Prolactin Receptor–Mediated Internalization of Imaging Agents Detects Epithelial Ovarian Cancer with Enhanced Sensitivity and Specificity

Advanced ovarian cancer with abdominal spread (stage III and IV) has 5-year survival rates of <30%, whereas cancers confined to the ovary and the pelvis (stage I and II) have 5-year survival rates of >70% (1). Because the ovaries and the fallopian tube are hidden in the peritoneal cavity, 75% of ovarian cancers remain undiscovered until stage III or IV, after the tumor has metastasized (2). Thus, specific tumor detection poses one of the most important challenges to ovarian cancer research and treatment. Two major goals for current detection methods include (i) detection of tumors when they are still small and curable and (ii) differentiation between benign and malignant ovarian tumors, which would drastically reduce the significant number of unnecessary surgeries (3). Bimanual pelvic examination, transvaginal ultrasound, and serum CA-125 levels have consistently failed to detect early ovarian malignancy (4), and advances in the serum biomarker field have been elusive (5, 6). New imaging modalities that have high specificity and sensitivity are therefore urgently needed.

MRI offers several advantages in the imaging of the pelvis, providing high-resolution images of anatomic structures with excellent soft tissue contrast without the use of ionizing radiation (7, 8). Upon intravenous injection of paramagnetic contrast agents (e.g., gadolinium-chelates; ref. 9), solid tumors become contrast-enhanced relative to surrounding tissue due to their increased permeability caused by altered vascular anatomy (10). However, the diagnosis of malignant ovarian cancer using gadolinium enhanced MRI still lacks sufficient specificity (11) to distinguish healthy from cancerous tissue, resulting in many unnecessary surgeries for a presumptive malignancy (3) and lacks the sensitivity (12) to detect small, early tumors. One potential strategy to improve sensitivity and specificity involves the development of targeted molecular probes—imaging agents conjugated to ligands that bind specifically to receptors that are overexpressed on cancer cells (13). However, due to lack of viable molecular targets, few targeted imaging probes have emerged for ovarian cancer. Here, we introduce the prolactin receptor (PRLR) as a vehicle for internalization of imaging agents into ovarian cancer cells, engendering a new strategy for targeted molecular imaging of ovarian cancer.

As a member of the cytokine receptor superfamily, PRLR activates kinase-mediated signaling networks as a dimer in complex with the pleiotropic protein hormones prolactin, growth hormone, or placental lactogen (14–17). Concomitantly, the heterotrimeric complex formed upon exogenous ligand binding to PRLR induces clathrin-mediated endocytosis (internalization) of the ligand:PRLR complex (18). On the basis of our own IHC analysis showing that 98% of all ovarian cancers express the human PRLR, we hypothesized that imaging agents conjugated to a human PRLR ligand will internalize selectively into ovarian cancer cells via the receptor's natural endocytic mechanism. Herein, we show that imaging conjugates of human placental lactogen (hPL), a specific and high-affinity PRLR ligand, internalize efficiently into PRLR positive (PRLR⁺) cancer cells in ovarian cancer mouse models and thereby enable detection of ovarian cancer with substantially greater specificity and sensitivity than currently used clinical contrast agents. Molecular PRLR imaging holds the potential to enable a more precise and earlier diagnosis of ovarian cancer and to reduce the number of unnecessary surgeries.

Tissue microarrays were assembled from 28 patients with Fédération Internationale de Gynécologie et d'Obstétrique (FIGO) early-stage I/II ovarian cancer and 124 patients with advanced stage III/IV ovarian cancer who had undergone surgery by a gynecologic oncologist at the University of Chicago after obtaining Institutional Review Board (IRB) approval (19). The slides were IHC stained as previously described (19) using 1:25 dilution of hPRLR antibody (sc-20992, Santa Cruz Biotechnology). Prolactin receptor (PRLR) expression was grouped into levels low, medium, or high. For tissues harvested from mice, IHC studies on formalin-fixed, deparaffinized sections (5 μmol/L) were performed using hPRLR antibody at 1:250 dilution (19). For frozen section analysis, human and mouse tissues were placed for confocal analysis in OCT fluid and immediately frozen at −80°C.

CaOV3 cells were purchased from ATCC in 2004. HeyA8 and SKOV3ip1 cells were provided by Dr. Gordon Mills (M.D. Anderson Cancer Center, Houston, TX) in 1995. T47D cells were a gift from Charles Clevenger (Northwestern University, Evanston, IL) and acquired between 2003 and 2006. Cell lines were authenticated using the commercial service, CellCheck (IDEXX Bioresearch). The alleles for nine short tandem repeat (STR) markers were determined and the results were compared to the profiles from DSMZ, ATCC, JCRB, and RIKEN STR databases. Samples were confirmed to be of human origin and no mammalian interspecies contamination was detected. Samples are tested approximately once per year.

Western blot analysis was performed as described (28). The following antibodies were used for our studies: phospho-ERK1/2 and ERK1/2 (Cell Signaling), actin (Sigma), PRLR (H-300, Santa Cruz Biotechnologies), and hPL (Thermo Fisher Scientific).

We grew cells to ∼70% confluency on 12 mm glass coverslips placed in 60 mm plates over 3 days and then starved with serum-free media overnight. The cells were then incubated with 500 nmol/L imaging agent for 6 hours at 37°C in serum-free media at indicated zinc concentration, washed thrice with PBS, fixed in 4% paraformaldehyde in PBS, permeabilized with 0.1% Triton X-100 in PBS, and mounted in Prolong Gold with DAPI (Invitrogen). For immunofluorescence studies, cells were permeabilized with 0.1% Triton X-100 in PBS, followed by incubation with 1:250 dilution of anti-hPRLR antibody (sc-20992) overnight, and briefly with a 1:1,000 dilution of Alexa Fluor 488-labeled mouse anti-rabbit antibody (Invitrogen) before mounting.

Frozen sections of human tissue specimen for hPL*-Cy5 binding were probed with 500 nmol/L hPL*-Cy5 in the presence of 20 μm zinc overnight, washed thoroughly with Dulbecco's phophate-buffered saline (DPBS), and mounted using the immnunofluorescence protocol. Data were analyzed with an Olympus FV1000 (Olympus America). After processing of frozen xenograft mouse tumors by the HTRC, we used mounted tissue protocol described above for confocal analysis. These samples were analyzed using a DSU Spinning Disk Confocal (Olympus America). All data were processed using ImageJ software (v1.42q, NIH).

For subcutaneous tumor imaging of ovarian cancer, we injected female, athymic nude mice (5–6 weeks old) near the thigh with 1 × 10⁶ normal or transfected HeyA8 cells, 2 × 10⁶ SKOV3ip1 cells, or 5 × 10⁶ CaOV3 cells with growth-factor reduced matrigel (BD Biosciences) at a 2:1 ratio. The xenograft model for metastases was previously described (19, 20). Briefly, 1 × 10⁶ GFP-Luc cells were intraperitoneally injected into 5- to 6-week-old, female, athymic nude mice. The Institutional Animal Care and Use Committee (IACUC) at the University of Chicago approved all protocols presented in these studies.

Mice (22.9 ± 0.7 g) with SQ HeyA8 tumors (normal and GFP-Luc) that weighed the indicated size were imaged. For specificity experiments, shScram (l. leg) and shPRLR (r. leg) HeyA8 cells were subcutaneously implanted on contralateral legs and tumors were allowed to grow for 7 to 10 days. For all experiments, we made intravenous injections of indicated imaging agents (0.21 μmol/kg Cy5.5) and analyzed animals using fluorescence molecular tomography (FMT) instrument (FMT1, VisEn Imaging) at the 680 nm channel. Total Cy5.5 (pmols) was calculated by drawing a 3D region of interest (ROI) around indicated tumor areas and values were plotted as a function of tumor mass—acquired as a wet mass (g) after euthanasia and dissection.

Mice were euthanized at indicated times after hPL*-Cy5.5 injection and organs were immediately dissected and analyzed in an OV100 Small Animal Imaging System (Olympus America) using the indicated filter. For intraperitoneal metastases imaging, hPL*-Cy5.5 (0.21 μmol/kg Cy5.5) was injected intraperitoneally in mice fourteen days after cell implantation, euthanized 8 hours postinjection (HPI), dissected organs, and analyzed using the OV100 system.

Ten minutes prior to imaging, 200 μL of 15 mg/mL D-luciferin (Gold Biotechnology) was injected as an intraperitoneal bolus in animals. Animals were imaged on the Xenogen IVIS 200 (Calipar) at indicated intensities. Tumor margins with representative BLI were correlated with two-dimensional (2D) MR images for approximation of 2D slice acquisition.

MR images were obtained from imaged mice (20.7 ± 0.7 g) with subcutaneous tumors (62.1 ± 13.5 mg) 7 to 10 days after implantation in a 30-cm horizontal bore Bruker BioSpec 9.4 Tesla Small Animal MR System (Bruker-Biospin). A homebuilt, half-open birdcage coil (21) was used to acquire T₁-weighted low-resolution 2D gradient echo images with the following parameters: fat suppression turned-on; flip angle, 30°; repetition time, 250 milliseconds; echo time, 5 milliseconds; field of view, 2.56 × 2.56 cm; matrix size, 128 × 128, two averages. High-resolution images (matrix size, 256 × 256 with two averages) were then acquired. We serially imaged the mice at time 0 HPI, immediately after intravenous injection (low dose, 0.36 μmol/kg gadolinium; high dose, 100 μmol/kg gadolinium), 4 HPI, and 24 HPI, with the best images acquired at 24 HPI. We acquired R₁ values using RAREVTR pulse-sequence. We analyzed R₁ values of tumors in slices with high T₁-weighted signal increases (regions of interest, ROI) by normalizing to surrounding muscle tissue. We then subtracted this value from the R₁ tumor to muscle ratio (same ROI) at 0 HPI. We harvested the organs 24 HPI for ICP-MS analysis.

The T₁ values of contrast agents were obtained in PBS using a 5.6 cm vertical bore, 11.7 Tesla vertical magnet (Oxford Instruments) using a Bruker DRX-500 MHZ Avance Spectrometer (Bruker Instruments) at the University of Illinois (Chicago, IL). A rapid acquisition with relaxation enhancement with variable repetition time (RAREVTR) pulse sequence was used to obtain T₁ values. Relaxivity (r₁) was calculated using linear regression analysis of 1/T₁ as a function of gadolinium or protein concentrations.

Synthesized MRI contrast agents were quantified for gadolinium in a trace-metal free 3% HNO₃ (Fluka) matrix solution by X series II (Thermo Electron) ICP-MS at Northwestern University (Evanston, IL). For cell analysis, we washed cells thrice with DPBS, gently removed them from plate surface using a cell lifter, dried at 80°C, digested with 100% HNO₃ and then diluted to 3% HNO₃. Likewise, animal tissues were harvested at 4 and 24 HPI of gadolinium contrast agents given as an intravenous bolus, weighed, digested with 100% HNO₃ and then diluted to 3% HNO₃ for gadolinium analysis.

For PRLR analysis in TMA studies (Table 1), the P-values were calculated using the Kruskal–Wallis tests except for age at diagnosis, stage, and grading where the P-value is from Spearman rank correlation analysis. For studies presented in Supplementary Table S1, McNemar test was performed to compare the proportion of ovarian tumors with a score of 3 to the proportion of omentum samples with a score of 3 (and likewise for the comparison of ovarian versus peritoneal samples). Linear regression analysis was used in the comparison of %ID to tumor mass and calculating R₁ values for contrast agents. In the case of comparing gadolinium uptake in HeyA8 cells, statistical comparisons were performed by analysis of variance. For all other data, Student t tests were used assuming equal variances to compare data pairs. Values of P ≤ 0.05 were considered significant and reported data as mean values ± SEM.

We conducted IHC analysis on tissue microarrays containing FIGO stage I to IV human ovarian cancer (n = 152) and normal ovarian tissue (n = 10) with categorization of PRLR expression into three groups: low, moderate, and high (Fig. 1A, Table 1, and Supplementary Fig. S1). Moderate-to-high PRLR expression was observed in 98% of samples and localized to the epithelial cancer cell compartment but not the tumor stroma such as fibroblasts and mesothelial cells (Supplementary Fig. S2A). Moderate-to-high PRLR expression was present regardless of histologic type, grade, and stage (Table 1). In contrast, analysis of normal ovarian tissue (n = 10) showed very little receptor expression (Fig. 1A, top left, and Supplementary Fig. S2B), consistent with a previous report (22). Furthermore, we observed no significant difference in PRLR expression between primary ovarian tumors and the corresponding tumor metastases in the peritoneum, suggesting that molecular imaging of PRLR should detect ovarian cancer independent of anatomic location (Supplementary Table S1). Collectively, these data establish PRLR as an upregulated cell-surface receptor in epithelial ovarian cancer and implicate it as a potential imaging target both for detecting ovarian cancer and for differentiating malignant ovarian tissue (PRLR⁺) from benign tissue (PRLR⁻).

To allow for attachment of imaging agents to hPL, we generated an isoleucine to cysteine mutation at residue 138 located in a flexible loop region (I138C, referred to hereafter as hPL*), expressed recombinant hPL* in Escherichia coli, and purified it using a C-terminal His₆-tag. According to the crystal structure of ovine placental lactogen (an hPL analogue) in complex with rat PRLR extracellular domain (a human PRLR analogue), the flexible loop resides outside the binding interface, suggesting that mutation within this region will not interfere with hormone binding to the receptor (Fig. 1B and ref. 17). Upon incubation with excess maleimide-Cy5 dye under mild-reducing conditions, hPL* becomes quantitatively labeled (1:1 stoichiometric ratio of dye:protein after purification) but not when pretreated with excess iodoacetamide (Supplementary Fig. S3), consistent with labeling at C138. As a control we constructed a variant of hPL*containing three alanine mutations (TM hPL*) known to reduce binding to the receptor (23). These mutations reside in the site 1 binding region, which includes two residues (E174, H18) in the ligand:receptor interface that interact with a zinc ion, a required cofactor for hPL binding to PRLR (Supplementary Fig. S4; ref. 23). The TM variant also conjugated to maleimide dyes at 1:1 stoichiometric ratios. Purified products indicated single species (Fig. 1C).

We incubated hPL*-Cy5 with frozen sections of normal and malignant human ovarian tissue in the presence of zinc (ZnCl₂). After washing the incubated tissue, the malignant cells but not the normal cells retained Cy5 fluorescence. This signal corresponded to PRLR expression levels and depended upon the presence of zinc, consistent with a signal resulting from hPL*-Cy5 binding directly to PRLR (Fig. 1A, bottom row). These results suggest that hPL*-conjugates may offer a strategy to differentiate between benign (PRLR⁻) tissue and malignant (PRLR⁺) tissue.

To test the efficacy of our concept for clinical MRI, we conjugated hPL* and TM hPL*to a gadolinium contrast agent, GdDTPA (Fig. 1B and C). We linked the DTPA moiety to the proteins using the bifunctional crosslinking agent SPDP, which allowed conjugation of the chelating agent by disulfide exchange. Under the same conditions, wild-type hPL, which lacks a free cysteine, did not react (Supplementary Fig. S5A). To load gadolinium ion into hPL*-DTPA and TM hPL*-DTPA, conjugates were treated with excess gadolinium chloride and dialyzed to remove free gadolinium ion. Isoelectric focusing showed a clear shift in pI after DTPA conjugation and gadolinium loading (Fig. 1C). Analysis by inductively coupled plasma mass spectrometry (ICP-MS) revealed ∼1:1 ratio of gadolinium to protein in each case (Supplementary Fig. S5B).

We first tested whether the hPL I138C mutation and subsequent conjugations affected ligand-induced PRLR-mediated signaling using both T47D cells, a breast cancer cell line overexpressing the long isoform of PRLR, and HeyA8 cells, an ovarian cancer cell line overexpressing a shorter PRLR isoform (Supplementary Fig. S6). Incubation of these cells with hPL, hPL*-GdDTPA, or hPL*-Cy5 in the presence of zinc induced ERK phosphorylation, a downstream PRLR signaling effect (Fig. 1D; ref. 24). In contrast, cells incubated with TM hPL*-GdDTPA, hPL*-GdDTPA without zinc, or hPL*-GdDTPA in the presence of the PRLR-mediated signaling inhibitor, cyclosporine A (CsA; ref. 25), showed attenuated ERK phosphorylation (Fig. 1D and Supplementary Fig. S7). Together, the results indicate that the observed signaling hinges on the integrity of the binding interface between receptor and hormone. To address concerns for possible side effects caused by hPL* activation of the signaling pathway, we assessed the tumorigenic potential of hPL* conjugate exposure, given as a single dose, by counting the colonies formed on soft agar plates of PRLR⁺ ovarian cancer cells. Our results indicate that a short-term exposure (6 hours)—a situation encountered for diagnostic purposes—does not increase tumorogenic activity, whereas long-term exposure (>15 days) appears to increase activity (Supplementary Fig. S8). Overall, our data show that the site-specific modification of the flexible loop region in hPL allows for construction of hPL* conjugates that retain the ability to induce receptor-mediated signaling.

Having demonstrated downstream signaling activity, we proceeded to test whether hPL* and associated conjugates internalize into PRLR⁺ cancer cells, which would allow for the accumulation and retention of imaging agents inside PRLR⁺ cancer cells compared to PRLR⁻ cells. We analyzed cell lysates after incubation of HeyA8 cells with hPL* at an endocytosis permissive temperature (37°C) and at an endocytosis restrictive temperature (4°C). Analysis of untreated cell lysates with an anti-hPL antibody confirmed that HeyA8 cells had very little, if any, endogenous hPL. However, lysate analysis after incubation of HeyA8 cells with hPL* showed significant intracellular levels of hPL* at 37°C compared to incubation at 4°C, suggesting active, rather than passive, internalization, and accumulation (Fig. 2A, top). These findings are consistent with results for PRLR-mediated internalization of prolactin (26, 27).

We next tested the imaging conjugates hPL*-Cy5 and hPL*-GdDTPA for internalization. Immunoblot analysis demonstrated zinc-dependent internalization of hPL*-Cy5 and hPL*-GdDTPA into PRLR⁺ HeyA8 cells. In contrast, TM hPL*-Cy5 and TM hPL*-GdDTPA, which have reduced affinity for the receptor, showed severely attenuated levels of uptake (Fig. 2A, bottom). These differences did not arise from biases in anti-hPL antibody for detection of hPL* conjugates compared with TM hPL* conjugates (Fig. 2A, middle). We further confirmed internalization of hPL*-Cy5 into HeyA8 cells by confocal microscopy (Fig. 2B, top) and internalization of hPL*-GdDTPA by measurement of gadolinium content of cell lysates using ICP-MS (Fig. 2C). Unconjugated GdDTPA (Magnevist), hPL*-GdDTPA in the absence of zinc, and TM hPL*-GdDTPA each showed an approximately four-fold reduction of internalization compared to hPL*-GdDTPA (Fig. 2C). We obtained similar results using T47D cells (Supplementary Fig. S9). These observations and the requirement for zinc strongly suggest active internalization of hPL* conjugates via PRLR-mediated endocytosis.

To directly establish the dependence of internalization upon the presence of the PRLR, we stably transfected HeyA8 cells lines with plasmids that express GFP and contain either a short hairpin (sh) RNA with a scrambled sequence (shScram) or a human PRLR shRNA (shPRLR) cassette. Analysis by Western blot analysis and immunoblotting showed that shPRLR cells expressed less PRLR than untransfected, shScram, or the GFP-luciferase expressing HeyA8 cells (GFP-Luc; ref. 28; Supplementary Fig. S10 and compare with Supplementary Fig. S6A). Cells transfected with the shPRLR plasmid exhibited reduced hPL*-Cy5 and hPL*-GdDTPA internalization compared to cells transfected with the shScram plasmid (Fig. 2A–C). Together these results show that hPL* conjugates internalize specifically into PRLR⁺ cancer cells through PRLR-mediated endocytosis.

Unmodified HeyA8, GFP-Luc HeyA8, and shScram HeyA8 cells formed solid, PRLR⁺ serous-papillary tumors when implanted subcutaneously in athymic nude mice, whereas solid, serous-papillary tumors formed from shPRLR HeyA8 cells had reduced PRLR expression (Fig. 3A and Supplementary Fig. S11). As expected, muscle tissue surrounding tumors showed undetectable PRLR expression in all implanted mice (Fig. 3A). We also observed PRLR expression in murine ovarian tissue (29) in contrast to normal human ovarian tissue, which has undetectable to low levels of PRLR (ref. 30 and Fig. 1A, top).

We then proceeded to test whether the near-infrared fluorescence (NIRF) imaging probe hPL*-Cy5.5 internalizes specifically and accumulates in PRLR⁺ ovarian cancer xenografts in vivo. Intravenous injection of hPL*-Cy5.5 into mice bearing >2 mm-sized HeyA8 xenografts followed by imaging using FMT at 24 HPI showed that hPL*-Cy5.5 localizes to PRLR⁺ GFP-Luc tumors (Fig. 3B). In contrast, TM hPL*-Cy5.5 showed reduced localization. Localization of NIRF signal from hPL*-Cy5.5 and TM hPL*-Cy5.5 correlated with tumor borders defined by bioluminescence imaging (BLI; Fig. 3B). Quantification of Cy5.5 uptake as a function of tumor weight confirmed the increased uptake of hPL*-Cy5.5 relative to TM hPL*-Cy5.5 (Fig. 3C). hPL*-Cy5.5 accumulation in PRLR⁺ HeyA8 tumors occurred in a time-dependent manner (Supplementary Fig. S12A), with maximal Cy5.5 signal at 72 HPI and clearance from the tumor by 192 HPI (Supplementary Fig. S12B). We observe exponential decay pharmacokinetics of hPL*-Cy5 in the blood, which is similar to other reported protein-imaging conjugates (Supplementary Fig. S12C; refs. 31–33). Ex vivo imaging of organs from mice at 24 HPI (Fig. 3D) indicated that beyond the expected localization to PRLR⁺ murine ovarian tissue, as well as liver and kidneys (due to contrast agent metabolism and clearance), hPL*-Cy5.5 preferentially localized to the PRLR⁺ tumor. In contrast, TM hPL*-Cy5.5 exhibited decreased localization to the tumor relative to hPL*-Cy5.5. Notably, hPL*-Cy5.5 does not localize to the PRLR⁻ muscle tissue (Fig. 3D and Supplementary Fig. S12D). Confocal imaging of GFP-Luc HeyA8 xenografts confirmed that hPL*-Cy5.5 localizes to and internalizes into PRLR⁺ tumors (Fig. 3E). As ovarian cancer is a heterogeneous disease with many different subtypes (20), we also tested CaOV3 cells, which form an adenocarcinoma subcutaneous xenograft tumor (34) with high expression of PRLR (Supplementary Fig. S10). We analyzed the CaOV3 xenograft for its capacity to internalize hPL*-Cy5.5 and confirmed tissue-specific uptake by NIRF and ex vivo imaging, consistent with our findings of hPL*-Cy5.5 uptake in PRLR⁺ HeyA8 xenografts (Supplementary Fig. S13). Together, the data strongly suggest that the PRLR mediates hPL*-Cy5.5 uptake in vivo.

To test the potential for hPL*-guided imaging to directly differentiate PRLR⁻ (benign) tissue from PRLR⁺ (malignant) tissue in vivo, we used hPL*-Cy5.5 to image mice implanted with both subcutaneous shPRLR and shScram tumors on contra-lateral hind-limbs with NIRF. As revealed by NIRF using FMT technology, shScram tumors accumulated significantly more hPL*-Cy5.5 at 8 HPI than did shPRLR tumors (Fig. 3F). Additional ex vivo images of excised tumors using Cy5.5 and GFP filters confirmed that the Cy5.5 signal in shPRLR tumors localized to non-GFP expressing tumor tissue (PRLR⁺) with very little colocalization to PRLR⁻ GFP expressing tissue. In contrast, Cy5.5 and GFP signals colocalized in shScram tumors (Supplementary Fig. S14). Thus, hPL*-Cy5.5 can distinguish between high and low PRLR expressing tumors within the same mouse and also can differentiate regions of high and low PRLR expression within the same tumor. Distribution of fluorescence signal in various organs corresponded to results obtained by hPL*-Cy5.5 imaging of live animals bearing GFP-Luc and normal HeyA8 tumors. Overall, these data demonstrate that solid, PRLR⁺ ovarian tumor xenografts internalize hPL*-Cy5.5 in a PRLR-dependent manner and suggest that molecular PRLR imaging can distinguish benign tissue from malignant ovarian cancer.

HeyA8 ovarian cancer cells and GFP-Luc HeyA8 cells form small, solid, PRLR⁺ disseminated peritoneal tumors after intraperitoneal implantation, mimicking the distribution of human ovarian cancer (Supplementary Fig. S15A; ref. 34). To test the sensitivity of molecular PRLR imaging, we targeted small, PRLR⁺ tumors formed from GFP-Luc HeyA8 cells in this intraperitoneal model of ovarian cancer. Ex vivo confocal imaging of peritoneal area 8 HPI of hPL*-Cy5.5 showed tumor nodules within abdominal fat with GFP and Cy5.5 signal colocalization (Supplementary Fig. S15B). Confocal microscopy of frozen tissue confirmed Cy5.5 signal in tumor tissue (Supplementary Fig. S15C) compared to tumor tissue from a noninjected animal. These data demonstrate sensitivity of hPL* conjugates for detecting small tumors in both an intraperitoneal and subcutaneous ovarian cancer model. In addition, we detected no differences in PRLR expression between intraperitoneal and subcutaneous implanted tumors (Supplementary Fig. S15A and Fig. 3A). We also note that PRLR expression was not biased by location in human ovarian cancer either (Supplementary Table S1).

Having demonstrated the efficacy of molecular PRLR imaging for detecting malignant tumors using NIRF, we proceeded to test PRLR directed imaging using MRI. In solution at 11.7 T, the two contrast agents, hPL*-GDTPA and TM hPL*-GdDTPA, showed small increases in relaxivities (r₁) compared to Magnevist (Fig. 4A and Supplementary Fig. S16A). The small increase in r₁ after conjugation of GdDTPA to hPL* or TM hPL* is consistent with the small differences observed by other high molecular weight gadolinium chelates at high magnetic field strengths (35, 36). Given recent concerns of gadolinium stability and toxicity (37), we verified that hPL*-GdDTPA is stable at 37°C for 24 hours compared to Omniscan, a gadolinium chelate with low stability (Supplementary Fig. S16B).

We then tested the ability of hPL*-GdDTPA to detect cancer cells by MRI. Using an in vitro MRI assay similar to that described for T47D cells (38) and previously used for imaging pancreatic islet cells by one of our laboratories (39), we incubated T47D and HeyA8 cells with hPL*-GdDTPA and compared the results to analogous incubations with GdDTPA and PBS. Results showed T₁-weighted signal enhancements and reduction of T₁ relaxation times for T47D and HeyA8 cells treated with hPL*-GdDTPA compared to controls (Supplementary Fig. S17).

We then tested the ability of hPL*-GdDTPA to detect solid, PRLR⁺ tumors formed by HeyA8 cells. In initial experiments, we imaged mice continuously under anesthesia after hPL*-GdDTPA injection. Subtraction of T₁-weighted images acquired at two and four HPI from precontrast images confirmed that T₁-weighted enhancement in the tumor was occurring as a consequence of hPL*-GdDTPA injection (Supplementary Fig. S18A). By comparison, attempts to image solid, PRLR⁺ GFP-Luc and PRLR⁺ SKOV3ip1 tumors (an ovarian tumor derived from subcutaneous implanted SKOV3ip1 cells) immediately after intravenous Magnevist injection at clinical doses (100 μmol/kg) showed poor signal enhancement compared to surrounding muscle tissue (Supplementary Fig. S18B and S18C). We observed T₁-weighted enhancement of PRLR⁺ tumors by hPL*-GdDTPA during serial imaging (Supplementary Fig. S19A).

We then compared hPL*-GdDTPA to TM hPL*-GdDTPA and Magnevist. In mice bearing PRLR⁺, GFP-Luc tumors, intravenous injection of hPL*-GdDTPA at 0.36 μmol/kg resulted in increased T₁-weighted image contrast at 4 and 24 HPI compared to equivalent dosage of Magnevist or TM hPL*-GdDPTA (Fig. 4B). BLI imaging showed that the contrast enhancement corresponded to locations of maximal luciferin signal and included the interior regions of the tumor tissue (Fig. 4B). Measured relaxivities (calculated as a ratio of tumor relaxivity to muscle relaxivity at 24 HPI subtracted from the ratio calculated before injection) confirmed T₁-weighted contrast enhancement at 24 HPI (Fig. 4C and Supplementary Fig. S19B). Tissue biodistribution of gadolinium qualitatively and quantitatively matched hPL*-Cy5.5 distribution (Fig. 3D and Supplementary Fig. S19C) and confirmed that tumors isolated from mice treated with hPL*-GdDTPA contained more gadolinium than did tumors isolated from mice treated with TM hPL*-GdDTPA or Magnevist (Fig. 4D and Supplementary Fig. S19D). These data strongly suggest that hPL*-GdDTPA localizes and accumulates to PRLR⁺ tumors in a manner similar to hPL*-Cy5.5 and enables detection by currently used clinical imaging.

In addition, we tested hPL*-GdDTPA for its ability to image a different solid, PRLR⁺ ovarian tumor with moderate-to-high PRLR expression derived from subcutaneous implanted SKOV3ip1 cells (Fig. 5A and B). Intravenous Injection of hPL*-GdDTPA increased signal intensity of tumors in T₁-weighted MR images compared to TM hPL*-GdDTPA (Fig. 5B), indicating that contrast enhancement provided by hPL*-GdDTPA was not unique to HeyA8 tumors. Finally, we attempted MR imaging of a subcutaneous tumor implanted near the abdomen and successfully detected an ovarian cancer tumor after intravenous injection of hPL*-GdDTPA (Supplementary Fig. S20). The results indicated signal enhancement in tumors compared to small colon, large colon, and peritoneum. We did not find PRLR in omental tissue and detected minimal localization of hPL*-GdDTPA to omentum (Fig. 4D).

Ovarian cancer remains the deadliest of gynecologic malignancies because current methods for detection have low specificity, confounding efforts to identify disease onset or disease recurrence after surgery (7). Moreover, the current practice of MR imaging of the pelvis following intravenous injection of the gadolinium contrast agent Magnevist often poorly distinguishes whether an observed mass contains benign or malignant tissue (11), further confounding efforts to diagnose and treat the disease (3). Here we have demonstrated proof-of-concept for a new, targeted molecular imaging approach for ovarian cancer based upon upregulation of the PRLR and its capacity to mediate ligand-induced endocytosis. Levine and colleagues previously showed that >80% of ovarian cancers express weak to high levels of PRLR regardless of stage, grade, and histology (22). Our own analysis confirmed and extended these observations, revealing that a majority (>98%) of ovarian cancers express moderate to high levels of the PRLR. In addition, we compared metastases to primary tumors and found little difference in PRLR expression between these two subgroups.

To exploit the receptor's capacity to mediate endocytosis, we established chemistry for conjugation of imaging agents to a flexible loop within hPL located outside the receptor interface without affecting the capacity for PRLR mediated endocytosis. hPL has several advantages as a vehicle for imaging: it binds specifically to PRLR with high affinity (23), promotes its own internalization via the PRLR (18), and allows informed design of bioconjugates because the binding epitope has been defined structurally (17). Moreover, hPL is expressed in the placenta and circulates in the serum of pregnant woman at high levels for several weeks (40). This suggests little competition of hPL*-conjugates with endogenous hormone, because pregnant patients are unlikely to be in the ovarian cancer patient population, and suggests that high levels of hormone are tolerated by patients without detrimental effects.

Despite these advantages, the capacity of hPL* conjugates to trigger PRLR signaling raises concerns about possible tumor induction. Given the half-life of our conjugate and clearance by 6 HPI, injection of small amounts of contrast agent required for diagnostic imaging is not expected to have significant effects on tumor growth. Our experiments suggest that only prolonged exposures to hPL* conjugates induced tumor growth. Before proceeding to clinical imaging of PRLR using hPL* conjugates, further validation and toxicity studies in nonmurine animal and nonhuman primate models must be completed.

Cell lines and tumors carrying PRLR efficiently internalized imaging agents fused to hPL, whereas those lacking the receptor or expressing low levels of the receptor due to RNAi knockdown did not, suggesting a specific mechanism of internalization via PRLR. Less efficient internalization of the weaker binding TM hPL* conjugates provides further support for a specific mechanism. In mouse models, the imaging agents accumulated into multiple ovarian cancer tumors expressing moderate to high levels of PRLR relative to surrounding tissue. The NIRF probe, hPL*-Cy5.5, readily detected tumors in mice imaged by FMT, whereas the contrast agent, hPL*-GdDTPA, readily detected tumors by MRI. We systematically show that a single gadolinium-chelate conjugated to hPL* allowed detection of PRLR⁺ ovarian cancer tumors, likely due to a combination of high affinity of hPL to PRLR and high-expression levels of PRLR in ovarian cancer. Although not directly tested in our work, we hypothesize that signal enhancement on T₁-weigted imaging may be augmented by the increased relaxivity of our agent in the intracellular environment, which is analogous to the liver-specific gadolinium chelates that are used clinically (41). In addition, we speculate that hPL*-GdDTPA may have a higher relaxivity at 1.5 or 3.0 T, magnetic field strengths used clinically, owing to the slower rotational correlation times of high molecular weight gadolinium chelates (42).

Our imaging platform achieved sufficient sensitivity to detect small tumors (10 mg) and metastatic ovarian cancer, conferring nearly a 100-fold improvement in the ovarian cancer detection threshold compared to GdDTPA, a contrast agent used in the clinic currently. Coupled with the expression of PRLR in early-stage tumors (Table 1), our ability to image multiple human ovarian cancer xenografts that express moderate to high levels of PRLR supports the potential for clinical translatability and possible improvement of the MRI detection sensitivity of human ovarian cancer in patients.

As we proceed toward translating molecular PRLR imaging clinically, further optimization of hPL* conjugates may be necessary. Testing the distribution and metabolism of hPL*-conjugates in nonmurine and primate animal models may indicate necessary alterations of the protein, chelating moiety, or conjugation chemistry. Other design considerations include using antibody fragments that bind specifically to PRLR as designed by one of our laboratories (43) or increasing the gadolinium payload by coupling hPL* to particles carrying multiple gadolinium ions (unpublished data).

Taken together, the PRLR targeted imaging platform may endow clinicians with a potential method to distinguish malignant from nonmalignant tissue with high specificity, engendering both a preoperative diagnostic tool and a postoperative surveillance tool for tumor detection and recurrence that could reduce the large number of unnecessary surgeries. Internalization of folate bioconjugates via folate receptor-mediated endocytosis has enabled detection of metastatic ovarian cancers in humans undergoing cytoreductive surgery (44), progressing beyond detection of tumors in animal models. Considering that PRLR upregulation occurs with greater frequency in ovarian cancer than does folate receptor upregulation (>98% vs. 72%, cf. this work and ref. 45), analogous approaches using PRLR-mediated internalization of hPL bioconjugates could potentially be beneficial to a greater fraction of ovarian cancer patients.

In summary, upregulation of PRLR in ovarian cancer and its use through targeted molecular imaging sets the stage for advances in the non-invasive diagnosis and treatment of ovarian cancer. In developing this technology further, it will be necessary to more fully define PRLR expression ranges with respect to ovarian cancer and other gynecologic pathologies. We envision adapting our approach to other imaging modalities such as PET or combined approaches such as MRI/PET. We speculate that our PRLR imaging concept would confer similar advantages to diagnosis of breast cancers, which have strongly upregulated PRLR expression compared to normal breast tissue (46). Beyond imaging, the capacity of PRLR targeting to internalize cargo in a cell-type-specific manner may hold promise as a therapeutic delivery system.

The University of Chicago filed a patent for the hPL-gadolinium conjugate on behalf of A.A. Kossiakoff, E. Lengyel, and J.A. Piccirilli. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K.M. Sundaram, A.K. Mitra, A.A. Kossiakoff, E. Lengyel, J.A. Piccirilli

Development of methodology: K.M. Sundaram, Y. Zhang, A.K. Mitra, J.-L.K. Kouadio, A.A. Kossiakoff, B.B. Roman, J.A. Piccirilli

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.M. Sundaram, Y. Zhang, A.K. Mitra, E. Lengyel, J.A. Piccirilli

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.M. Sundaram, Y. Zhang, A.K. Mitra, J.-L.K. Kouadio, K. Gwin, J.A. Piccirilli

Writing, review, and/or revision of the manuscript: K.M. Sundaram, Y. Zhang, J.-L.K. Kouadio, B.B. Roman, E. Lengyel, J.A. Piccirilli

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.M. Sundaram, E. Lengyel, J.A. Piccirilli

Study supervision: B.B. Roman, E. Lengyel, J.A. Piccirilli

Other (review of pathology including immunohistochemical studies and TMAs): K. Gwin

We thank the Quantitative Bioelemental Imaging Core in the Chemistry of Life Processes Institute (Northwestern University) for use of the ICP-MS. We are grateful to the laboratory of H. Singh for allowing use of the FV1000 confocal microscope and to L. Gerhold and J. Souris at the Optical Imaging Core Facility for assistance in NIRF imaging. X. Fan at the Magnetic Resonance Imaging and Spectroscopy Laboratory and the University of Chicago Comprehensive Cancer Center assisted with MR image acquisition. We are grateful to C. Gong at the Human Tissue Resource Center for immunohistochemical staining and A. Oto in the Department of Radiology for insightful comments.

This work was supported in part by the Medical Scientist Training Program (K.M. Sundaram), Howard Hughes Medical Institute (J.A. Piccirilli), Woman's Board Grant (E. Lengyel and J.A. Piccirilli), Burroughs Wellcome Fund (E. Lengyel), Noreen Fraser Foundation (E. Lengyel), Ovarian Cancer Research Fund (E. Lengyel), NIH Grant R01GM088656 (J.A. Piccirilli), and the BSD Imaging Research Institute Pilot Grants (J.A. Piccirilli, E. Lengyel, and B.R. Roman).

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.