2.1.

Fluorophores & Antibodies

Alexa Fluor 647 (AF647; Thermo Fisher Scientific, Waltham, Massachusetts, United States) and Cy3B (GE Healthcare Life Sciences, Little Chalfont, United Kingdom) were purchased in their N-hydroxysuccinimide ester form and then solubilized in anhydrous dimethyl sulfoxide (DMSO) at 10 mM for antibody conjugation. For the targeted probe, PSMA targeted monoclonal antibody (Mab) J533 IgG was used. The mouse hybridoma Prost J533 (HB-12127) was purchased from American Type Culture Collection (ATCC; Manassas, Virginia, United States). The hybridoma was used to produce the J533 antibody as the IgG, K isotype, which was purified using mercaptoethylpyridine/protein A chromatography by the Monoclonal Antibody Core at Oregon Health and Science University (OHSU). For the untargeted probe, donkey anti-rabbit (DkRb) IgG (Jackson ImmunoResearch, West Grove, Pennsylvania, United States) was used.

2.2.

Cell Lines

The human prostate carcinoma cell line LNCaP clone FGC was cultured in RPMI 1x (Gibco/Thermo Fisher Scientific, Waltham, Massachusetts, United States) with 10% fetal bovine serum ([FBS], Seradigm, Sanborn, New York, United States) and 1% penicillin–streptomycin–glutamine (Thermo Fisher Scientific). The human prostate grade IV adenocarcinoma cell line PC-3 was cultured in F-12K 1x (Gibco/Thermo Fisher Scientific) with 10% FBS and 1% penicillin–streptomycin–glutamine. All cell lines were cultured to $\sim 90\%$ confluence prior to use.

2.3.

Antibody-Fluorophore Conjugation

The targeted probe J533 IgG was conjugated to AF647, and the untargeted probe DkRb IgG was conjugated to Cy3B. Each probe was prepared individually, explained briefly as follows. First, the antibodies were buffer exchanged into 1x phosphate-buffered saline (PBS) at pH 8.0. Following this, 10 mM AF647 in anhydrous DMSO was added to the J533 IgG, and 10 mM Cy3B in anhydrous DMSO was added to the DkRb IgG, resulting in a 5:1 fluorophore-to-antibody molar ratio in a total volume of 1 mL for each solution diluted in 1x PBS at pH 8.0. The solutions were shaken gently for 3 h at room temperature (RT), protected from light. The resulting solutions were then concentrated in 10 kDa molecular weight cut-off spin filters (Amicon Ultra 0.5 mL 10 kDa, Fisher Scientific) into clean microcentrifuge tubes to remove any unconjugated fluorophore. The fluorophore-to-protein ratio for each conjugated antibody was quantified using absorbance spectroscopy (SpectraMax M5 Microplate Reader, Molecular Devices, Sunnyvale, California, United States). Antibody absorbance was measured at 280 nm (J533 and DkRb IgG estimated extinction coefficient $(\epsilon )=\mathrm{210,000}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$). AF647 absorbance was measured at 650 nm (AF647 $\epsilon =\mathrm{270,000}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$). Cy3B absorbance was measured at 560 nm (Cy3B $\epsilon =\mathrm{130,000}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$). Calibrated absorbance and fluorescence spectra and the Beer–Lambert law were used to determine the final concentrations of fluorophore and antibody for each conjugated antibody. All antibody conjugates had a fluorophore-to-antibody ratio of $\sim 2:1$.

2.4.

Flow Cytometry

LNCaP and PC-3 cells were trypsinized, counted, and resuspended to $1\times {10}^7$ cells each in 1 mL of 1x PBS. Cells were then fixed in 4% paraformaldehyde (PFA) for 10 min, followed by 3-min permeabilization in 1x PBS + 0.1% Tween-20. The cells were then washed with 1x PBS and blocked with 5% FBS in 1x PBS for 15 min. $20\mu \mathrm{g}/\mathrm{mL}$ of the J533-AF647 conjugate was added to each sample resulting in a final staining concentration of $10\mu \mathrm{g}/\mathrm{mL}$. The cells were washed with 1x PBS $+0.5\%$ Triton-X for 5 min, followed by two washes with 1x PBS for 5 min each. The cells were resuspended in $500\mu \mathrm{L}$ of fresh 1x PBS immediately prior to analysis on a Becton Dickinson LSR Fortessa (Becton, Dickinson and Company, Franklin Lakes, New Jersey, United States). To detect AF647, the flow cytometer was configured to the 640-1 (650/665) Cy5 channel. A minimum of $1\times {10}^5$ cells were counted for each sample. To quantify PSMA receptor number, Quantum^TM AF647® molecules of equivalent soluble fluorophore beads (Bangs Laboratory, Inc., Fishers, Indiana, United States) were quantified prior to the LNCaP and PC-3 PSMA stained samples.

2.5.

Spinning Disk in Vitro Immunofluorescence

LNCaP and PC-3 cells were trypsinized, counted, and resuspended to a final concentration of $1.5\times {10}^5$ cells per well into a 96 well glass bottom plate (Cellvis P96-1.5H-N, Mountain View, California, United States). Cells were fixed with 4% PFA for 15 min at RT, followed by two washes with 1x PBS for 5 min each at RT. Cells were then blocked with 5% FBS in 1x PBS for 15 min at RT. Blocking solution was then carefully removed without additional washing. The cells were then stained with the J533-AF647 conjugate for 30 min at RT with final staining concentrations of 1, 5, 10, and $20\mu \mathrm{g}/\mathrm{mL}$. Cells were washed with 1x PBS + 0.5% Triton-X for 5 min at RT, followed by a wash with 1x PBS for 5 min at RT. Cell staining with the J533-AF647 conjugate was compared to an unstained control sample. A Hoechst nuclear stain (Thermo Fisher Scientific) was performed on all samples with a final staining concentration of $1\mu \mathrm{g}/\mathrm{mL}$. The cells were then imaged at $20\times$ on a Yokogawa CSU-X1 Zeiss Axio Observer microscope (Oberkochen, Germany) in the UV and Cy5 channels.

2.6.

Mice

Separate cohorts of male athymic nude mice (32 to 38 days old, Homozygous 490, Charles River Laboratories, Wilmington, Massachusetts, United States) weighing 19 to 21 g were used to grow LNCaP and PC-3 tumor xenografts. All animal studies were approved by the Institutional Animal Care and Use Committee at OHSU (Protocol No. TR01_IP0000202).

2.7.

Tumor Implantation and Growth

All mice were anesthetized with a 100 mg/kg ketamine (Hospira Inc., Lake Forest, Illinois, United States) and 10 mg/kg xylazine (AnaSed, Shenandoah, Indiana, United States) solution injected intraperitoneally. The toe pinch method was utilized to ensure the mice were fully anesthetized prior to tumor implantation. The lower peritoneal region of each mouse was prepared in a sterile field using povidine-iodine (Purdue Products, Stamford, Connecticut, United States). For LNCaP and PC-3 xenografts, $200\mu \mathrm{L}$ of cell suspension in fresh media ($1\times {10}^6\mathrm{cells}$) was injected into both hind flanks of each mouse subcutaneously. The mice were monitored weekly following implantation for tumor growth and general well-being.

2.8.

Tumor Resection & DDSI Staining

All mice were euthanized via carbon dioxide asphyxiation followed by cervical dislocation prior to tumor resection, which is compliant with the recommendations by the panel on euthanasia of the American Veterinary Medical Association. LNCaP tumor bearing mice were euthanized after 6 weeks of tumor growth or a maximum tumor size of $1{\mathrm{cm}}^3$. PC-3 tumor bearing mice were euthanized after 4 weeks of tumor growth or a maximum tumor size of $1{\mathrm{cm}}^3$. For each cell line, two cohorts of $n=6$ tumors were grown in $n=3$ mice and were extracted for DDSI staining. Thus, in total, $n=12$ tumors per tumor type were grown in $n=6$ mice per tumor type. Each extracted tumor was bisected for DDSI staining, where the cut faces were imaged. For the first cohort of mice, corresponding normal tissues, including a peritoneal adipose sample and a peritoneal muscle sample, were extracted from each mouse for comparison. For the second cohort of mice, corresponding normal tissues, including a peritoneal adipose sample, a peritoneal muscle sample, and normal prostate tissue, were extracted from each mouse for comparison. Each bisected LNCaP and PC-3 tumor sample was blocked, stained, and washed together in a single Eppendorf tube according to an optimized, previously published procedure,³⁷ explained briefly as follows. Each xenograft type was blocked with 1 mL of 5% bovine serum albumin (BSA) solution in 1x PBS at pH 7.4 for 2 min, stained with 1 mL of 200 nM J533-AF647 + 200 nM DkRb-Cy3B solution in 1x PBS at pH 7.4 containing 1% BSA and 0.1% Tween-20 for 1 min, and then washed in 50 mL of 1x PBS containing 0.1% Tween-20 for 3 min. The tissue samples were imaged immediately after conclusion of the wash step with the bisected tumor cut-face oriented towards the light source and camera.

2.9.

DDSI Macroscopic Imaging

White light and fluorescence images of tumor, adipose, muscle, and prostate tissues were collected using a previously described custom-built wide-field fluorescence imaging system consisting of a QImaging EXi Blue monochrome camera (Surrey, British Columbia, Canada) for fluorescence detection with a removable Bayer filter and a near infrared (NIR) PhotoFluor II light source (89 North, Burlington, Vermont, United States).⁵⁷ The broad band light source (360 to 800 nm) was filtered using a $545\pm 12.5\mathrm{nm}$ or a $620\pm 30\mathrm{nm}$ bandpass excitation filter for fluorescence excitation of Cy3B and AF647, respectively. Fluorescence images were collected with a $605\mathrm{nm}\pm 35\mathrm{nm}$ or a $700\mathrm{nm}\pm 37.5\mathrm{nm}$ bandpass emission filter for Cy3B or AF647, respectively. All filters were obtained from Chroma Technology (Bellows Falls, Vermont, United States). To facilitate image calibration, an aliquot of the staining solution was placed in a covered optical well plate (Greiner Bio-One, Monroe, North Carolina, United States) and imaged at the beginning of each DDSI tissue staining experiment.

2.10.

DDSI Image Processing

The collected targeted and untargeted fluorescence images were processed using custom written MATLAB code (MathWorks, Natick, Massachusetts, United States) to generate DDSI maps of each set of tumor, adipose, muscle, and prostate tissues described as follows. First, the median background signal from a user-selected region of interest (ROI) absent of resected tissue was subtracted from the entire image. Separate ROIs were then selected within an equal volume of each probe in solution that was imaged in a 96 well plate to calculate the mean intensity value of the targeted and untargeted staining solutions. These values were subsequently used to normalize the targeted and untargeted images permitting calculation of DDSI tissue map. Each pixel value within the ROI in both targeted and untargeted fluorescence images was divided by the average intensity value of the ROI representative of the DDSI staining solution in their respective imaging channels. Each tissue sample ROI was selected to encompass the bulk of tissue without containing any of the tumor edges so as to ensure that all pixels selected corresponded to the tissue type. User-created manual masks were used to identify tissue sample ROIs containing tumor or normal (i.e., adipose, muscle, or prostate) tissues based on white light images alone to minimize selection bias that could come from processing the fluorescence images. These user-created manual masks on white light images were overlaid onto each normalized fluorescence image (targeted and untargeted), and final DDSI images were then calculated as $I_{\text{DDSI}}=(I_{\text{Targeted}}-I_{\text{Untargeted}})/I_{\text{Untargeted}}$, where $I$ = signal intensity/pixel.

2.11.

H&E Staining and Immunohistochemistry

Following DDSI, each tumor and normal tissue set were flash frozen in optical cutting temperature compound to preserve the tissue and its orientation. Frozen blocks were sectioned at $10\mu \mathrm{m}$ thickness onto superfrost plus slides, where serial sections were used for immunofluorescence (IF) and gold standard hematoxylin and eosin (H&E) staining described briefly as follows. For IF-stained slides, standard indirect IF staining protocols were used, where unconjugated J533 antibody and AF488 donkey antimouse antibody were used for detection and imaging. H&E staining was performed on serial sections by the OHSU Histology Shared Resource. Both IF- and H&E-stained slides were imaged using the Zeiss AxioScan.Z1 microscope (Zeiss) at $10\times$ magnification. Images of the entire field were stitched together using the ZEN software.

2.12.

Statistical Analysis

All statistical analysis was performed using custom written MATLAB code. Tumor-to-normal adipose, muscle, and prostate tissue diagnostic detection ability was determined by ROC curves generated for the untargeted probe images, targeted probe images, and calculated DDSI images. The perfcurve function in MATLAB was used to calculate area under the curve (AUC) values on a pixel-by-pixel basis with individual pixel values for each tissue type used as the response variable input. A threshold variable was generated with a linearly increasing value from the minimum to maximum pixel intensity with 100 times fewer number of values than the total number of pixel values. The true positive rate (percentage of tumor pixels greater than threshold), false positive rate (percentage of normal pixels greater than threshold), ROC curves, and corresponding AUC values were then generated and plotted at each threshold value. Histogram plots for tumor-to-normal adipose, muscle, and prostate tissue were generated, on which the optimal threshold point determined via ROC curve analysis was plotted.

3.1.

In Vitro PSMA Expression of Prostate Cancer Cell Lines

Prostate cancer cell lines with varied PSMA expression in vitro were used, where the highly PSMA expressing LNCaP cell line ($1.7\times {10}^5$ receptors per cell) was quantitatively compared with the minimally PSMA expressing PC-3 cell line [$6.6\times {10}^3$ receptors per cell, Fig. 1(a)]. Adherent LNCaP and PC-3 cells were stained for PSMA in vitro, where the expected positive membrane-staining pattern for LNCaP cells was seen with decreasing intensity as primary antibody concentration decreased. PC-3 cells showed minimal staining even at the highest primary antibody concentration [Fig. 1(b)], demonstrating the PSMA antibody specificity. IF staining of the resected LNCaP and PC-3 xenograft tumors confirmed that PSMA expression was retained when the prostate cancer cell lines were grown as tumor xenografts [Fig. 1(c)].

Fig. 1

Prostate cancer cell line PSMA receptor quantification. (a) Flow cytometry-based quantification of LNCaP and PC-3 PSMA receptor number, where $n=3$ samples per cell line were used to quantify expression. (b) Fixed cell in vitro IF staining of PSMA expression was completed in LNCaP and PC-3 cells stained with J533-AF647 at varied concentrations (1, 5, 10, and $20\mu \mathrm{g}/\mathrm{mL}$). PSMA = green, Hoechst = blue. All images are shown with equivalent contrast and brightness. (c) Validation of PSMA expression in representative tumor tissue samples from LNCaP (top) and PC-3 (bottom) xenografts. Serial sections of representative resected specimens from each xenograft type were stained with gold standard H&E and indirect IF using the J533 antibody to validate PSMA expression in the model tissue. The black square on the H&E image represents the magnified region shown in the PSMA IF stained images. PSMA = green, DAPI = blue, H&E = hematoxylin and eosin.

3.2.

PSMA DDSI Staining of Prostate Tumor Xenografts

The PSMA positive prostate cancer-specific DDSI probe pair (J533-AF647 + DkRb-Cy3b) and the previously published DDSI staining protocol optimized for breast cancer were used to establish the accuracy of DDSI for prostate cancer detection. LNCaP ($n=6$), and PC-3 ($n=6$) tumor xenograft samples were stained with the DDSI staining protocol with corresponding normal adipose, muscle, and prostate tissues. Notably, little difference was seen between the targeted and untargeted probe uptake of the highly PSMA expressing LNCaP tumors versus the minimally PSMA expressing PC-3 tumors due to nonspecific probe uptake after topical administration. In addition, substantial untargeted probe uptake was seen in both tumor types as well as in the normal tissues [i.e., adipose, muscle, and prostate, Figs. 2(a) and 3(a)], again due to nonspecific probe uptake after topical administration. The similar nonspecific uptake between the targeted probes in the xenografts with substantial difference in PSMA expression as well as the substantial uptake of untargeted probe clearly demonstrated the need for ratiometric imaging for quantitative, molecular-specific margin assessment.

Fig. 2

PSMA DDSI staining pattern and quantification in LNCaP tumor xenografts. (a) Representative white light, untargeted fluorescence (Cy3B), PSMA targeted fluorescence (AF647), and calculated DDSI images of LNCaP tumor versus normal adipose, muscle, and prostate tissues. LNCaP tumor and normal (b) prostate, (c) adipose, and (d) muscle tissue pixel intensity histograms for untargeted, targeted, and DDSI images following staining. All images and histograms are shown for a representative tumor and normal (i.e., prostate, adipose, and muscle) set of tissues and are representative of the data collected for $n=6$ tumor versus benign normal tissue pairs.

Fig. 3

PSMA DDSI staining pattern and quantification in PC-3 tumor xenografts. (a) Representative white light, untargeted fluorescence (Cy3B), PSMA targeted fluorescence (AF647), and calculated DDSI images of PC-3 tumor versus normal adipose, muscle, and prostate tissues. PC-3 tumor and normal (b) prostate, (c) adipose, and (d) muscle tissue pixel intensity histograms for untargeted, targeted, and DDSI images following staining. All images and histograms are shown for a representative tumor and normal (i.e., prostate, adipose, and muscle) set of tissues and are representative of the data collected for $n=6$ tumor versus benign normal tissue pairs.

DDSI showed substantial contrast difference between the PSMA positive LNCaP tumors and their respective normal tissues [Fig. 2(a)]. In addition, DDSI showed substantial contrast difference between the highly PSMA expressing LNCaP and minimally PSMA expressing PC-3 tumors [Figs. 2(a) and 3(a)] due to the ratiometric correction for the nonspecific uptake patterns in resected tissues. Notably, LNCaP tumors presented with two phenotypes, which appeared as highly vascular versus relatively avascular, where the vascular LNCaP tumors showed lower uptake of the targeted and untargeted antibody stains compared to relatively avascular LNCaP tumors. However, these apparent fluorescence differences were largely corrected using the ratiometric DDSI protocol [Fig. 2(a)]. Notably, since visible fluorophores were utilized for the DDSI staining protocol, evaluation of PSMA expression was largely limited to the tissue surface, which is representative of the surgical margin to be assessed. Histograms of targeted pixel intensities across the sample set (LNCaP, $n=6$) showed strong overlap between the tumor and normal tissues [i.e., prostate and adipose tissues, Figs. 2(b) and 2(c)]. Of note, some targeted pixel intensity separation was seen between the LNCaP tumors and the normal muscle tissue [Fig. 2(d)]. However, when the DDSI pixel intensities were evaluated across the LNCaP sample set, substantial robust quantitative separation between tumor tissues and all normal tissue types (i.e., prostate, adipose, and muscle) was seen [Figs. 2(b)–2(d)].

The PC-3 tumor phenotypes were more homogenous, where similar targeted and untargeted uptake was seen between all representative tumors and stained normal tissues, with the exception of the prostate which had relatively low targeted and untargeted uptake. DDSI images showed minimal contrast difference between the PC-3 tumors and the normal tissues, since all tissues had relatively low PSMA expression [Fig. 3(a)]. Notably, histogram pixel intensities across the sample set (PC-3, $n=6$) showed separation of tumor to prostate signal for both the untargeted and targeted probe uptake [Fig. 3(b)]. By comparison, the histogram pixel intensities showed substantial overlap between the tumor and normal tissues (i.e., adipose and muscle) for both the targeted and untargeted uptake [Figs. 3(c) and 3(d)]. However, the DDSI histogram pixel intensities normalized all differences in signal between the PC-3 tumors and all normal tissue types (i.e., prostate, adipose, and muscle), the expected result given the low PSMA expression in these tissues [Figs. 3(b)–3(d)].

3.3.

ROC Curve Analysis of DDSI Accuracy

The pixel intensity data from combined LNCaP tumors ($n=6$) and the combined PC-3 tumors ($n=6$) as well as their normal tissue counterparts were used to construct ROC curves of the targeted, untargeted, and DDSI contrast (Fig. 4). Minimal difference between the LNCaP tumors and their corresponding normal tissues was seen with either the targeted or untargeted probes alone. However, high diagnostic accuracy was obtained using the DDSI metric for LNCaP tumor to prostate (ROC AUC = 0.99), adipose (ROC AUC = 0.97), and muscle [ROC AUC = 0.98, Figs. 4(a)–4(c)]. Similarly, minimal difference between the PC-3 tumors and their corresponding normal tissue was seen with either the targeted or untargeted probes. However, given the minimal expression of the PSMA biomarker, using the DDSI metric for the PC-3 tumors did not improve the tumor to prostate (ROC AUC = 0.34), adipose (ROC AUC = 0.18) or muscle (ROC AUC = 0.21), which was not diagnostically useful [Figs. 4(d)–4(e)].

Fig. 4

ROC curve analysis. ROC curves and calculated AUC values for untargeted, targeted, and DDSI images of LNCaP versus normal (a) prostate, (b) adipose, and (c) muscle following PSMA staining. ROC curves and calculated AUC values for untargeted, targeted, and DDSI images of the PC-3 tumors versus normal (d) prostate, (e) adipose, and (f) muscle following PSMA staining. All ROC curves and AUC calculations represent the collective data for all $n=6$ tumor versus normal tissue pairs. UT, untargeted; T, targeted.