Preclinical evaluation of PSMA-targeted ultrasound contrast agents in an orthotopic model of prostate cancer in rabbits

Abstract

The localization of prostate cancer by ultrasound remains limited by the lack of B-mode conspicuity and the confinement of clinically approved microbubbles (MBs) to the vasculature. This precludes differentiating viable tumor, necrotic tissue, and margin-associated disease. We investigated prostate-specific membrane antigen (PSMA)-targeted lipid-shelled perfluorocarbon nanobubbles (PSMA-NBs) in an orthotopic rabbit model using a clinical contrast-enhanced ultrasound (CEUS) system. We implanted PSMA-positive PC3pip-GFP tumors into the prostates of immunosuppressed New Zealand White rabbits and performed transabdominal imaging with PSMA-NBs, MBs, and Plain-NBs using identical protocols. To address tumor heterogeneity and ultrasound boundary ambiguity, regions of interest were defined from baseline B-mode images and segmented into the tumor core, rim, and a peritumoral area. Pixel-wise parametric and decorrelation time (DT) maps were generated and compared with whole-slide histology (H&E) and, in an exploratory and non-specific analysis, with PSMA IHC. Compared to MBs at the doses used, PSMA-NBs exhibited higher peak intensities in the tumor core and rim (1.60-fold and 1.50-fold, respectively) and improved retention (mean transit time [MTT]: 4.20 to 5.40-fold higher) for up to 10 min in the tumor and peritumoral areas. In an exploratory analysis constrained by cohort size, PSMA-NB kinetics, notably MTT, tracked histology-defined tumor viability, and DT mapping showed spatially heterogeneous retention at the tumor periphery. Compared to Plain-NBs, PSMA-NBs also exhibited improved retention (MTT +21% overall) in the rim and peritumoral areas. This study demonstrates the potential of PSMA-NBs to characterize prostate cancer by molecularly targeted CEUS beyond that achieved with MBs at the doses used.

Graphical abstract

Highlights

• •PSMA-NB time–intensity-curve kinetics differentiate viable tumor from necrotic/stromal regions in orthotopic prostate tumors.
• •Decorrelation-time mapping reveals prolonged, zone-specific PSMA-NB retention in viable tumor rim, providing a novel readout of intratumoral contrast kinetics on a clinical scanner.
• •PSMA-NBs show higher peak CEUS signal and markedly prolonged persistence versus clinically available microbubbles (Lumason®).
• •Ligand-mediated targeting improves retention-sensitive metrics versus untargeted nanobubbles, indicating benefits beyond passive extravasation.
• •Pixel-wise multiparametric maps capture intratumoral heterogeneity and highlight biologically active rim/peritumoral zones on a clinical ultrasound platform.

1. Introduction

Prostate cancer is the second leading cause of cancer-related deaths in men in the United States, with over 30,000 deaths occurring every year [1,2]. Despite the progress made in the diagnostic and therapeutic modalities, the annual medical expenditure for prostate cancer in the United States has been estimated at US$24.8–39.2 billion, making it imperative to develop more accurate and cost-effective diagnostic modalities [3]. The conventional diagnostic modalities involve the use of ultrasound (US)-guided prostate biopsy in patients with elevated levels of prostate-specific antigen or abnormal digital rectal examination findings [[4], [5], [6]]. Magnetic resonance imaging (MRI)-US fusion-guided biopsies are considered the gold standard, but their limited availability outside of Europe and North America means that B-mode US remains the primary worldwide method for guiding prostate biopsies [[5], [6], [7], [8]]. However, the low sensitivity and resolution of the ultrasound modality to distinguish between normal and malignant tissues make it necessary to obtain multiple biopsy cores, thereby increasing the risk of complications, including significant infection rates [[9], [10], [11]].

To improve soft tissue contrast and structure visualization with US, microbubble (MB) ultrasound contrast agents (UCAs) were introduced, primarily for cardiovascular applications [12,13]. More recently, FDA-approved UCAs (e.g., Lumason®) have been employed for the detection of neoplastic lesions [14]. However, the intravascular nature of MBs, due to their 1-5 μm diameter, restricts their utility in tumors to the vasculature. Nonetheless, there may be a benefit to additional assessment beyond vascular characteristics. Nanobubble (NB) UCAs were developed as an alternative to MBs and can provide complementary information that has the potential to probe the extravascular tumor biology [15,16]. With diameters of 200-500 nm, NBs can extravasate from pathologically permeable blood vessels, commonly seen in inflammatory conditions, providing information about the surrounding environment and parenchyma[[17], [18], [19]].[17], [18], [19] [[17], [18], [19]]

As tumors grow, they frequently develop a necrotic core due to inadequate vascular supply [20,21]. This core, characterized by metabolic stress and inflammation, is a hallmark of advanced disease and presents a significant challenge for imaging technologies, which struggle to delineate tumor boundaries and differentiate between viable and necrotic tissue [20,22]. An ideal contrast agent should not only identify the tumor, but also distinguish its regions, including the core, rim, and surrounding inflamed tissue, to improve both diagnostic accuracy and therapeutic planning. NBs are excellent candidates for this purpose due to their ability to extravasate and linger in the tissues[17,19].

Prostate-specific membrane antigen (PSMA) is a molecular target in prostate cancer that has been validated in the clinic by its high-level expression in a significant proportion of clinically significant prostate cancer and metastatic sites, as well as its established role in imaging and therapeutic interventions [23,24]. In current clinical practice, PSMA positron emission tomography (PET) has become a mainstay in the evaluation of prostate cancer patients, including the initial evaluation of those with unfavorable intermediate- and high-risk prostate cancer, as well as the evaluation of patients with suspected biochemical recurrence [24,25]. In these contexts, the modality is often found to impact the overall management strategy by increasing the sensitivity to detect nodal and metastatic sites compared to other imaging modalities [24]. Moreover, PSMA-targeted theranostics, such as radioligand therapy using PSMA-targeting agents, underscore the importance of PSMA expression and the need to accurately characterize PSMA in prostate cancer[23].

In the context of the current research, PSMA-targeted NBs (PSMA-NBs), which have shown promising results in preclinical studies [[26], [27], [28]], could potentially further improve the accuracy of prostate cancer imaging before, during and after an interventional procedure such as prostate biopsy or focal therapy. The submicron hydrodynamic diameter of these NBs enables their extravasation into the extracellular matrix and their retention in PSMA-expressing tissues for long periods compared to MBs, especially in the setting of molecular targeting [27]. This is particularly relevant because ultrasound remains broadly accessible, yet lacks an established clinically deployed molecular imaging analogue to PSMA PET; a PSMA-targeted contrast-enhanced ultrasound (CEUS) approach could therefore complement current imaging workflows by enabling bedside, real-time assessment of PSMA-expressing disease. The long retention times are essential because they create opportunities for both imaging and therapeutic applications of these NBs. Previous studies have shown that prostate cancer cells expressing PSMA can be preferentially visualized using PSMA-NBs with CEUS, mediated by pharmacokinetics that are distinct from those of plain/untargeted NBs. These NBs interact with cell surface receptors and are internalized via receptor-mediated endocytosis [[27], [28], [29]]. However, this targeted-NB imaging work, spanning PSMA- and other receptor-targeted agents, has been limited to mouse models [26,[29], [30], [31]], which do not adequately mimic the prostate cancer environment in humans[32].

The present study aims to validate PSMA-NB CEUS in a large-animal model that is more comparable to human prostate cancer, and can accommodate an orthotopic tumor to enable the comparison of agent activity and detection sensitivity in cancerous versus normal prostate tissue. While orthotopic mouse models have been explored for this purpose, the murine normal prostate has a highly distinct anatomy and it cannot accommodate the size of lesions which would be of clinical relevance, and the tumors do not develop in a manner that is consistent with human prostate cancers. Following a previously established rabbit orthotopic model [33], here we aim to demonstrate the ability of PSMA-NBs to differentiate between necrotic and viable tumor areas, PSMA-expression in tumors as well as normal prostate tissue, with the potential to improve diagnostic accuracy. With the validation of PSMA-NBs in a large-animal model, we are a step closer to clinical trials.

2. Methods

Industry Support: Siemens® Healthineers (Erlangen, Germany) provided the ultrasound system used in this study (Acuson S3000 with 18L6 transducer). The company had no involvement in study design, data analysis, or manuscript preparation. All authors had full control of the data and submission, and none are affiliated with Siemens® Healthineers.

Ethics Approval: All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) and were conducted in accordance with the National Institutes of Health (NIH) guidelines and relevant institutional regulations for the care and use of laboratory animals.

2.1. Ultrasound contrast agent preparation

The preparation and characterization of the Plain-NB (untargeted) and PSMA-NBs (targeted) contrast agents followed established protocols [15,27]. In summary, the lipids DBPC (Avanti Polar Lipids Inc., Pelham, AL, United States), DPPE (Corden Pharma, Switzerland), DPPA (Corden Pharma, Switzerland), and mPEG-DSPE2000 (Laysan Lipids, Arab, AL, United States) were dissolved in a mixture with propylene glycol (PG, Sigma Aldrich, Milwaukee, WI, United States), and glycerol (Thermo Fisher Scientific, Fair Lawn, NJ, United States). Following dissolution, the lipids were dispersed in prewarmed phosphate-buffered saline (PBS, Gibco Life Technologies, Grand Island, NY, United States) to form an oil-in-water emulsion. The emulsion was placed in a 3 mL vial and sealed, and the vial was evacuated and filled with perfluoropropane gas (C₃F₈, AirGas, Radnor, PA, United States). For PSMA-NB, 25 μL of DSPE-PSMA-1 at a concentration of 1 mg/mL was added into the vial along with the lipid dispersion prior to evacuation. The DSPE-PSMA-1 conjugate was custom-synthesized in-house; synthesis and characterization of the PSMA-1 peptide ligand and the DSPE-PEG-PSMA conjugate were described previously [29,34]. Following the gas exchange process, the vial was agitated using a Vialmix® device (Lantheus Medical Imaging, Canada) for 45 s, inverted, and centrifuged at 50 RCF for 5 min. 400 μL of the resultant Plain-NBs or PSMA-NBs were collected from each vial, and multiple vials were combined to achieve the required volume for imaging studies. A flowchart summarizing the NB synthesis process is provided in Supplementary Fig. S1.

Characterization using resonant mass measurement (RMM) and dynamic light scattering (DLS) demonstrated that PSMA-NBs have a hydrodynamic diameter of 266–288 nm, a concentration of 3.9 × 10¹¹ ± 2.82 × 10¹⁰ NBs/mL, and a slightly negative zeta potential, which ensures stability. For comparison, untargeted NBs were shown to have an average diameter of 274 ± 8 nm with a concentration of 4.07 × 10¹¹ ± 3.15 × 10¹⁰ NBs/mL. Prior side-by-side characterization matched the two formulations in diameter (277 ± 11 vs 281 ± 2 nm), concentration (3.9 vs 4.0 × 10¹¹ NBs/mL), and zeta potential (−4.21 vs −2.15 mV), differing only by the DSPE-PSMA-1 conjugate [29]. Furthermore, in vitro cellular uptake experiments defined the optimal PSMA ligand density as incorporation of 25 μg of PSMA-1 into the formulation, corresponding to approximately 35 × 10³ PSMA molecules per NB and yielding robust receptor-mediated targeting [15,29]. For comparative purposes, commercially available Lumason® MBs (Bracco, Italy) were reconstituted with 5 mL of 0.9% sodium chloride (saline) per vial and prepared according to the manufacturer's instructions[35].

2.2. Cell culture

PSMA-positive human PC3pip cells (human prostate cancer) were retrovirally transduced [36] and subsequently modified to express green fluorescent protein (GFP). The authenticity of the cell line was previously verified via Western blot analysis [34]. These cells were maintained in a humidified incubator at 37 °C and 5% CO₂ in RPMI 1640 medium (Gibco Life Technologies, Grand Island, NY), supplemented with 10% fetal bovine serum (FBS - Invitrogen Life Technology, Grand Island, NY) and 1% penicillin/streptomycin (Gibco Life Technologies, Grand Island, NY).

2.3. Tumor inoculation

All animal procedures were performed according to protocols approved by the IACUC, adhering to relevant guidelines and regulations. Immunosuppression was induced with daily subcutaneous injections of cyclosporine (10 mg/kg), starting one day prior to tumor cell injection and continued until the study concluded (Fig. 1A). Nine sexually mature male New Zealand White rabbits (Charles River Laboratories, Ashland, OH, United States) were utilized. Anesthesia was induced and maintained with 1–2% isoflurane throughout all procedures. Veterinary staff monitored animal health and tolerability throughout. Heart rate, respiratory rate, peripheral oxygen saturation (SpO₂), and body temperature were recorded at pre-anesthetic, intra-procedural, and post-procedural time points during each session; body weight and a systems-based clinical examination (cardiovascular, respiratory, gastrointestinal (GI), urogenital, neurological, integumentary, lymphatic, and head/eyes/ears/nose/throat) were performed recurrently. Contrast-agent identity and injection times came from the study imaging log; vital signs, examinations, and adverse events were collected from the institutional veterinary medical record. We define an imaging session as one anesthetic episode with one or more contrast agents administered; a scan is one analyzable region-of-interest (ROI) measurement, so a session could yield a tumor scan, a prostate scan, or both, depending on whether each structure was captured in the imaging plane.

Fig. 1.

Study protocol summary. (A) Schematic diagram of the experimental setup and study methodology. (B) Schematic image of the percutaneous tumor cells inoculation and relevant anatomical structures. (C1-2) Representative images of magnetic resonance imaging, (C3-4) B-mode ultrasound, (D1-2) H&E histology and (D3-4) prostate-specific membrane antigen (PSMA) immunohistochemistry (IHC) for intraprostatic tumors (C1, C3, D1, D3) and extraprostatic tumors (C2, C4, D2, D4). Scale bars are shown on each imaging panel (schematic panels are not to scale).

For tumor inoculation, rabbits were placed in a supine position, and the lower abdomen was shaved and prepared using aseptic techniques. After shaving, additional hair was removed with potassium thioglycolate cream (Veet Industries, Warren, MI). The prostate gland was located using the US, and 8 × 10⁶ PC3pip-GFP cells suspended in PBS were injected into the prostate gland using a 50.8 mm, 21-gauge PrecisionGlide™ needle (BD, Franklin Lakes, NJ, United States). The needle was introduced percutaneously and advanced to the prostate while avoiding the iliac vessels using US guidance by an interventional radiologist (Fig. 1B). Once the needle tip reached the central portion of the prostate, 100 μL of the cell suspension was injected. Both the syringe and needle were primed with cells prior to injection[33].

2.4. B-mode ultrasound and MRI

Ultrasound imaging, including B-mode and Doppler, for guided tumor cell injection and subsequent weekly monitoring, was conducted using an Acuson S3000 US scanner (Siemens Healthineers, Germany) equipped with an 18 MHz (18L6) transducer. Trans-abdominal images were acquired. MRI scans were performed using a 3T Magnetom Vida scanner (Siemens Healthineers, Germany) with the following parameters: echo time (TE) = 85 ms, repetition time (TR) = 7000 ms, flip angle = 160°, bandwidth = 203 Hz/pixel, voxel size = 0.4 × 0.4 × 2.0 mm³, field of view (FOV) = 120 × 86.3 mm (sagittal) and 120 × 97.5 mm (transverse), base resolution = 320, 26 slices. A board-certified abdominal radiologist interpreted the MRI images to determine the tumor's location and size. Rabbits were anesthetized with 2% isoflurane during US procedures. For the MRI, rabbits were induced with 2% isoflurane and anesthetized with ketamine (10 mg/kg) intramuscularly 10 min before scanning. Additional ketamine (5 mg/kg) was injected intravenously under veterinary supervision as needed to maintain sedation.

2.5. Contrast enhanced ultrasound imaging

CEUS imaging, using the same US system described above, was performed once a week and employed the standard nonlinear contrast pulse sequencing (CPS) mode; all acquisition settings were held constant across agents and are summarized in Table 1. Plain-NBs, PSMA-NBs and MBs at a dose of 0.8 mL/kg were injected intravenously via an ear vein catheter. Each agent was administered at 0.8 mL/kg (3.2 mL in a 4 kg animal) as a slow intravenous injection over 30–40 s. Lumason® was reconstituted per the manufacturer's instructions.

Table 1.

CEUS acquisition settings.

System	Mode	f0	MI	DR	Gain	Focal depth	FR	Duration
Siemens Acuson S3000	CPS	8.0 MHz	0.13	55 dB	−5 dB	1.5 cm	1 s−1	10 min

Contrast-enhanced ultrasound (CEUS) acquisition settings used for all three agents (PSMA-NB, Plain-NB, and Lumason® MB). CPS, contrast pulse sequencing (nonlinear contrast mode); f₀, transmit (center) frequency; MI, mechanical index; DR, dynamic range; FR, frame rate. Imaging depth was adjusted per animal to encompass the prostate. Acquisition was continuous for 10 min, beginning with a 15-s pre-injection baseline.

The injected bubble volume was selected on the basis of preliminary in vivo optimization in this species and is consistent with the volume-dependence of tumor enhancement reported for this model, in which peak signal intensity increased monotonically with injected volume across the range tested without reaching a plateau [33]; the same weight-normalized volume was used for all three agents. Because intravascular MB and an extravasating, receptor-binding NB derive their tumor signal from different physical processes, no single normalization — by bubble number, gas volume, or scattering cross-section — renders them equivalent. The targeting comparison, by contrast, was fully matched: PSMA-NBs and Plain-NBs share size, concentration, zeta potential, shell composition, and administered volume, and differ only in the presence of the targeting ligand. CEUS imaging was conducted continuously for 10 min, with a 15-s background recording prior to bubble injection (Supplementary Video 1). Rabbits were alternately injected with PSMA-NBs and MBs (N = 6) or Plain-NB (N = 3) each week (Fig. 1A), with the order of injection reversed every week. A 20-min clearance period was maintained between the administration of each UCA (Fig. 1A).

2.6. Animal enrollment, exclusion criteria, and data acquisition

One animal was euthanized approximately four weeks after tumor inoculation due to GI complications related to cyclosporine, in compliance with IACUC and the Animal Welfare Act guidelines and therefore not included in the analysis. 14 tumors were successfully induced in the eight remaining enrolled animals. At each weekly session, we acquired contrast data from only one 2D plane. As a result, animals with multiple spatially separated tumors could not have all lesions imaged at every time point. In some scans, the prostate had to be excluded from the imaging plane to capture the tumor, resulting in scans that contained only tumor images. In total, 14 tumor scans and 9 prostate scans were performed with PSMA-NBs, 9 tumor scans and 9 prostate scans were conducted using MBs and 3 tumor scans were conducted using Plain-NBs. Plain-NBs, size- and shell-matched to PSMA-NBs but lacking the PSMA-1 ligand, were included in a subset of animals as a confirmatory control to separate molecular targeting from size-dependent extravasation; following the 3Rs principle, this comparison was restricted to few animals and is exploratory. Five imaging sessions used PSMA-NBs only (no MBs): in two, an MRI plus two contrast injections could not fit within one anesthesia session; in three, two animals were imaged the same day, allowing one agent each. Scans from the first two weeks were excluded from analysis due to the tumors being too small to be adequately visualized, which hindered proper probe positioning.

2.7. Tumor segmentation and time-intensity curve analysis

To minimize bias related to contrast enhancement patterns, regions of interest (ROIs) were defined on baseline B-mode images acquired prior to UCA injection, rather than on contrast-enhanced sequences. Because tumor conspicuity and margin delineation on B-mode ultrasound are limited, the baseline tumor ROI was treated as an approximate “tumor-containing” region rather than a definitive histologic boundary; accordingly, we incorporated a tumor-adjacent compartment to explicitly sample the interface between the contoured lesion and surrounding tissue (Fig. 2B–C). Two researchers independently delineated the tumor ROI on the baseline B-mode image using pre-specified criteria (hypoechoic and/or heterogeneous regions suspicious for tumor while avoiding clearly identifiable non-tumor structures, such as vessels, when possible). Inter-observer agreement was assessed, and discrepancies were resolved by consensus through joint review to generate a single final tumor ROI per acquisition.

Fig. 2.

PSMA-NB and MB CEUS output and tumor segmentation strategy. (A) Representative images of B-Mode and non-linear contrast mode in vivo of PSMA Nanobubbles (PSMA-NBs) and clinically approved MBs during a 10 min scan in the same tumor of the same animal. A scale bar is shown on the B-mode and baseline (pre-contrast) frames; the contrast-enhanced frames at 1, 2, 4, 6, and 10 min are displayed at the same scale as the baseline frame. (B-C) Example region of interest (ROI) for the tumor core (red), tumor rim (green) and periphery tissues surrounding the tumor (blue) and their respective time-intensity curves (TICs) for (B) PSMA-NBs and (C) MBs.

In order to identify biologically significant spatial heterogeneity while maintaining a reproducible and non-contrast-guided approach, the consensus tumor ROI was subdivided using an automated, geometry-based MATLAB algorithm. From the consensus tumor boundary, the ROI perimeter was contracted inward by morphological erosion of the binary mask until the interior held 50% of the total ROI area. This interior was the tumor core, and the outer annulus between the original and contracted boundaries the tumor rim, giving two equal-area compartments per tumor. This approach is consistent with established gradients in perfusion and viability that are typically seen in solid tumors, wherein the outer rim is likely to harbor viable and relatively well-perfused tissue, and the core is likely to harbor nonviable tissue due to lack of vascular supply[[37], [38], [39], [40], [41]].[37], [38], [39], [40], [41] [[37], [38], [39], [40], [41]]

Moreover, an ROI for the peritumoral area was created automatically as a band outside the tumor ROI, with a cross-sectional area equal to 20% of the area of the tumor ROI and not overlapping with it. In recognition of the degree of uncertainty in boundary definition based solely on B-mode characteristics, an ROI was created to sample tissue immediately adjacent to the delineated tumor, potentially including a mix of benign prostatic tissue, inflammatory cells, and microscopic tumor[42,43].

A specific ROI, representing the non-tumor prostate tissue, was manually defined in an area with normal anatomical appearance when possible, avoiding the presence of large vessels and imaging artifacts. All ROIs (tumor core, tumor rim, surrounding/peritumoral tissue, and prostate reference) were then mapped onto the corresponding nonlinear contrast (NLC) cine acquisition obtained during UCA injection within the same imaging session. Time-intensity curves (TICs) were generated by calculating the mean NLC signal intensity within each ROI over time, and quantitative TIC parameters were derived from these curves for subsequent comparisons across ROIs and experimental conditions.

For each ROI, TIC parameters were computed from the ROI-mean nonlinear-contrast signal (illustrated in Fig. 4; the parameter framework was adapted from Gu et al. [44]). “Wash-in” was defined as the first frame at which the ROI-mean signal rose more than 30% above the pre-injection baseline. Peak intensity was the maximum ROI-mean signal, and time to peak (TTP) the interval from wash-in to that maximum. Area under the curve (AUC) was the integral of the signal over the acquisition; area under the wash-out curve (AUC_wo) the integral from the peak to the end of the acquisition; and area under the rising curve (AUC_r) the complementary rising-phase integral (AUC − AUC_wo). The end-frame average signal was the mean background-subtracted signal over the final frames. MTT was defined operationally as the interval between wash-in and the point at which the descending (wash-out) signal first returned to 50% of its peak value; so defined, MTT is a retention-weighted residence time, shorter for rapidly cleared agents and longer for retained ones. All analysis was carried out using MATLAB® (The MathWorks Inc. Natick, Massachusetts. Version R2025b).

Fig. 4.

TIC of CEUS in different tumor regions. Time intensity curves (TIC) of the ultrasound non-linear contrast mode average signal in the tumor core (red), tumor rim (green) and periphery tissues surrounding the tumor (blue). (A) TIC for PSMA-NBs. (B) TIC for MBs. (C) Graphical representation of the TIC parameters (defined in Methods; framework adapted from Gu et al. [44]) (D) Parametric graphs for TIC. AUC: Area under the curve. AUC_r: Area under the rising curve. AUC_wo: Area under the wash-out curve. PSMA: Prostate-specific membrane antigen. Mean ± SEM; the PSMA-NB-versus-MB comparison uses n = 11 (PSMA-NB)/9 (MB) tumor scans from the five animals imaged with both agents. Region-level PSMA-NB-versus-MB differences were significant in all regions in the log-scale linear mixed-effects model with animal and tumor random effects (peak intensity all p ≤ 0.016; AUC all p < 0.0001).

2.8. Size stratification

To evaluate the influence of tumor size on CEUS parameters, tumors were stratified into two groups based on cross-sectional area: small tumors (≤0.55 cm²) and large tumors (>0.55 cm²; Table S1). This threshold was derived from biological and anatomical considerations. Specifically, the normal rabbit prostate has an approximate cross-sectional area of 5 cm²; thus, a 0.55 cm² tumor represents roughly 11% of the prostate cross-section. This proportion is consistent with early-stage prostate tumors in humans, which typically occupy approximately 4% of total prostate volume—equivalent to ∼11% of the cross-sectional area assuming a spherical geometry[45,46].

2.9. Multiparametric mapping and decorrelation time

Multiparametric maps of AUC, decorrelation time (DT), and peak signal intensity were also generated using a pixel-based approach (Fig. 6, Fig. 8C–D), as described by Wegierak et al. [47] DT is an ultrasound-derived biomarker that captures how long the contrast signal at a given pixel remains similar over time. In practice, higher DT indicates more persistent (slowly changing) contrast signal, consistent with reduced motion and/or prolonged local retention (e.g., extravasation, constrained transport within the tumor interstitium, or agent adherence), whereas lower DT reflects rapidly varying signal, consistent with faster-moving intravascular contrast or freely perfused tissue. DT maps were computed from the NLC image sequence on a pixel-by-pixel basis using the same thresholding approach described by Wegierak et al. [47] generating spatial maps that highlight intratumoral heterogeneity in contrast dynamics beyond intensity-only metrics. Analysis used the log-compressed grayscale NLC data: each pixel time–intensity curve was z-normalized, and its temporal autocorrelation was computed over the entire acquisition (not a sliding sub-window, so no inter-window overlap). DT was the lag time, obtained by converting the autocorrelation lag index with the acquisition frame rate, at which the normalized autocorrelation first fell to a threshold of 0.5; this was repeated at every pixel to form the map, with no spatial smoothing. Unlike the analysis shown in Wegierak et al., no image registration, respiratory gating, or motion compensation was applied to the current data set; residual physiological motion may therefore contribute to pixel-level decorrelation and is acknowledged as a limitation. To characterize intratumoral heterogeneity of the DT signal, the per-pixel DT values within each tumor zone (core, rim, surrounding) were compiled for each animal and agent and summarized as relative-frequency histograms and the within-zone coefficient of variation (Fig. S7).

Fig. 6.

Representative CEUS images and pixel-based parametric maps highlight intratumoral heterogeneity. (A) Large intraprostatic tumor (>0.55 cm²): baseline B-mode anatomy and peak-intensity CEUS frames for PSMA-NB (left) and Lumason® MB (right), with tumor ROI overlay, alongside corresponding pixel-based maps of peak intensity and AUC for each agent. (B) Small intraprostatic tumor (≤0.55 cm²): analogous B-mode images, peak-intensity CEUS frames, and peak-intensity/AUC maps for PSMA-NB and MB. Multiparametric maps illustrate spatially heterogeneous enhancement patterns within and around the ultrasound-defined tumor region. Within each column, a scale bar is shown on the baseline B-mode image; the corresponding peak-intensity CEUS frame and the peak-intensity and AUC parametric maps are spatially co-registered with and displayed at the same scale as that B-mode image.

Fig. 8.

CEUS decorrelation-time mapping and PSMA immunohistochemistry across tumor sizes. (A) Representative whole-section PSMA immunohistochemistry (IHC) from a large tumor (>0.55 cm²), with pixel-based classification into PSMA-positive and PSMA-negative regions (segmented into core and rim compartments). (B) Representative PSMA IHC from the single small tumor (≤0.55 cm²), with the same pixel-based classification. (C) Baseline B-mode images and corresponding decorrelation-time (DT) maps for PSMA-NB and MB from the representative large tumor in (A). (D) Baseline B-mode images and corresponding DT maps for PSMA-NB and MB from the representative small tumor in (B). In (C) and (D), a scale bar is shown on each baseline B-mode image; the DT maps are spatially co-registered with and displayed at the same scale as the B-mode image above them and therefore do not repeat the bar. In (A) and (B), the pixel-based PSMA classification is displayed at the same scale as its paired IHC section. An exploratory comparison of DT with PSMA positivity is provided in Fig. S6.

2.10. Histology and pathologic correlations

Five weeks post-tumor injection, rabbits were euthanized, and bladder and tumor tissues were harvested. Ex vivo imaging was performed using a Maestro In Vivo Imaging System (PerkinElmer, Waltham, MA, United States) with a blue filter (excitation 445–490 nm, emission filter 515 nm longpass) to detect GFP. Bladder, accessory sex glands, and tumor tissues were fixed in 10% formalin, embedded in paraffin, cut at 5 μm thickness, and stained with hematoxylin-eosin (H&E) following standard protocols. For immunohistochemistry (IHC), tissues were processed by the pathology core, with PSMA expression assessed using a monoclonal antibody (ab) against PSMA (ab19071, Ms mAb to PSMA, Abcam, Cambridge, MA, USA).

Slides were digitized using a Nanozoomer S60 scanner (Hamamatsu, Japan). Whole-slide images (WSI) were analyzed in QuPath (v0.5.1) using a Random Trees pixel-classification workflow [48]. For H&E WSIs, pixel classification was used to segment the tumor into viable tumor, necrosis, and stroma (Fig. 7). Using the same QuPath Random Trees pixel-classification approach, PSMA IHC WSIs were segmented into PSMA-positive and PSMA-negative regions to quantify spatial PSMA expression within the tumor (Fig. 8; correlation with DT in Fig. S6). All segmentation outputs were independently reviewed and validated by a veterinary pathologist.

Fig. 7.

Histopathologic validation of CEUS kinetic metrics across tumor sizes. CEUS–histology correlations were performed using the terminal (euthanasia) imaging session. (A1) Representative whole-section hematoxylin and eosin (H&E) slide from a large intraprostatic tumor (>0.55 cm²), with pixel-based classification into viable tumor, necrosis, and stroma. (A2) Correlation of CEUS-derived kinetic parameters with the percentage of viable tumor quantified from whole-section H&E segmentation in the large-tumor cohort: MTT (left) and AUC_wo (right), shown for PSMA-NB (top row) and Lumason® MB (bottom row). Each plot reports Spearman's rank correlation coefficient (ρ) and p value; correlations were computed on the large-tumor cohort (3 animals, 6 ROIs) and are interpreted as exploratory. (B1) Representative whole-section H&E slide from the single small tumor (≤0.55 cm²), with the same pixel-based classification. (B2) MTT and AUC_wo versus % viable tumor for the small tumor (colored symbol) overlaid on the large-tumor cohort (grey symbols) for PSMA-NB (top) and MB (bottom); in (B2) the coefficients were recomputed with the small tumor added to the large-tumor cohort (n = 8 ROIs), whereas the large-tumor-only coefficients (n = 6 ROIs) are those shown in (A2). In (A1) and (B1), the pixel-based tissue classification is derived from and displayed at the same scale as the paired whole-section H&E image. Large-tumor cohort n = 6 ROIs from 3 animals (small tumor (B) shown for reference); coefficients are Spearman's ρ, p as shown.

For regional histologic correlations with imaging, each tumor WSI was additionally partitioned into two equal-area compartments based on the histology-defined tumor boundary: an inner core and an outer rim, each comprising 50% of the total tumor area. A “surrounding tissue” compartment was not defined on histology because histopathology was treated as the reference standard for tumor extent and margin definition; therefore, regional analyses were restricted to core–rim compartmentalization within the tumor boundary. CEUS acquisition planes were referenced to fixed anatomical landmarks, and the prostate was sectioned to approximate the imaging plane, giving regional (core/rim) correspondence between the CEUS maps and histology; this correspondence is approximate rather than a precise slice-to-slice co-registration.

2.11. Statistical analysis

All statistical analyses were performed using Prism® (Version 10.4.1, GraphPad, La Jolla, CA). To compare the average tumor infiltration correlation parameters for PSMA-NBs versus Plain-NBs or MBs across the four regions, i.e., surrounding tissue, rim, core, and prostate, we used two-way analysis of variance (ANOVA) with a mixed-effects model. This statistical test was chosen to suitably address the repeated measures, since the same animals contributed data to more than one zone, and to address the missing data points, e.g., when visualization was not possible or when the data were limited to one structure. Geisser–Greenhouse correction was also employed to address the potential violation of sphericity, thereby satisfying the assumptions for the ANOVA test. To account for the repeated weekly sessions and the multiple tumors per animal, region-level peak intensity and AUC were modeled on the natural-log scale with a linear mixed-effects model (contrast agent × region fixed effects, imaging week as a covariate) with random intercepts for each animal and for each tumor nested within animal (R 4.5, lme 4); agent effects are reported as geometric-mean ratios with 95% confidence intervals (Satterthwaite degrees of freedom). Within each TIC parameter and region, raw p-values for pairwise agent comparisons were adjusted with the Benjamini–Hochberg procedure to control the false discovery rate at 5%; q-values are reported alongside raw p-values (Tables S2 and S3), and whether each comparison remained significant after correction is indicated in the text. Measurement repeatability was assessed with the intraclass correlation coefficient (ICC) computed from the mixed-effects variance components, using the repeated weekly acquisitions of the same tumors.

For post hoc pairwise comparisons, the Tukey method was used to control family-wise error rates while maintaining sufficient statistical power, including direct comparisons between PSMA-NBs and MBs within the same tumor zone. For the small exploratory Plain-NB control cohort (n = 3; Fig. 3), per-ROI comparisons between agents were assessed by paired t-tests across the three animals (a conservative per-animal repeated-measures analysis), and the pairwise comparisons among tumor zones within each agent were Tukey-adjusted for the family of three comparisons. Because the correlation cohorts were small (n = 3–8), used bounded percentage variables, and were expected to be monotonic rather than strictly linear, associations between imaging parameters and histologic readouts (percentage of viable tumor and percentage of PSMA-positive pixels) were assessed with Spearman's rank correlation (ρ). Because each animal contributed two ROIs to the pooled viability correlations, a repeated-measures correlation (rmcorr) with the animal as the clustering unit served as a sensitivity analysis; given the cohort size, all correlations are exploratory.

Fig. 3.

Comparison of PSMA-NB versus Plain-NB (untargeted) kinetics across tumor subregions. (A) Representative TICs from a single rabbit tumor, demonstrating prolonged signal persistence (increased half-life/MTT) of PSMA-NB compared with untargeted nanobubbles (Plain-NB) within the tumor core, rim, and peritumoral (surrounding tissues) ROI. (B) MTT for PSMA-NB versus Plain-NB in each ROI across the three animals imaged with both agents. (C) AUC_wo for PSMA-NB versus Plain-NB across the same ROIs and animals. In panels (B2) and (C2), the between-agent comparisons (PSMA-NB versus Plain-NB) are per-animal paired t-tests across the three animals, and the within-agent comparisons among tumor zones are the same paired analysis with Tukey adjustment for the three pairwise zone comparisons. MTT was significantly longer for PSMA-NB in the rim (p = 0.009) and surrounding tissue (p = 0.022); the AUCwo differences were directionally consistent across animals but did not reach significance between agents in any region (all p > 0.30) and are therefore interpreted descriptively only.

All values are presented as mean ± standard error of the mean (SEM). Statistical significance was set at p < 0.05. Significance is denoted *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

3. Results

3.1. PSMA-NB versus untargeted NB

To assess the specificity of molecular targeting with PSMA-NBs, a set of studies was performed in which a distinct cohort of animals was imaged with PSMA-NB and Plain-NB during a single imaging session. Since one of the main goals of ligand-based targeting is to improve molecular specificity and prolong the intratumoral retention of the therapeutic agent, the retention-sensitive TIC parameters were given greater emphasis: MTT and AUC_wo.

When compared to the Plain-NB formulation, the PSMA-NB formulation demonstrated a 21.0% longer MTT value overall across all tumor regions. However, the greatest differences were seen in the rim and surrounding tissue ROI. MTT was significantly longer for PSMA-NBs in the rim (p = 0.009) and surrounding tissue (p = 0.022), with a consistent but non-significant difference in the core (p = 0.186) (Fig. 3). The wash-out behavior of the formulations was also evaluated by measuring the AUC_wo. The PSMA-NB formulation was observed to have a greater AUC_wo value compared to the Plain-NB formulation. The greatest differences were seen in the tumor core and rim ROI, with 34.7% (SEM: 25.2%) and 25.0% (SEM: 17.0%) greater AUC_wo, respectively, and a smaller difference in the surrounding tissue (9.8%, SEM: 11.3%). All differences did not reach significance (all p > 0.30).

3.2. TIC analyses in tumor zones with comparative evaluation of PSMA-NBs and MBs

In the PSMA-NB and MB cohort, PSMA-NBs consistently demonstrated higher signal intensity across all tumor zones when compared to MBs. Specifically, PSMA-NBs exhibited approximately 1.60-fold and 1.50-fold higher peak signal intensity than MBs in the tumor core and rim, respectively (descriptive mean ratios; the formal agent comparison is given by the mixed-effects model below). Additionally, PSMA-NBs showed significantly longer MTTs than those of MBs in both the tumor core (4.20-fold; p < 0.0001) and rim (4.50-fold; p < 0.0001). A similar effect was observed in the tumor surrounding tissues, where PSMA-NBs displayed a 1.60-fold increase in peak intensity and a 5.40-fold increase in MTT (p < 0.0001) compared to MBs (Fig. 4). Further analysis of TIC parameters revealed that PSMA-NBs exhibited longer MTTs (indicative of extended tissue retention), particularly in the tumor surrounding tissues and rim. In contrast, the tumor core displayed a TIC profile more similar to prostate tissue. These region-level differences were confirmed in the mixed-effects model with animal and tumor-within-animal as random effects across sessions: contrast agent remained highly significant for both peak intensity (F₁,₆₄ = 32.6, p < 10⁻⁶) and AUC (F₁,₆₅ = 139, p < 10⁻¹⁵), with PSMA-NBs exceeding MBs in every region (peak geometric-mean ratios 2.0–2.9×, all p ≤ 0.016; AUC 6.2–8.7×, all p < 0.0001) and no significant agent × region interaction (p = 0.73). After Benjamini–Hochberg correction across the family of PSMA-NB-versus-microbubble regional comparisons in peak intensity, AUC, and MTT, all principal differences remained significant (all q ≤ 0.016; Table S2). Wash-out area (AUC_wo) was treated as a descriptive parameter and was not included in the multiplicity-controlled families. Within the smaller, exploratory PSMA-NB-versus-Plain-NB comparison, the MTT differences in the rim and surrounding tissue also remained significant after correction (q = 0.027 and 0.033), whereas the core comparison did not (q = 0.186; Table S3).

The AUC was significantly higher for PSMA-NBs than for MBs, with the greatest differences observed in the tumor rim (440 ± 80 vs. 60 ± 10 a. u., p < 0.001) and surrounding tissues (580 ± 90 vs. 80 ± 10 a. u., p < 0.001) (Fig. 4D). At the conclusion of the imaging period (end-frame signal), PSMA-NBs remained detectable in the tumor rim (6.0 ± 3.0 a. u.) and surrounding tissues (12.0 ± 5.0 a. u.), whereas no signal from MBs was observed in any tumor zone.

When comparing tumor zones, PSMA-NBs demonstrated differentiation of surrounding tissue from prostate tissue. In the tumor surrounding tissues, PSMA-NBs exhibited a significantly higher AUC (580 ± 90 vs. 250 ± 100 a. u., p = 0.003) (Fig. 4D). No significant difference in AUC was observed between the tumor rim and the prostate (440 ± 80 vs. 250 ± 100 a. u., p = 0.157) or the tumor core and prostate (250 ± 60 vs. 250 ± 100 a. u., p = 0.999). MBs did not show any significant differences in AUC between surrounding tissues and prostate.

PSMA-NBs also showed higher AUC_wo values in the tumor surrounding tissues (470 ± 80 a. u., p = 0.015) compared to the prostate (210 ± 90 a. u.) whereas the difference between the tumor rim and prostate was not significant (p = 0.243) (Fig. 4D). Similarly, no significant difference was observed in AUC_wo between the tumor core and prostate (p = 0.904). MBs also did not exhibit significant zone-to-prostate differences for AUC_wo in the surrounding tissues (53.0 ± 12.0 a. u. vs. 16.0 ± 6.0 a. u., p = 0.999), tumor rim (47.0 ± 11.0 a. u. vs. 16.0 ± 6.0 a. u., p = 0.997) or tumor core (24.0 ± 8.0 a. u. vs. 16.0 ± 6.0 a. u., p = 0.948).

At the end of the imaging period (10 min), PSMA-NBs showed higher signal levels in the tumor surrounding tissues compared to the prostate (12.0 ± 5.0 vs. 2.0 ± 1.0 a. u., p = 0.011) (Fig. 4D). However, no significant differences were observed for the tumor rim (6.0 ± 3.0 a. u., p = 0.449) or tumor core (3.0 ± 1.0 a. u., p = 0.999). In contrast, MBs showed no detectable signal in any of the tumor zones, surrounding tissue, or prostate by the end of the 10-min imaging period.

3.3. Tumor size stratification and pharmacokinetic variations

The impact of tumor size on the pharmacokinetics of the UCA was investigated by comparing small tumors defined by an area of ≤0.55 cm² (n = 5) and large tumors defined by an area >0.55 cm² (n = 6). The cut-off was set a priori (Table S1). For all small and large tumors, the PSMA-NBs demonstrated a distinct TIC pattern compared to the MBs, with increased peak signal and retention (longer MTT).

For small tumors, the signal was homogeneous for the different ROI for each type of UCA, with the core, rim, and surrounding tissue ROI for each type of UCA showing similar TIC patterns, and there were no differences between the different regions for the PSMA-NBs or the MBs (Fig. 5A1-B1). When the PSMA-NBs were compared to the MBs for the small tumors, the PSMA-NBs maintained the pattern seen for the entire group.

Fig. 5.

Time-intensity curves (TIC) of the ultrasound non-linear contrast mode average signal in the tumor core (red), tumor rim (green), and periphery tissues surrounding the tumor (blue). TIC for PSMA-NBs in (A1) tumors smaller than 0.55 cm² and (A2) larger than 0.55 cm2 (B1) TIC for MBs in tumors smaller than 0.55 cm² and (B2) larger than 0.55 cm². Data are mean ± SEM; n = 5 small and n = 6 large tumors, from the five animals imaged with both PSMA-NB and MB. Within-agent zone comparisons used the two-way mixed-effects ANOVA with Geisser–Greenhouse correction and Tukey post hoc test (see Methods); in large tumors, PSMA-NB MTT differed by zone (core vs surrounding p < 0.001; core vs rim p = 0.005).

In larger tumors, the kinetics of UCA become non-uniform, indicating the onset of biological zoning. With PSMA-NBs, the TICs in the regions of interest demonstrate a trend where the maximum MTT is in the surrounding tissue (4.60 ± 0.60 min), followed by the rim (3.70 ± 0.50 min), and the minimum in the core (2.10 ± 0.40 min) (core vs surrounding tissue p < 0.001; core vs rim p = 0.005; Tukey-corrected) (Fig. 5A2). With contrast-enhanced MBs, the contrast washes out rapidly in all areas, with smaller differences in MTT among the ROIs: the rim and core values were similar (0.61 ± 0.15 min and 0.53 ± 0.14 min, p = 0.19), whereas the surrounding tissue differed from the tumor core (p = 0.025; Tukey-corrected). These data indicate that small tumors tend not to demonstrate zonal differences for both contrast agents, while in larger tumors, biological zoning is best demonstrated with PSMA-NBs, maintaining high enhancement in the biologically active areas.

3.4. Correlation of contrast enhancement TIC parameters with tumor necrosis

We evaluated radiologic–pathologic correlations between CEUS-derived contrast enhancement metrics and histology-defined tumor viability, stratified by tumor size (Fig. 5A1–A2 and Fig. 7A1–A2). To ensure temporal alignment between imaging and pathology in this fast-growing model, only the final CEUS acquisitions performed immediately prior to euthanasia (PSMA-NB and MB) were used for correlation analyses.

By the terminal time point, only one animal remained in the small tumor category (≤0.55 cm²), whereas the remaining animals in the PSMA-NB/MB cohort had large tumors (>0.55 cm²). Accordingly, statistical correlations were evaluated for the large-tumor group, where TIC-derived parameters showed size-dependent associations with histologic viability. In large tumors, MTT correlated strongly with viability for both agents, with a stronger association for PSMA-NB (Spearman ρ = 1.00, p = 0.003) (within-animal r_rm = 0.95, p = 0.054; n = 6 ROIs from 3 animals; exploratory) and more weakly for MB on the same cohort (ρ = 0.49, p = 0.36), a pattern that is hypothesis-generating rather than evidence of targeting specificity, because PSMA-NBs and MBs differ in size as well as targeting, targeting is isolated only by the size-matched Plain-NB comparison, and the clustering-aware analysis did not reach significance. AUC_wo was only weakly associated with viability for both agents (PSMA-NB ρ = 0.37; MB ρ = −0.37; both p = 0.50) (Fig. 7A2; complete Spearman correlation matrices for all TIC parameters are provided in Fig. S2).

3.5. Pixel-based mapping and regional distribution of contrast agents

Finally, multiparametric maps were created using a pixel-based approach [47] (Figs. 6 and 8C-D), rather than relying solely on ROI-averaged signal metrics. These pixel-wise maps (peak intensity, AUC, and DT) highlighted spatial heterogeneity in contrast behavior and showed that PSMA-NBs exhibited higher signal intensity within histology-defined viable tumor regions (Fig. S3). To quantify this heterogeneity for the DT signal, we examined the distribution of per-pixel DT within each tumor zone for every animal (Fig. S7). DT was consistently longer for PSMA-NBs than for MBs and lengthened from the tumor core toward the surrounding tissue, while the within-zone pixel-level coefficient of variation was greatest in the core (PSMA-NB ∼126% in the core versus ∼78% in the surrounding tissue), identifying the core as the most heterogeneous compartment. These pixel-level distributions are exploratory and were not subjected to formal statistical testing. Notably, DT maps also suggested prolonged signal behavior extending beyond the ultrasound-defined tumor margin.

As an exploratory analysis, we examined how pixel-based DT related to PSMA expression from IHC (Fig. S6). Because histologic PSMA% is defined only within the tumor boundary and each tumor contributed few, non-independent ROIs, we did not treat DT–PSMA as a formal quantitative correlation: in the large-tumor cohort any positive association was non-specific, with the microbubble comparator matching or exceeding PSMA-NB (Fig. S6). DT is therefore better read qualitatively — pixel-based DT maps showed prolonged, spatially heterogeneous retention of PSMA-NBs at the viable tumor rim and periphery, including regions adjacent to the ultrasound-defined tumor boundary (Fig. 8C and D), and in the single small tumor DT was more than two-fold higher for PSMA-NB than for the microbubble. Only DT was compared with PSMA IHC in this exploratory analysis; peak intensity was not.

3.6. Safety and tolerability

Safety and tolerability were assessed in the five animals for which complete veterinary records were available; all five received both PSMA-NB and MB. PSMA-NB was administered at every imaging session contributing to the analysis, and MB at all but five of those sessions. In sessions in which both agents were given, MB was injected first and PSMA-NB approximately 20-30 min later, with each injection followed by at least 10 min of continuous imaging under veterinary monitoring. Thus each animal received, per weekly imaging session, one PSMA-NB injection and, in all but the five sessions noted, one MB injection (approximately two contrast administrations per session) across the repeated weekly series, although a per-injection total was not separately logged. Contrast-enhanced imaging sessions performed before a tumor was sonographically detectable were not included in the analysis, and no contrast agent was administered on the day of tumor inoculation. No adverse reaction attributable to PSMA-NB or MB administration was observed at any administration. Body temperature remained within, or marginally at the lower bound of, the normal rabbit range (100.8-102.4 °F) at all recorded sessions, and SpO₂, where recorded, was 99-100%; no desaturation was documented in any animal.

Cardiorespiratory findings were infrequent, isolated, and non-reproducible. One animal developed bradycardia with transient arrhythmia shortly after anesthetic induction that returned to normal sinus rhythm within minutes of reducing the isoflurane concentration, before any contrast was administered. One animal exhibited mildly increased respiratory effort under anesthesia that improved with repositioning of the ultrasound probe. One animal developed intra-procedural tachycardia that was managed with intravenous fluid support, with all other vital signs within normal limits, and that resolved at recovery. None of these events recurred during any subsequent imaging session in the same animal, and none began in temporal association with a contrast injection. A fourth cardiorespiratory finding, bradycardia on awake pre-anesthetic examination, was recorded in the animal excluded from the imaging analysis and is shown in Table 2 for completeness.

Table 2.

Safety and tolerability data.

Animal	ΔWa	GI eventsb	Cardioresp. Event			Outcome
			Event	Sessions affectedc	Recurrenced
1	−13.0%	Stasis/ileus	Tachycardiae	1	None	Completed; terminal endpoint
2	−27.0%	Stasis/ileus	↑ Respiratory effortf	1	None	Completed; terminal endpoint
3	−18.9%	Stasis/ileus	Bradycardia, arrhythmiag	1	None	Completed; terminal endpoint
4	NR	Stasis/ileus	None	0	-	Completed; terminal endpoint
5	NR	Stasis/ileus	None	0	-	Completed; terminal endpoint
6	−15.7%	Acute anorexia	Bradycardiah	1	-	Euthanized ∼4 wk; excluded

Per-animal safety and tolerability for the five animals receiving both PSMA-NB and MB with complete veterinary records. All animals received daily cyclosporine A (10 mg/kg) and underwent multiple weekly contrast-enhanced imaging sessions, each receiving one PSMA-NB injection and, in most sessions, one MB injection per session. No adverse reaction attributable to PSMA-NB or MB administration was observed in any animal at any administration. The excluded animal did not complete the imaging series and is shown for completeness.

^aRelative to body weight at inoculation, per veterinary record. NR, not recorded.

^bResolved with supportive care (subcutaneous fluids, prokinetics, vitamin B12).

^cNumber of imaging sessions in which the event occurred.

^dOccurrence in any subsequent imaging session in the same animal.

^eIntra-procedural; managed with IV fluids; resolved at recovery.

^fUnder anesthesia; resolved with probe repositioning.

^gPost-induction, transient; resolved on reducing isoflurane, before any contrast administration.

^hOn awake pre-anesthetic examination. GI, gastrointestinal; MB, microbubbles; NR, not recorded; PSMA-NB, PSMA-targeted nanobubbles.

GI stasis or ileus and progressive weight loss were observed across the cohort during the period of daily cyclosporine A immunosuppression. Serial body weights recorded by veterinary staff documented losses of 13.0%, 18.9%, and 27.0% from the time of tumor inoculation in three animals; body weight was not recorded with sufficient granularity to quantify change in the remaining two animals (Table 2). These findings were managed with supportive care, including subcutaneous fluid therapy, prokinetic treatment, and vitamin B12 supplementation, and did not preclude completion of the imaging series. One additional animal, enrolled but excluded from the imaging analysis, was euthanized following cyclosporine-associated anorexia and GI stasis.

4. Discussion

In this exploratory study, PSMA-NBs showed prolonged retention relative to conventional, clinically approved MBs at the doses used, and their kinetics tracked histology-defined tumor viability in a small correlative analysis, suggesting the potential to differentiate viable from necrotic tumor tissue and adjacent structures. Because PSMA-NBs (∼270 nm) and MBs (1–5 μm) differ markedly in size, the large PSMA-NB-versus-MB differences in retention and washout likely arise mainly from nanoscale extravasation through hyperpermeable tumor vasculature and interstitial persistence rather than targeting per se; PSMA targeting is instead isolated by the size-matched Plain-NB comparison, which differs only in the ligand [29]. PSMA-NBs also showed improved retention in PSMA-positive tumors relative to untargeted NBs, supporting a measurable contribution of ligand-mediated targeting rather than passive NB behavior alone. PSMA-NBs demonstrated superior signal retention and prolonged presence within the tumor, particularly in the tumor rim and surrounding tissues (e.g., tumor invasion, adjacent fat, vas deferens or fibrous tissue), compared to clinically approved MBs and untargeted NBs. Supplementary Video 1 shows the real-time appearance of PSMA-NB and MB enhancement across the full acquisition in a representative animal. Importantly, in an exploratory analysis these enhanced imaging capabilities tracked the percentage of viable tumor tissue, underscoring their potential to enable more accurate diagnosis and targeted therapeutic applications. The enhanced signal is driven primarily by nanoscale extravasation and interstitial retention, with a contribution from cell-specific molecular targeting, a high NB concentration per imaging voxel, and strong nonlinear oscillations of PSMA-NBs under insonation, supported by their flexible phospholipid shell, which may enable large volumetric deformation and prolonged acoustic response [49,50]. Multiparametric imaging further indicated that PSMA-NBs preferentially accumulate in viable tumor zones and exhibit prolonged retention; decorrelation-time patterns at the tumor periphery were spatially consistent with, but not specific to, regions of PSMA expression. These findings support the diagnostic potential of PSMA-NBs in advancing prostate cancer management.

The 10-min continuous acquisition was a research protocol to characterize the full retention kinetics, not a clinical requirement: early-phase peak intensity already discriminated PSMA-NBs from MBs within the first minutes at the doses used (Fig. 4), and the retention divergence is captured by a brief delayed acquisition. This supports real-time, intra-procedural molecular confirmation during targeted biopsy or focal therapy, a capability that, unlike PET/MRI fusion or micro-ultrasound, could in principle provide molecular information during rather than before the intervention, although this was not evaluated here, where molecular information is acquired before, not during, the intervention[51].

Significantly, in contrast to the earlier studies performed in mice, the present study extends the evaluation of PSMA-NBs to an orthotopic large-animal prostate tumor model and includes an untargeted NB control group (Plain-NB). CEUS was previously reported in this model only as a brief, untargeted feasibility and dose-selection study quantified by peak intensity alone [33]. To our knowledge, no molecularly targeted ultrasound contrast agent has been evaluated in an orthotopic large-animal prostate cancer model, nor has targeted NB retention in such a model been quantified by parametric mapping and correlated with whole-slide histology, a translational gap recently noted for the NB CEUS literature [52]. Such a study design will allow for a more rigorous evaluation of whether the PSMA-NB performance will translate to a setting with clinically relevant anatomy, depth of image acquisition, and peritumoral complexity (Fig. 3, Fig. 5). Consistent with extensive prior characterization of this formulation—including matched physicochemical properties between targeted and untargeted NBs, validated ligand conjugation, receptor-specific binding in vitro, and improved in vivo tumor retention and accumulation—our rabbit data similarly demonstrate enhanced persistence and retention-sensitive kinetics for PSMA-NBs relative to Plain-NBs, consistent with a contribution of active molecular targeting beyond passive accumulation, although the small size-matched Plain-NB cohort (n = 3) makes this exploratory rather than confirmatory [26,28,29,53]. Although similar results have been observed in murine models, which include orthotopic [29] and flank models [26], these systems have inherent translational constraints due to important differences in prostate histology/anatomy and ultrasound appearance relative to humans.

In contrast, the rabbit prostate more closely mirrors human anatomy and ultrasound characteristics (e.g., baseline echogenicity and spatial relationships to adjacent structures) [32], making it a more relevant intermediate model for assessing tumor conspicuity and peritumoral behavior. This also distinguishes the present work from other PSMA-directed nanomaterial imaging platforms, including PSMA-targeting molecularly imprinted nanogels [54], which to date have been evaluated exclusively in murine models.

Building on the translational relevance of this orthotopic large-animal model, we observed that PSMA-NBs exhibited significantly longer retention within prostate tissue than clinically approved MBs. The nonlinear-contrast signal followed a multiphase course: wash-in, wash-out, and, for the NB agents, a prolonged retention phase. MB signal washed out rapidly and was undetectable by the end of acquisition, whereas PSMA-NB signal persisted throughout the 10-min scan, an extended, clinically useable imaging window rather than a transient or unstable signal. This prolonged persistence suggests potential utility beyond tumor conspicuity, particularly for applications that rely on sustained parenchymal enhancement—such as monitoring tissue response after ablative therapies (e.g., high-intensity focused ultrasound [HIFU]), where MBs are currently used for post-procedural assessment [55]. In this context, the extended prostate signal provided by PSMA-NBs could improve sensitivity for detecting treatment-induced tissue destruction and for tracking post-therapy evolution over time. Retained, immobilized NBs also remain within the insonated field for the entire acquisition, whereas freely circulating agent flows into and out of the beam. Bound bubbles therefore accumulate acoustic exposure across the window, and we hypothesize that the accelerated late-phase decay of PSMA-NB signal reflects, in part, progressive acoustic destruction of the retained population rather than intrinsic instability.

Interpretation of enhancement in the “surrounding tissue” regions, however, should be made in the context of a key limitation of anatomic ultrasound: B-mode imaging alone cannot precisely define tumor boundaries, and regions labeled as peritumoral may partially overlap with the biologic tumor rim or include microscopic tumor extension not apparent on grayscale imaging [42,43,56]. This ambiguity is particularly relevant in prostate cancer, where chronic inflammation in non-tumoral prostate—frequently associated with malignancy—has been linked to increased odds of cancer and is more common in higher-grade disease [57], potentially increasing vascular permeability and favoring retention of nanoscale agents. In addition, the rapid tumor growth observed in our model (Fig. S4) likely imposed mechanical compression and altered regional perfusion, further influencing contrast-agent distribution and washout. Larger lesions would also develop elevated interstitial fluid pressure, which opposes convective transport of nanoscale agents into the interstitium, along with greater heterogeneity and, in hypoxic necrotic regions, reduced viable-cell density and correspondingly lower total PSMA availability. Together these would limit both extravasation and receptor-mediated retention of PSMA-NBs in the core, a plausible account of the shorter core MTTs in the large-tumor subgroup[58].

Consistent with the segmentation and biologic ambiguity of the peritumoral compartment discussed above, we also observed pronounced spatial heterogeneity in contrast enhancement across tumor regions, with signal increasing from the tumor core toward the rim (Supplementary Video 1). This is consistent with established models of tumor growth in which the rate of growth outstrips the capacity of the vascular supply to meet the metabolic demands of the tumor cells, resulting in central hypoxia, nutrient deprivation, and subsequent areas of necrosis. Thus, there is decreased perfusion-dependent contrast signal in the tumor core compared to the rim [59,60]. Because the core and rim compartments are defined purely geometrically and interrogated with the same contrast agent, the core-to-rim gradient reflects regional differences in perfusion and viability rather than molecular targeting. The model also explains the observed phenomenon of less clear delineation of core, rim, and other regions of interest in smaller tumors because they do not develop clear gradients of necrosis or clear viability in the core compared to the rim.

Notably, the greatest enhancement is observed in the tissue immediately adjacent to the contoured tumor, which is likely related to peritumoral inflammation and possibly microscopic tumor invasion that is difficult to distinguish using B-mode guidance alone. The association of inflammation/reactive stroma in the peritumoral tissues with aggressive tumor biology and poor outcomes has also been established [[61], [62], [63], [64]]. Thus, the ability of PSMA-NBs to enhance these tissues could potentially be of value in identifying tumors that have more aggressive invasive potential. These observations support the preferential uptake of PSMA-NBs in the biologically active regions of the tumors and their rims, which is consistent with the potential of these agents to enable characterization of tumors using CEUS [65]. This is in line with the observed patterns of regional enhancements that have already been described in the above sections. The sustained presence of the PSMA-NBs in the viable tumor tissues as well as in the margin-associated regions is likely to be beneficial in improving the characterization of prostate cancer tissues using molecularly targeted CEUS.

The sustained presence of the PSMA-NBs in the tissues could also have implications in the context of theranostic approaches that require sufficient residence times to enable the ultrasound-mediated delivery of the therapeutic agents [66,67]. Notably, the sustained enhancements of the PSMA-NBs in the prostate tissues while concurrently clearing over time is consistent with retention of the targeted agents, though this pattern alone cannot separate specific targeting from nanoscale extravasation and retention. The results of earlier preclinical studies on the PSMA-NB formulations have provided support to the translation of the agents to the human context. These studies have demonstrated favorable stabilities of the agents in physiological media as well as biodistribution patterns that are consistent with the reticuloendothelial system-mediated clearance of the agents. These patterns have included the uptake of the agents by splenic macrophages [68,69]. The perfluoropropane gas core is cleared by pulmonary excretion and the lipid-shell components through this reticuloendothelial (hepatosplenic) route; organ-level biodistribution was not assessed here and remains an important direction for future work. Notably, a preliminary toxicology report in Sprague-Dawley rats indicated no overt adverse effects of the agents after administration of a high dose of 70 mg/kg, with only mild non-adverse changes in hematologic parameters and spleen tissue consistent with agent clearance[70].

The transient rise in whole-blood echogenicity reported for these agents has been attributed partly to their association with red blood cell surfaces rather than intrinsic bubble stability, though other mechanisms may contribute; PSMA-NB stability in serum at 37 °C was not measured here [68]. Size and concentration were quantified by resonant mass measurement and dynamic light scattering (methods well suited to sub-micron gas-core bubbles); detailed morphology was not re-examined here. High-resolution morphology has been established previously by cryo-electron microscopy, which resolves a spherical gas core in a lipid monolayer shell and confirms the sub-300 nm architecture (Fig. S5) [15,71]. Conventional ambient-temperature transmission and scanning electron microscopy are poorly suited to these agents, whose gas core is susceptible to dehydration-, vacuum-, and stain-induced collapse; cryo-EM and resonant mass measurement are therefore the preferred tools for shell-stabilized NBs [16]. Finally, because the ligand is incorporated modularly, the platform could in principle be redirected to other PSMA-expressing malignancies; however, PSMA in most non-prostatic tumors is expressed mainly on the neovascular endothelium rather than the malignant epithelium, so the accessible target would be intravascular rather than the extravascular, epithelial PSMA engaged here, and the contrast kinetics would likely differ[72].

The GI morbidity and weight loss in this cohort are most consistent with the daily cyclosporine A immunosuppression required to support human tumor xenografts, rather than with contrast-agent administration. Cyclosporine produces a dose-dependent, species-specific GI toxicity in rabbits, in which weight loss varies as a linear function of cyclosporine dose [73], and dose-limiting cyclosporine-associated morbidity has been reported in rabbit human-tumor xenograft models using the same daily 10 mg/kg regimen employed here [74]; rabbits are additionally predisposed to GI stasis as hindgut fermenters under systemic illness or drug exposure [75]. Weight loss did not scale with contrast exposure: two animals with comparable exposure differed more than twofold (13.0% versus 27.0%). No adverse reaction was temporally associated with any contrast administration.

Several limitations should be noted. First, the study was small (N = 9) and not powered a priori, with 14 tumor scans from 8 animals treated as non-independent (animal as the unit of analysis) and an exploratory n = 3 Plain-NB arm. Second, imaging was trans-abdominal 2D, limiting consistent plane matching and imaging–histology correspondence. Third, molecular specificity of the PSMA-NB platform was previously established [27,[76], [77], [78], [79]], thus the present study characterized in vivo contrast and retention kinetics rather than repeating these controls [27]. Moreover, the PC3pip xenografts grow rapidly and are vascularly hyperpermeable (an enhanced permeability and retention [EPR]-like effect [17,80]). Notably, the contrast enhancement and retention behavior of this NB platform has been observed in vivo across four ultrasound platforms from three manufacturers (Toshiba AplioXG SSA-790A [28], the VisualSonics Vevo 2100 [81] and Vevo 3100 [47,65], and the Siemens Acuson S3000 used here), confirming that this behavior is not platform-specific. Decorrelation-time reflects temporal signal stability and is therefore motion-sensitive; at the 1 Hz frame rate used, no respiratory gating or registration was applied, so motion may contribute to the measured values, with gating or registration a clear next step. Finally, safety monitoring was clinical rather than a prespecified toxicological endpoint: organ histopathology, hematology, serum biochemistry, immune profiling, serum thermal-stability time-courses, and a dedicated biodistribution study were beyond the scope of the current study. As such, the presented tolerability data are supportive rather than definitive, although ultrasound contrast-agent safety is, in general, well established [82,83].

In conclusion, this orthotopic large-animal study shows that PSMA-NBs preferentially enhance viable tumor and margin-associated regions and demonstrate prolonged retention relative to both clinically approved MBs and untargeted NBs. These findings support continued development of PSMA-NBs as a molecularly targeted CEUS approach for prostate cancer characterization and justify future studies using volumetric imaging and more clinically representative disease models to define diagnostic performance and evaluate theranostic applications.

Declaration of AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used Claude LLM (Anthropic, San Francisco, CA) in order to review grammar, cohesion and clarity of the text. After using this tool, the authors reviewed and edited the work and take full responsibility for the content of the publication.

CRediT authorship contribution statement

Felipe M. Berg: Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing – original draft. Eric C. Abenojar: Data curation, Formal analysis, Investigation, Methodology, Project administration, Validation, Visualization, Writing – original draft. Pinunta Nittayacharn: Investigation, Writing – review & editing. Nathan K. Hoggard: Formal analysis, Validation, Writing – review & editing. Sidhartha Tavri: Investigation, Methodology, Writing – review & editing. Jing Wang: Data curation, Investigation, Writing – review & editing. Felipe Matsunaga: Investigation, Methodology, Writing – review & editing. Michaela B. Cooley: Formal analysis, Writing – review & editing. Dana Wegierak: Software, Writing – review & editing. Xinning Wang: Methodology, Resources, Writing – review & editing. Thomas J. Rosol: Formal analysis, Validation, Writing – review & editing. James P. Basilion: Conceptualization, Funding acquisition, Methodology, Resources, Supervision, Writing – review & editing. Agata A. Exner: Conceptualization, Data curation, Formal analysis, Funding acquisition, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – original draft.

Ethics approval and consent to participate

All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) (approval number 2018-0022) and were conducted in accordance with the National Institutes of Health (NIH) guidelines and relevant institutional regulations for the care and use of laboratory animals.

Funding

We acknowledge funding and support for this study from the Case-Coulter Translational Research Partnership and Wallace H. Coulter Foundation (A.A.E and J.P.B). This work is also supported by the National Institutes of Health via the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under Award No. 5R01-EB025741 (A.A.E and J.P.B). Felipe M. Berg was supported by the Marcos Lottenberg and Marcus Wolosker International Fellowship for Physicians Scientist. Michaela B. Cooley was supported by the National Institute of General Medical Sciences (T32GM007250) and National Heart Lung and Blood Institute (F30HL160111).

Declaration of competing interest

The authors declare the following financial interests (e.g., any funding for the research project)/personal relationships (e.g., the author is an employee of a profitable company) which may be considered as potential competing interests: Agata A. Exner and James P. Basilion are the co-founders of Visano Theranostics. The remaining authors declare no competing interests. Siemens® Healthineers (Erlangen, Germany) provided the ultrasound system used in this study (Acuson S3000 with 18L6 transducer). The company had no involvement in study design, data analysis, or manuscript preparation. All authors had full control of the data and submission, and none are affiliated with Siemens® Healthineers.

Appendix A. Supplementary data

The following are the Supplementary data to this article: