spleen is perhaps the best benchmark. Overall, the increased contrast of ⁸⁹Zr-^ssdual-5B1 compared with ⁸⁹Zr-DFO-5B1 is entirely apparent in the spleen whereas the decrease in liver contrast is negligible, suggesting ⁸⁹Zr-^ssdual-5B1 will likely offer advantages in a clinical setting.

In Vivo Imaging with ^ssDFO-5B1, ^ssFL-5B1, and ^ssDual-5B1.

After confirming the ability of each construct to bind CA19.9 in vitro, the next step was to assess each of the three constructs, using in vivo murine models of pancreatic cancer. It has been established that convincingly blocking the binding of radiolabeled 5B1 to CA19.9 using an excess of cold 5B1 is very difficult due to the exceptionally high copy number of CA19.9 in BxPC3 xenografts. In a previous study, a statistically significant level of blocking was achieved by coinjecting an excess of unmodified 5B1, yet the uptake of the 5B1 radiotracer at 24 h postinjection (p.i.) remained in excess of 50% ID/g (24). For that reason, we chose to use a bilateral, s.c. xenograft model, using both a CA19.9-positive (BxPC3) and a CA19.9-negative (MIAPaCa-2) xenograft in each mouse for our PET and NIRF imaging studies to more clearly show the specificity of the site-specifically modified constructs. This allowed each mouse to serve as its own control and reduced the number of mice required. To that end, athymic, nude mice (n = 3 per construct) bearing both a MIAPaCa-2 and BxPC3 s.c. xenografts on either the left or the right flank, respectively, were used to evaluate ⁸⁹Zr-^ssDFO-5B1, ^ssFL-5B1, and ⁸⁹Zr-^ssdual-5B1 in vivo.

Both the ^ssDFO-5B1 and ^ssdual-5B1 exhibited exceptional uptake in BxPC3 tumors that increased over time and negligible uptake in MIAPaCa-2 tumors (Fig. 2 B and C and Figs. S3 and S4). Regions of interest (ROI) drawn on the PET images (n = 3 per construct) suggested that the uptake in the BxPC3 xenografts was more than eightfold higher than in the MIAPaCa-2 xenografts at 120 h for both constructs (Fig. S5). Large molecules like antibodies are known to accumulate in xenografts, due to the enhanced permeability and retention (EPR) effect, which is likely the reason for the nominal uptake of the radiotracers in the MIAPaCa-2 tumors.

Fig. S3.

Maximum-intensity projections from serial PET of mice with bilateral BxPC3 (T+, right flank, antigen positive) and MIAPaCa-2 (T−, left flank, antigen negative) that were injected with ⁸⁹Zr-^ssDFO-5B1 (244 ± 9 μCi).

Fig. S4.

Maximum-intensity projections from serial PET of mice with bilateral BxPC3 (T+, right flank, antigen positive) and MIAPaCa-2 (T−, left flank, antigen negative) that were injected with ⁸⁹Zr-^ssDFO-5B1 (92 ± 14 μCi).

Fig. S5.

Graph of tracer uptake determined from regions of interest drawn on PET images.

Although visualizing two-dimensional tumor slices allows for a quantitative “snapshot” of the tracers’ distribution, considering the maximum intensity projections (MIPs) provides a more complete picture of the PET imaging results. In the case of ⁸⁹Zr-^ssDFO-5B1 and ⁸⁹Zr-^ssdual-5B1, the MIPs indicated that the tumor to background contrast was exceptional in all cases for each of the mice (Figs. S3 and S4). With both constructs, analysis of the MIPs showed some uptake in nontarget tissue that varied between mice, but the nonspecific uptake was quite low and was not consistently high in any one tissue other than the CA19.9-positive xenografts. The higher specific activity of the ⁸⁹Zr-^ssDFO-5B1 construct reduced the time required to obtain quality images, but the overall quality of the images that were acquired with ⁸⁹Zr ^ssdual-5B1 was equal and the uptake values from ROI analysis of the images were slightly higher, reinforcing the results of the acute biodistribution study (Fig. S5).

Concurrent with the PET imaging studies, NIRF imaging was performed with ^ssFL-5B1 (Fig. 3A and Fig. S6) and ⁸⁹Zr-^ssdual-5B1 (Fig. 3B and Fig. S7), the latter using the same cohort of mice from the PET studies discussed above. ^ssFL-5B1 (50 μg) was injected into athymic, nude mice (n = 3) with bilateral tumors as described above. The NIRF signal from the tracers in the CA19.9-positive tumors increased over time whereas signal from the nontarget organs and CA19.9-negative tumors was negligible, yielding images of remarkable quality. Similar to the trend seen in the PET images, ⁸⁹Zr-^ssdual-5B1 matched or exceeded its singly modified counterpart, ^ssFL-5B1.

Fig. 3.

(A and B) Serial NIRF imaging using ^ssFL-5B1 (50 μg) (A) and ⁸⁹Zr-^ssdual-5B1 (50 μg, 92 ± 14 μCi) (B). NIRF images of a representative mouse from each cohort (n = 3) bearing CA19.9-negative (T−, left flank) and CA19.9-positive (T+, right flank) tumors that were acquired at 48 h and 120 h are shown. The scale of the fluorescence signal is reported as radiant efficiency [(p/s/cm²/sr)/(μW/ cm²)].

Fig. S6.

White light (Top) and fluorescence (Bottom) from serial NIRF imaging (excitation 750 nm/emission 800 nm) of mice with bilateral BxPC3 (T+, right flank, antigen positive) and MIAPaCa-2 (T−, left flank, antigen negative) injected with ⁸⁹Zr-^ssdual-5B1.

Fig. S7.

White light (Top) and fluorescence (Bottom) from serial NIRF imaging (excitation 750 nm/emission 800 nm) of mice with bilateral BxPC3 (T+, right flank, antigen positive) and MIAPaCa-2 (T−, left flank, antigen negative) injected with ⁸⁹Zr-^ssFL-5B1.

After removal of the tumors, the MIAPaCa-2 tumor and remaining organs were harvested for further imaging and analysis ex vivo. NIRF images of the tumors (Fig. S8A) and organs (Fig. S8B) were acquired to compare the relative fluorescence signal from the whole organs without obstruction by other organs. The average fluorescence signal from BxPC3 tumors was more than 18-fold higher than that of the MiaPaCa-2 tumors and the tumor to organ ratios were all in excess of 25:1 (Fig. S8C). The contrast achieved in both the in vivo and ex vivo NIRF images, particularly with the ⁸⁹Zr-^ssdual-5B1 construct, further confirmed the specificity of the tracers and provided us further impetus to carry the ⁸⁹Zr-^ssdual-5B1 forward to studies in more advanced murine models of PDAC.

Fig. S8.

(A–C) Ex vivo NIRF images of resected tumors (A, all mice injected with ⁸⁹Zr-^ssdual-5B1) and organs (B, from M2 only) (B, bone; H, heart; K, kidney; LI, large intestine; Lu, lung; Lv, liver; M, muscle; P, pancreas; SI, small intestine; St, stomach) along with ROI analysis (C).

In Vivo Imaging ⁸⁹Zr-^ssDual-5B1 in an Orthotopic Model of PDAC.

After establishing the potential of ⁸⁹Zr-^ssdual-5B1 in initial PET and NIRF imaging studies, we assessed it in a mouse model that more accurately recapitulates the tumor environment found in PDAC. We transitioned to an orthotopic pancreatic cancer model in which the lesions are formed directly in the pancreas, using a different human PDAC cell line, Suit-2. Suit-2 tumors are known to shed CA19.9 into the blood, which occurs more often than not in PDAC and is the basis of the CA19.9 blood test that is the current standard for evaluating PDAC patients in a clinical setting (31, 32). Furthermore, Suit-2 tumors that are inoculated in the pancreas are known to metastasize and thus provide a better platform for evaluating the potential for clinical translation of these radiotracers (31).

Our goal was to study the ⁸⁹Zr-^ssdual-5B1 in a model that provides the best recapitulation of CA19.9-positive PDAC in a clinical setting that is currently possible with a murine model. Each of the orthotopically implanted mice developed primary pancreatic tumors and one mouse, which was selected for further PET-computed tomography (PET/CT) imaging, had additionally developed multiple metastases in the abdominal cavity that could be clearly delineated in the PET images (Fig. 4A). PET/CT imaging provided an accurate anatomic localization of the PET signal, suggesting at least two large metastases in the abdominal cavity (Fig. 4B). Additionally, the images clearly delineated the sentinel lymph nodes, suggesting potential metastasis or lymphatic drainage from the tumors due to shedding CA19.9. The involvement of the lymphatic system in metastases has been well documented, and the ability to map the sentinel lymph nodes noninvasively—as well as in the context of biopsy or surgical resection—could prove beneficial in a clinical setting. Stage 2 PDAC is characterized by spread to the lymph nodes, and ^ssdual-5B1 could reliably guide the biopsy of the appropriate sentinel lymph nodes. Although ^ssdual-5B1 cannot directly determine whether malignancy has spread to the sentinel lymph nodes, the information provided by an image-guided biopsy could prove crucial to the staging and, consequently, treatment of PDAC.

Fig. 4.

PET, PET/CT, and NIRF imaging of ⁸⁹Zr-^ssdual-5B1 (50μg, 84 ± 6 μCi) in an orthotopic PDAC model. (A) Slices (Top, coronal; Bottom, transverse) of a representative mouse at 120 h p.i. show high uptake in the primary tumor, a metastasis, and sentinel lymph nodes (LN, lymph node; M, metastasis; T, tumor). (B) A rendering of the PET/CT data allows for the anatomization of the lesions. (C and D) NIRF imaging of the sentinel lymph nodes (C) and the open thoracic cavity (D) after removal of the LNs demonstrates the potential of ⁸⁹Zr-^ssdual-5B1 in an intraoperative setting.

After PET/CT imaging at 120 h, the mouse was killed and NIRF images of the intact mouse were acquired. The initial images showed the primary tumor, sentinel lymph nodes, and the primary metastases that had been identified in the PET/CT images. The NIRF signal was used to aid in the localization and removal of several of the sentinel lymph nodes (Fig. 4C). An image with an expanded view of the open abdominal cavity after removal of the sentinel lymph nodes shows the remaining lymph nodes as well as the primary tumor (Fig. 4D). Interestingly, NIRF signal was also apparent in numerous micrometastases that were not delineated in the PET/CT scans and were in most cases not obvious to the naked eye. After NIRF imaging was completed, a portion of a large liver metastasis and the pancreas, including the primary tumor, were collected for ex vivo analysis. Due to the large number of micrometastases and partial infiltration of the primary tumor into the spleen, it was not possible to remove all of the tumor tissue. Nonetheless, this study did illustrate the remarkable potential of ⁸⁹Zr-^ssdual-5B1 to serve as a guide for the staging, treatment planning, and resection of PDAC. In particular, the delineation of micrometastases that were not apparent to the naked eye or in the PET imaging could prove beneficial. The diagnosis of metastatic disease precludes patients with PDAC from resection. However, it is not uncommon for metastases to go undetected, resulting in futile resection procedures that greatly decrease the quality of life of a patient that has no hope of being cured. The application of ⁸⁹Zr-^ssdual-5B1 in resection candidates could prevent such needless resections by aiding in the detection of difficult to identify metastases.

Ex Vivo Evaluation of ^ssDFO-5B1, ^ssFL-5B1, and ^ssDual-5B1.

To demarcate the areas of specific uptake and demonstrate preferential localization of the ⁸⁹Zr-^ssdual-5B1, tissues of interest were harvested for ex vivo analysis directly after the completion of the imaging studies. A representative BxPC3 tumor from a mouse injected with ⁸⁹Zr-^ssdual-5B1 is shown in Fig. 5A. The histologic staining and the autoradiography confirmed that the uptake of ⁸⁹Zr-^ssdual-5B1 was focused in areas of the highest CA19.9 expression. Autoradiography of MIAPaCa-2 tumor slices showed no detectable signal, suggesting the small amount of ⁸⁹Zr-^ssdual-5B1 seen in the PET and NIRF imaging was due to nonspecific accumulation (Fig. S9A).

Fig. 5.

Histology (A–C, Left), autoradiography (A–C, Center), and fluorescence microscopy (A–C, Right) of resected tumor tissue from mice injected with ⁸⁹Zr-^ssdual-5B1. (A) A BxPC3 xenograft, (B) metastatic foci from a mouse with a Suit-2 orthotopic xenograft, and (C) the primary tumor with surrounding healthy pancreas tissue from the same mouse are shown, confirming colocalization of the tracer with CA19.9 expression.

Fig. S9.

(A) Hematoxylin and eosin (Left) and autoradiography (Right) of the MIAPaCa-2 s.c. xenograft from the same mouse that is shown in Fig. 2B and Fig. S3 (mouse 2), indicating there was no specific uptake in the MIAPaCa-2 tumor. (B) Hematoxylin and eosin (Left), autoradiography (Center), and immunohistochemistry (Right) of the right brachial lymph node that was collected from the mouse shown in in Fig. 4, indicating the uptake was not specific.

Tumor tissues were also harvested from the mouse with an orthotopic Suit-2 xenograft that underwent PET/CT and postmortem NIRF imaging. A portion of the metastases located in the liver (Fig. 5B) was harvested as was the primary pancreatic tumor with the surrounding pancreas tissue (Fig. 5C). The distribution of ⁸⁹Zr-^ssdual-5B1 in the small metastasis matched the expression of the CA19.9, and the same was true of the primary tumor in the pancreas, which was easily distinguished from the healthy pancreas (Fig. 5C, Inset). Examination of the right brachial lymph node, which showed a strong signal by PET, showed that the accumulation was nonspecific (Fig. S9B), and histological evaluation confirmed no tumor cells were present.

Conclusions

Currently, there are no clinically available, PDAC-specific molecular imaging tools, and FDG is not well suited for imaging PDAC. We have generated an array of three modular tools to do exactly that by targeting the most common clinical biomarker in PDAC, CA19.9. The three constructs that we evaluated—⁸⁹Zr-^ssDFO-5B1, ^ssFL-5B1, and ⁸⁹Zr-^ssdual-5B1—displayed excellent uptake in CA19.9-positive xenograft models of PDAC, suggesting that each of them has the potential to improve outcomes in patients with PDAC. One of these constructs, ⁸⁹Zr-^ssdual-5B1, combines the strengths of PET and optical imaging into a single agent, using a robust, reproducible, and modular methodology that does not compromise the binding to CA19.9. Evaluation of ⁸⁹Zr-^ssdual-5B1 in an orthotopic PDAC model demonstrated that the dual-modal construct has multiple applications, including PET imaging to stage PDAC and identify small metastases, visualization of malignant tissue during tumor resection, and localization of sentinel lymph nodes using PET and/or optical modalities. Although these results represent a major step toward developing a clinically useful toolkit for the management of PDAC, it is worth noting that mice do not naturally express CA19.9, making it impossible to determine how the background expression of CA19.9 will influence imaging contrast. Inflammation associated with pancreatitis is known to increase CA19.9 expression and could also reduce image contrast. We are collaboratively working to develop a mouse model in which CA19.9 is constitutively expressed to provide a platform to study these effects and further optimize our CA19.9 probes.

Materials and Methods

Construct Preparation and Characterization.

The 5B1 was modified to incorporate four azido groups (^ss5B1-N₃) by a previously reported method (25, 26) and then was incubated with DIBO-DFO, DIBO-FL, or a 1:1 mixture of the two constructs to create the completed immunoconjugates: ^ssDFO-5B1, ^ssFL-5B1, and ^ssDual-5B1. An SDS/PAGE assay was performed, as previously described (25, 26), to demonstrate the specificity of the conjugation. For biodistribution and imaging studies, the ^ssDFO-5B1 and ^ssDual-5B1 were radiolabeled following published procedures to generate ⁸⁹Zr-^ssDFO-5B1 and ⁸⁹Zr-^ssdual-5B1 (13, 28, 29). Radiolabeled constructs were analyzed by radio-TLC to assess the radiochemical purity and a cell-based binding assay to assess the immunoreactivity, using previously described methods. The DOL for ^ssFL-5B1 was determined via UV-Vis spectrophotometry at 280 nm and 774 nm per the dye manufacturer’s instruction.

Murine Models.

All experiments involving laboratory animals were performed in accordance with the Memorial Sloan Kettering Institutional Animal Care and Use Committee (protocol 08–07-013). BxPC3 and MIAPaCa-2 xenografts were grown 18–21 d postimplantation. Orthotopic pancreas tumors were induced via intrapancreatic injection of Suit-2 cells into the body of the pancreas and were allowed to develop for 14 d before PET, PET/CT, and NIRF imaging.

Acute Biodistribution.

The acute biodistribution of ⁸⁹Zr-^ssdual-5B1 was determined in mice with BxPC3 s.c. xenografts. Mice (n = 4) were injected with ⁸⁹Zr-^ssdual-5B1 via the lateral tail vein and euthanized at 48 h and 120 h p.i. before collection of 13 tissues, including the tumor. The mass of each organ was determined and then each sample was counted using an automatic gamma counter. Counts were converted into activity (% ID/g)—after decay and background correction—by normalization to the total activity injected into the respective animal.

In Vivo Imaging.

For imaging, mice (n = 3) were injected via the lateral tail vein with an equal mass (50 μg) of the appropriate imaging agent. At 24 h, 48 h, 72 h, and 120 h p.i., PET and/or NIRF images of mice were acquired. Mice with orthotopic Suit-2 xenografts were injected with ⁸⁹Zr-^ssdual-5B1 and images were acquired at 48 h and 120 h on a small animal PET/CT scanner.

Ex Vivo Analysis.

Tissues were resected, embedded, and frozen in optimal cutting temperature compound, and sequential 10-μm sections were cut for analysis. Autoradiography was performed in a film cassette on a phosphor imaging plate. Immunohistochemistry (IHC) was performed with unmodified 5B1 that was visualized with a fluorescently labeled goat anti-human secondary antibody. A sequential section was submitted to the Memorial Sloan Kettering Cancer Center (MSKCC) Molecular Cytology Core Facility for automated hematoxylin and eosin staining.

SI Materials and Methods

Reagents and General Procedures.

Unless otherwise noted, all chemicals were purchased from Sigma-Aldrich and used without any additional purification. Ultrapure water (>18.2 MΩ⋅cm⁻¹) and dry, molecular biology grade dimethyl sulfoxide and N,N-dimethylformamide were used. All activity measurements were performed using a Capintec CRC-15R dose calibrator. Instant TLC (ITLC) for radioITLC experiments was performed on strips of glass-fiber, silica-impregnated paper (PallCorp.), read on a Bioscan AR-2000 radioTLC plate reader, and analyzed using Winscan Radio-TLC software (Bioscan). DBCO-FL was purchased from Li-COR. MabVax Therapeutics generously provided the 5B1 mAb. The GalT(Y289L) enzyme, UDP-GalNAz, and DIBO-DFO were purchased from Thermo Fisher Scientific (25). All instrumentation was used, maintained, and calibrated according to the manufacturer’s recommended procedures. All experiments involving laboratory animals were performed in accordance with the Memorial Sloan Kettering Institutional Animal Care and Use Committee (protocol 08–07-013).

Site-Specific Modification of 5B1.

Glycans modification.

The 5B1 (2.5 mg, 5 mg/mL) underwent a buffer exchange into pretreatment buffer (50 mM Na-phosphate, pH 6.0), using a microspin column prepared with P30 resin (Bio-Rad 732-6008, 1.5 mL bed volume). The column was first equilibrated in 50 mM Na-phosphate, pH 6.0, and then spun for 3 min at 850 × g. A total of 500 µL 5B1 antibody was added and then spun down for 5 min at 850 × g. The resultant antibody solution was supplemented with 40 µL of β-1.4-galactosidase [from Streptococcus pneumoniae (2 mU/µL), obtained from Prozyme] and placed in an incubator at 37 °C for 6 h.

GalNAz labeling.

After the β-1.4-galactosidase treatment, 75 μL H₂O, 12.5 μL 1 M Tris buffer (pH 7.6), 25 μL GalT(Y289L) (from a stock of 2 mg/mL in 50 mM Tris, 5 mM EDTA, pH 8), 2.5 μL 1 M MnCl₂ (in 0.1 M HCl), and 10 μL UDP-GalNAz (from a stock of 40 mM in H₂O) were added to the reaction solution to bring the final volume up to 250 μL. This reaction solution contained concentrations of 10 mg/mL 5B1, 10 mM MnCl₂, 1.6 mM UDP-GalNAz, and 0.2 mg/mL GalT (Y289L) and was incubated for 16 h at 30 °C.

DIBO-DFO and DBCO-FL ligation.

The solution from the GalNAz labeling step was purified via centrifugal filtration, using 2 mL Amicon Ultra centrifugal filters with a 50,000-Da molecular mass cutoff (Millipore) and TBS buffer, pH 7.4. After centrifugation, the modified 5B1 antibody (1 mg in 800 μL TBS buffer, pH 7.4) was combined with a mixture of 100 μL DIBO-DFO (from a 2-mM stock in DMSO) and 100 μL DBCO-FL (from a 2-mM stock in DMSO) to yield a final solution containing 1 mg/mL 5B1, 0.2 mM DIBO-DFO, and 0.2 mM DBCO-FL. This solution was then incubated for 16 h at 25 °C.

Purification.

After the click labeling, the completed antibody was purified via size exclusion chromatography (PD10 column; GE Healthcare) and concentrated via centrifugal filtration, using 2-mL Amicon Ultra centrifugal filters with a 50,000-Da molecular mass cutoff (Millipore).

Determination of Degree of Fluorescent Labeling.

To determine the degree of labeling (DOL) of the fluorophore-labeled antibodies, UV-Vis absorbance measurements were taken at 280 nm and 774 nm for three separate antibody concentrations. The degree of labeling was calculated using the following formulas:

$${\text{A}}_{\mathrm{mAb}}={\text{A}}_{280}\mathrm{-}{\text{A}}_{\max }\left(\mathrm{CF}\right)$$

$$\mathrm{DOL}\mathrm{=}\left[{\text{A}}_{\max }\mathrm{\ast }{\mathrm{MW}}_{\mathrm{mAb}}\right]/\left[{\left[\mathrm{mAb}\right]}^{\ast }{\mathrm{\epsilon }}_{\mathrm{Dye}800}\right].$$

The correction factor (CF) for Dye800 was given as 0.03 by the supplier, MW_5B1 = 150,000, ε_774,Dye800 = 240,000, and ε_280,5B1 = 225,000.

SDS/PAGE Analysis of the Antibody Conjugation.

Denaturing SDS/PAGE experiments were used to analyze the antibody conjugation methodology. To this end, 1 μg antibody (2 μL of a 0.5-mg/mL stock) was combined with 7.5 μL 4× electrophoresis buffer (NuPAGE LDS Sample Buffer; Life Technologies), 3 μL 500 mM DTT (NuPAGE 10× Sample Reducing Agent; Life Technologies), and 18.5 μL H₂O. This solution was then denatured by heating to 90 °C for 10 min on a heat block. After denaturing, 25 μL of each sample was then loaded alongside an appropriate molecular weight marker (Mark12 unstained standard; Life Technologies) onto a 1-mm, 10-well 4–12% Bis-Tris protein gel (Life Technologies) and run in Mops buffer for ∼2 h at 10 V/cm. The finished gel was then washed three times with H₂O, stained for 1 h using SimplyBlue SafeStain (Life Technologies), and destained overnight in H₂O. After destaining, the gel was then analyzed using an Odyssey Infrared Gel Scanner (Li-Cor Biosciences).

PNGase F Treatment of Antibody.

To further study the modification of the antibodies, the cleavage of the heavy chain glycans by the enzyme PNGaseF was studied. To this end, 5B1 antibody construct (1 μg) in 10 µL TBS was denatured with 0.5% SDS and 40 mM DTT by adding 17 µL H₂O and 3 µL 10× Glycoprotein Denaturation Buffer (New England Biolabs) and incubated at 90 °C for 10 min on a heat block. Subsequently, 18 µL H₂O, 6 µL 10% Nonidet P-40, and 6 µL 500 mM sodium phosphate, pH 7.5 (G7 reaction buffer from New England Biolabs), were added. This solution was then split into two aliquots, and one aliquot was supplemented with 1 µL PNGaseF (New England Biolabs) and incubated overnight at 37 °C. After incubation, the PNGaseF-treated and untreated samples were analyzed by SDS/PAGE as described above.

Radiolabeling of 5B1 Constructs with ⁸⁹Zr.

Preparation of ⁸⁹Zr-labeled DFO-5B1 and dual-5B1 was achieved in accordance with previously described methods (13, 28, 29). The ⁸⁹Zr-oxalate in oxalic acid (1 M) was neutralized to pH 7.0–7.2, using Na₂CO₃ (1 M) followed by addition of the appropriate construct in PBS (pH 7.4). The mixture was incubated at room temperature for 1–2 h and monitored using radio-ITLC, eluting with an aqueous solution of EDTA (50 mM, pH 5.5). Upon satisfactory radiolabeling, the reaction was quenched by addition of the same EDTA solution and the labeled construct was purified using gel-filtration chromatography (Sephadex G-25, PD10 desalting column; GE Healthcare) into 0.9% saline. Radiochemical purity was assessed by ITLC in an aqueous solution of EDTA (50 mM, pH5.5). The ⁸⁹Zr is produced at MSKCC via the ⁸⁹Y(p,n)⁸⁹Zr transmutation reaction on a TR19/9 variable-beam energy cyclotron (Ebco Industries) (28).

Cell Culture.

BxPC3 cells were grown in RPMI medium modified to contain 4.5 g/L glucose, 1.5 g/L sodium bicarbonate and supplemented with 10% (vol/vol) heat-inactivated FCS, 100 IU penicillin, 100 μg/mL streptomycin, 10 mM Hepes, and 10 cc/L nonessential amino acids. MIAPaCa-2 cells were grown in Dulbecco’s modified essential medium (DMEM) modified to contain 4.5 g/L glucose, 1.5 g/L sodium bicarbonate and supplemented with 10% (vol/vol) heat-inactivated FCS, 100 IU penicillin, and 100 μg/mL streptomycin. Suit-2 cells were grown in RPMI medium supplemented with 10% (vol/vol) heat-inactivated FCS, 100 IU penicillin, and 100 μg/mL streptomycin. All media were purchased from the Media Preparation Facility at Memorial Sloan Kettering Cancer Center.

Immunoreactivity Measurements.

The immunoreactivity of the ⁸⁹Zr-labeled constructs was determined using CA19.9-positive (BxPC3) and CA19.9-negative (MIAPaCa-2) human pancreatic cancer cells via a previously reported method. Briefly, 50 μL of a 2-μCi/mL stock of the radiolabeled construct (PBS, pH 7.4, with 1% BSA) was added to suspensions of the appropriate cells at a range of concentrations (5 × 10⁵–5 × 10⁶ cells/mL, 500 μL in PBS, pH 7.4) and incubated with gentle shaking at room temperature for 1 h. The cells were then pelleted (600 × g for 2 min) and washed three times with PBS, and the activity of the cell pellet was counted on a gamma counter to assess the amount of ⁸⁹Zr-labeled mAb bound to the cells. Linear regression analysis of the background-corrected data was performed by plotting the ratio of the total to bound (total/bound) activity against the inverse of the normalized cell concentration (1/normalized cell concentration).

A second binding assay was performed to determine the specificity of binding. Briefly, 1 × 10⁷ cells were suspended in 100 μL of PBS, pH 7.4, and an aliquot of ⁸⁹Zr-labeled construct (20 μL of a 2-ng/mL stock in PBS, pH 7.4, with 1% BSA) was added. The mixture was incubated at 4 °C for 1 h with gentle mixing before the cells were pelleted, the supernatant was removed, and the cells were washed in ice-cold PBS (three times). The activity bound to the cells was compared with the combined radioactivity collected from the supernatant and washes to calculate the total percentage bound. All immunoreactivity and binding experiments were performed in triplicate and the results are reported as an average.

Cell Binding Assay.

For cell binding assays, BxPC3 were seeded in 96-well plates (30,000 cells per well) and cultured overnight to ∼90% confluence. The imaging agents—^ssdual-5B1 or ^ssFL-5B1—were diluted in media and added to wells immediately after changing media in the wells at concentrations ranging from 0.98 nM to 1,000 nM (n = 4 wells per concentration per construct). The plates were incubated with gentle shaking for 1 h at room temperature. The antibody solution was then removed and the wells were washed with PBS three times. The plate was immediately scanned on an Odyssey Infrared Imaging System (LI-COR Biosciences) to measure the fluorescence signal intensity. The data were processed and analyzed with GraphPad Prism v6.0 (GraphPad Software) to determine the K_d and B_max values.

Murine Xenograft Models of PDAC.

s.c. xenograft models.

BxPC3 and MiaPaCa-2 female athymic homozygous nude mice, strain Crl:NU(NCr)-Foxn1^nu (Charles River Laboratories), age between 6 wk and 8 wk, were xenografted s.c. with 5 × 10⁶ cells, suspended in 150 μL of a solution containing a 1:1 mixture of Matrigel (Becton Dickinson) and cell culture medium. BxPC3 tumors were grown for 21–28 d postimplantation and MiaPaCa2 xenografts were grown 18–21 d postimplantation before imaging.

Orthotopic xenograft model.

For orthotopic pancreas xenografts, athymic homozygous nude mice, strain Crl:NU(NCr)-Foxn1^nu (Charles River Laboratories), age 6–8 wk were used. Mice were anesthetized with 1–2% isoflurane gas in medical air at a rate of 2 L/min and surgery was performed on a heated platform to help maintain body temperature. Bupivicaine, a local anesthetic agent, was injected intradermally in the area surrounding the incision line. Skin was prepped for surgery, using alternating scrubs of povidone-iodine and 70% ethanol. A longitudinal incision (0.5–1 cm in length) was made in the skin and the peritoneum, allowing for the spleen and pancreas to be exteriorized. Suit-2 cells (7.5 × 10⁵ cells in 30 μL containing 1:1 cell media and Matrigel) were slowly injected into the parenchyma of the pancreas. The spleen and pancreas were returned to the peritoneal cavity, the peritoneal wall was closed using 4-0 Vicryl sutures, and the skin was closed using sterile wound clips. Buprenorphine was administered before recovery and the dosage was repeated postsurgery as needed every 4–6 h. Tumors reached optimal size within 14–21 d postimplantation.

In Vivo Imaging.

Female, athymic nude mice with bilateral MIAPaCa-2 (CA19.9-negative) and BxPC3 (CA19.9-positive) tumors were used for the imaging experiments. In all cases, between 0.3 nmol and 0.4 nmol (50–60 μg) of the appropriate 5B1 construct was prepared in 200 μL of sterile 0.9% saline solution for injection. For all imaging procedures, animals were anesthetized with 2% isoflurane/oxygen mixture beginning 5 min before imaging and remained under anesthesia throughout the imaging procedure. For all constructs, PET and/or NIRF imaging was performed at 24 h, 48 h, 72 h, and 120 h postinjection. Animal experiments were carried out in accordance with an MSKCC Institutional Animal Care and Use Committee approved protocol (08-07-013).

Near-infrared fluorescence imaging.

All mice that underwent NIRF imaging were placed on a low autofluorescence diet (AIN-76; Harlan Laboratories) at least 5 d before the first imaging time point. Near-infrared fluorescent images were acquired with an IVIS Spectrum Preclinical Imaging System (Perkin-Elmer), using excitation/emission settings of 745 nm/800 nm. All live-animal images were acquired with a 500-ms exposure time. Images were processed using Living Image (v4.4; Perkin-Elmer) and values are reported as radiant efficiency to allow direct comparison of images from each experiment. All region-of-interest measurements were generated via automated analysis to avoid user bias.

PET imaging.

For experiments with the bilateral s.c. xenograft model, mice (n = 3) were administered either ⁸⁹Zr-^ssdual-5B1 [3.0–4.0 megabecquerels (MBq) (80–110 μCi)] or ⁸⁹Zr-^ssDFO-5B1 [9.2–10.0 MBq (250–270 μCi)] via tail vein injection, ensuring the same amount of mass was injected into all mice. Static scans were recorded at the various time points with a minimum of 12 million coincident events (8–25 min total scan time). Images were recorded on a microPET Focus scanner (Concorde Microsystems). An energy window of 350–700 keV and a coincidence timing window of 6 ns were used. Data were sorted into two-dimensional histograms by Fourier rebinning, and transverse images were reconstructed by filtered back projection (FBP) into a 128 × 128 × 63 (0.72 × 0.72 × 1.3 mm) matrix. The image data were normalized to correct for nonuniformity of response of the PET, dead-time count losses, positron branching ratio, and physical decay to the time of injection but no attenuation, scatter, or partial-volume averaging correction was applied. Activity concentrations [percentage of dose per gram of tissue (% ID/g)] were determined by conversion of the counting rates from the reconstructed (filtered back-projection) images. Maximum intensity projection (MIP) images were generated from three-dimensional ordered subset expectation-maximization reconstruction. All of the resulting images were analyzed using ASIPro VM software.

PET/CT.

Mice (n = 3) were administered ⁸⁹Zr-^ssdual-5B1 [2.8–3.3 MBq (77–89 μCi)] via tail vein injection and images were acquired at 48 h and 120 h. Static scans were recorded on an Inveon PET/CT scanner (Siemens Healthcare Global) at the various time points with a minimum of 30 million coincident events (10–30 min total scan time). Data were sorted into two-dimensional histograms by Fourier rebinning, and the images were reconstructed using a 2D ordered subset expectation-maximization (2DOSEM) algorithm (16 subsets, four iterations) into a 128 × 128 × 159 (0.78 × 0.78 × 0.80 mm) matrix. The image data were normalized to correct for nonuniformity of response of the PET, dead-time count losses, positron branching ratio, and physical decay to the time of injection but no attenuation, scatter, or partial-volume averaging correction was applied. Activity concentrations [percentage of dose per gram of tissue (% ID/g)] and MIPs were determined by conversion of the counting rates from the reconstructed images. Whole-body CT scans were acquired with a voltage of 80 kV and 500 μA. A total of 120 rotational steps for a total of 220° were acquired with a total scan time of 120 s and 145 ms per frame exposure. Combined PET/CT images were processed using Inveon Research Workplace software and optimized to show localization of the PET signal in the lymph nodes, primary tumor, and metastases.

Acute Biodistribution.

The acute biodistribution of ⁸⁹Zr-^ssdual-5B1 was determined using a BxPC3 s.c. xenograft (right flank, ∼100 mm³) model in athymic, nude mice. Mice (n = 4) were randomized and warmed gently using a heat lamp before being injected with ⁸⁹Zr-^ssdual-5B1 [7.4–8.1 MBq (20–22 μCi) in 200 μL of 0.9% sterile saline] via the lateral tail vein. Care was taken to ensure the same amount of mass that was injected as was in the imaging experiments by addition of cold 5B1 to the stock solution from which the doses were drawn (total mass = 50–60 μg per mouse). At 48 h and 120 h postinjection, mice (n = 4) were euthanized by asphyxiation, using CO₂ (g). After euthanization, 13 tissues including the tumor were collected, dried in open air for 5 min, and placed into preweighed tubes. The mass of each organ was determined and then each sample was counted using a Wizard² automatic gamma counter that was calibrated for ⁸⁹Zr. A calibration curve that was generated from standards of known activity was used to convert counts into activity. The counts from each sample were decay and background corrected from the time of injection and the activity in each sample was converted to % ID/g by normalization to the total activity injected into the respective animal.

Ex Vivo Analysis.

Autoradiography.

After euthanasia, organs (tumor, muscle, and pancreas) were excised, embedded in Tissue–Plus OCT compound (Scigen), and frozen on dry ice. Tissues were cut in a series of 10-μm sections. Autoradiography was performed to determine radiotracer distribution in tissue sections by placing the sections in a film cassette beneath a phosphor imaging plate for 48–72 h at −20 °C (BASMS-2325; Fujifilm). Phosphor imaging plates were then read on a Typhoon 7000IP plate reader (GE Healthcare) at a pixel resolution of 25 μm.

Immunohistochemistry.

Tissue sections were washed in PBS and fixed with 4% paraformaldehyde for 12 min. Sections were then washed with PBS and incubated in blocking buffer composed of 1% BSA (Sigma Aldrich), 5% goat serum (Sigma Aldrich), and 0.1% Triton-X100 (Sigma Aldrich) in PBS for 30 min. Blocking buffer was removed and slides were incubated in 5B1 antibody (10 μg/mL in blocking buffer) for 1 h, washed with PBS, and then incubated in Alexa Fluor488 goat anti-human IgG (Thermo Fisher Scientific) as a secondary antibody (20 μg/mL in blocking buffer) for 45 min. Slides were washed three times in PBS and one time in H₂O and allowed to air dry.

Hematoxylin and eosin staining.

After autoradiography was performed on a number of sections, a sequential section was submitted to the Molecular Cytology Core Facility and Memorial Sloan Kettering Cancer Center for automated hematoxylin and eosin staining.

Microscopy.

All microscopic images were acquired using a Zeiss Axioplan2 fluorescence microscope connected to a CCD camera and equipped with a motorized stage (Prior Scientific Instruments). MetaMorph software (Molecular Devices) was used to control the microscope and to render the captured image frames into a montage of the entire tumor section. Postacquisition processing was performed using Photoshop CS6 software (Adobe Systems) and image rebinning and pixel-by-pixel correlations were performed as previously described.

Statistical Analysis.

An unpaired, two-tailed Student’s t test was used to analyze the data. In all cases, a 95% confidence level (P < 0.05) was considered to represent a statistical difference in the data.