Narrowing of Glioma Vascular Caliber Via Chronic VEGFR2 Blockade Improves the Uniformity of Focused Ultrasound-Mediated Small Molecule Drug Delivery

Abstract

Glioblastoma (GBM) is a devastating disease, with standard-of-care therapies still yielding dismal survival outcomes. GBM cells are protected by the blood-brain and blood-tumor barriers, which severely limit therapeutic agent delivery from the bloodstream. Focused ultrasound (FUS), in combination with microbubbles (MBs), addresses this challenge by enhancing drug delivery. However, dilated and tortuous brain tumor vasculature disrupts MB flux and oscillation, which may limit FUS-mediated delivery. Here, we evaluated whether normalizing tumor vasculature via chronic neoadjuvant VEGFR2 inhibition (aVEGFR2, DC101) improves subsequent FUS-mediated small molecule drug delivery. After aVEGFR2 administration had pre-normalized GL261 glioma vasculature through reduced permeability and vascular caliber, T1 mapping MRI of FUS-delivered Multihance (MH) contrast agent, a model small molecule drug, yielded no change in total delivery. However, radiomic analysis of the T1 maps indicated that FUS-mediated model drug penetration into otherwise poorly accessible tumor regions was improved with aVEGFR2 pre-treatment. This improvement was accompanied by acoustic signatures suggestive of more stable MB oscillation. These results were then compared to those achieved with acute aVEGF pre-treatment, a regimen that copied the permeability reduction of chronic aVEGFR2 without reducing vascular caliber. This comparison identified reduced vascular caliber as the probable mechanism of improved delivery uniformity, perhaps acting through a shift in MB oscillation toward more stable regimes. Our results indicate that neoadjuvant aVEGFR2 cooperates with FUS-mediated small molecule drug delivery through a unique biophysical mechanism. This mechanism may be leveraged to further augment the efficacy of combination therapies against GBM that entail blocking VEGF signaling.

Graphical Abstract

Introduction

Glioblastoma (GBM) accounts for about 40–50%[1] of primary brain tumors diagnosed. Current standard-of-care therapy includes maximal safe resection followed by post-operative radiation therapy and temozolomide chemotherapy[2,3]. Despite this aggressive treatment regimen, the median survival time for patients is less than 16 months post diagnosis[4]. Virtually all patients experience tumor recurrence, resulting in a survival time of less than 1 year after recurrence, and overall, a 5-year mortality rate of over 90%[4]. This low survival outcome is attributed, at least in part, to the limited passage of most systemically administered therapeutics from the bloodstream to the tumor microenvironment.

At the capillary level, the “blood-brain barrier” (BBB) is comprised of, and supported by, cells of the neurovascular unit, including endothelial cells, pericytes, and astrocytes[5]. The BBB supports brain function and preserves brain homeostasis via protection of neural tissue from both endogenous and exogenous elements in blood[6]. Primary and metastatic brain tumors disrupt BBB structure and function throughout their progression, promoting pro-angiogenic cellular and molecular crosstalk [7,8]. A key mediator of aberrant tumor angiogenesis is the overexpression of vascular endothelial growth factor-A (VEGF)[9]. High neovascularization and upregulation of VEGF and one of its receptors, VEGF receptor 2 (VEGFR2), are associated with more aggressive, high-grade tumors[10,11]. VEGF anti-angiogenic bevacizumab (Avastin^®) is currently approved for recurrent GBM; however, it imparts limited, transient radiological responses without improving overall survival, emphasizing the need for improved treatment strategies[12,13].

Multiple obstacles to the successful treatment of GBM exist due to the combined influences of the BBB and its dysregulated tumor counterpart, coined the blood-tumor barrier (BTB) [9,14]. Tumor regions with intact BBB, resembling healthy brain vasculature, limit both diffusive and convective drug delivery[15,16]. Conversely, the dilated and leaky microvessels of the BTB yield both high tumor interstitial pressure and poor convective transport gradients[17]. Dysregulated vascular structure, connectivity, and tortuosity result in heterogeneous tumor perfusion, leading to variable levels of exposure to circulating therapies and contributing to therapeutic resistance and poor responses[17,18]. Further, highly infiltrative cells at the tumor margins, often the source of tumor recurrence, are difficult to target because they are protected from circulating therapies and not visible on conventional imaging modalities due to the BBB[12,19,20]. Low-intensity pulsed focused ultrasound with microbubbles (MBs), henceforth referred to simply as “FUS”, is a non-invasive technique to transiently open the BTB and facilitate the targeted delivery of systemically administered anti-cancer agents, such as chemotherapies and immunotherapies[21]. Pre-clinical and clinical applications employ a range of targeted FUS techniques, including magnetic resonance image (MRI)-guided, neuronavigation-guided, and implantable FUS technologies[22,23]. With these approaches, circulating MBs oscillate upon exposure to FUS and exert mechanical stresses on the surrounding microvasculature. This, in turn, opens the BBB/BTB for augmented payload delivery[21]. FUS is employed in the clinic for a wide range of brain pathologies, with 35 active or completed clinical trials employing BBB opening in brain tumors[24]. Of these trials, three aim to assess FUS to enhance bevacizumab delivery in recurrent GBM, with one reporting repeated treatment as safe and feasible[25].

Despite the current success of FUS for augmenting drug delivery to brain tumors, the approach depends upon the presence of oscillating MBs in the tumor vasculature. Aberrant tumor vasculature may diminish MB flux and energy transmission to vessel walls, which would limit the ultimate success of FUS-mediated drug delivery. To investigate how the dysregulated brain tumor vasculature impacts FUS, we pre-treated tumor vessels with low-dose anti-angiogenics that target the VEGF signaling cascade. The vascular normalization hypothesis suggests that low-dose anti-angiogenic therapy can temporarily restore the structure and function of tumor vasculature without ablating tumor vessels[26]. The strengths of this form of vascular modulation are that it can increase tumor vessel perfusion[27], decrease interstitial pressure to enhance convective transport gradients [28], and modify vascular caliber[29]. Notably, all of these changes elicited by vascular normalization have the potential to improve FUS-mediated delivery to brain tumors.

Here, we use two approaches to inhibit the VEGF signaling cascade to evaluate model drug delivery and distribution. The first approach uses a pre-treatment strategy with the monoclonal antibody DC101 (aVEGFR2), which blocks VEGFR2 and induces vascular modulation[30]. VEGFR2 is a promising target because it is overexpressed on tumor endothelial cells, reducing the need for deep tumor tissue penetration. The second approach uses a single dose of a monoclonal antibody that inhibits VEGF (aVEGF), akin to bevacizumab treatment. This strategy allows the examination of how changes in vascular permeability, but not vascular structure, affect therapeutic delivery. As VEGF is secreted by a wide variety of cells within the tumor microenvironment, it represents an important, though broad and widely distributed, target[31]. Overall, the goal of these studies was to use MRI to assess how vascular pre-treatment impacts small-molecule FUS-mediated drug delivery, which contrasts with many current efforts that are employing FUS to enhance the delivery of bevacizumab itself.

Moreover, while the evidence for enhanced delivery to brain tumors with FUS is abundant [21,32–34], few studies have examined how the uniformity of delivery with FUS can be increased[35,36]. Many studies focus on engineering drug carriers to better penetrate the tumor the microenvironment[37,38] and some focus on using FUS to enhance the delivery of anti-angiogenics [39], but none discuss how vascular modulation may be beneficial in enhancing the ability of MBs to interact with tumor endothelium for increased barrier opening. As such, the studies herein used this framework to test how normalizing tumor vasculature via modulation of VEGF signaling impacts subsequent FUS-mediated small-molecule drug delivery.

Results

Baseline Characterization of GL261 Tumor with VEGFR2 Inhibition

We first characterized how chronic VEGFR2 inhibition with low-dose DC101 alters baseline GL261 tumor vessel properties. Tumor-bearing mice were treated with 3 doses of 10mg/kg DC101 or IgG isotype control. MRI using MH contrast agent as a low molecular weight (MW) model drug was performed 6 days post-treatment initiation (Figure 1A). Reductions in permeability to MH due to VEGFR2 blockade were clear on T1 maps (Figure 1B). Baseline permeability of the tumor vessels to MH was reduced by ~34%, with the mean concentration of 0.047 mM ± 0.006 vs. 0.071 mM ± 0.005 for IgG control (Figure 1C, p = 0.0210). We then employed immunofluorescence (IF) to assess changes in vascular structure. VEGFR2 inhibition altered tumor vessel morphology in comparison to IgG-treated tumors (Figure 1D). The percent CD31+ tumor area significantly decreased in the DC101-treated group with 3.38 ± 0.37% vs 1.85 ± 0.25% area (p = 0.0117) for IgG and DC101, respectively (Figure 1E). This corresponded to a significant reduction in tumor vessel size, with mean sizes of 251.7 ± 22.37 μm² in the IgG group compared to 117.2 ± 17.25 μm² in the DC101 treatment group (p = 0.0104) (Figure 1F), but no difference in the number of vessels per tumor area, with 1.362×10⁻⁴ ± 1.33×10⁻⁵ and 1.550×10⁻⁴ ± 2.16×10⁻⁵ vessels per μm² (p = 0.4846) for IgG and DC101 groups, respectively (Figure 1G). Overall, this indicates that VEGFR2 blockade decreases vessel caliber without vessel pruning. Of note, in a separate cohort of animals, VEGFR2 blockade improved neither survival nor tumor control (Figure S1A–B). This was expected, as others report little to no tumor control at this dose of DC101.

Figure 1. Baseline characterization of GL261 brain tumors with chronic VEGFR2 inhibition.

A) Overview of treatment time course for T1-Mapping and arterial spin labeling (ASL). MRI window is marked with a red square. B) Representative images of baseline contrast enhanced (left) and corresponding concentration map (right) for IgG- and DC101-treated tumors 6 days post-treatment initiation. C) Corresponding baseline intratumor concentration of Multihance for IgG (n=3) and DC101 (n=5) treated tumors. D) Representative immunofluorescence images showing DAPI stained cell nuclei and CD31⁺ tumor vessels. Scale bars = 20 μm. E-G) Bar graphs of percent tumor area positive for CD31 (E), average CD31⁺ tumor vessel size (F), and vessel profiles per tumor area (G) for IgG- (n=5) and DC101-treated (n=5) tumors. H) Representative ASL blood flow maps for IgG (top) and DC101 (bottom) treated tumors. I-K) Bar graphs of intratumor blood flow (I), integrated blood flow (J), and maximum tumor blood flow over contralateral brain blood flow (K) for IgG-(n=5) and DC101-treated (n=5) tumors. Welch’s t tests (two-tailed). Means ± S.E.M.

We also employed ASL MRI in a separate cohort of GL261 tumor-bearing mice to test whether these changes in vessel structure affect tumor blood flow (Figure 1A). Qualitatively, tumor blood flow is far lower than flow in the surrounding brain parenchyma (Figure 1H). There was no difference between the average (Figure 1I, p = 0.6925) or integrated (Figure 1J, p = 0.8419) blood flow between the IgG and DC101 treated groups. When calculated as a fraction of maximum blood flow in normal contralateral brain tissue, maximum tumor blood flow was 0.697 and 0.644 for IgG- and DC101-treated mice, respectively (Figure 1K, p = 0.4983). Further, in a separate cohort of mice, there was no difference between DC101 and IgG control groups when examining the fraction of perfused vessels, as detected via lectin-perfusion at day 14 post-tumor implantation (Figure S2).

Impact of VEGFR2 Blockade Pre-Treatment on Low Molecular Weight FUS-mediated Drug Delivery

We next tested how pre-treatment with chronic VEGFR2 inhibition impacted FUS-mediated delivery of MH. As shown in Figure 1A, mice were treated with FUS the day after baseline imaging to enable paired post-FUS measurements. Representative post-FUS concentration maps for IgG and DC101-treated mice qualitatively show that, as expected, FUS increases the concentration of MH in both groups (Figure 2A). Quantification of these concentration maps suggests no change in overall MH delivery with FUS across the treated or control group, with mean MH concentrations of 0.157 ± 0.005 mM in IgG-treated tumors and 0.147 ± 0.023 mM in DC101-treated tumors (Figure 2B, p = 0.3421). Importantly, this suggests that the change in baseline vessel permeability did not limit FUS BTB opening or FUS-mediated low-molecular-weight (MW) drug delivery. In fact, there was a trend towards DC101-treated tumors having a higher fold-increase in total concentration 12 minutes after MH injection, with an average 2.24-fold vs 3.51-fold increase for IgG and DC101, respectively (Figure 2C, p = 0.1040).

Figure 2. Chronic VEGFR2 inhibition before FUS does not increase the magnitude of Multihance (MH) delivery.

A) Representative intratumor contrast enhanced images (left) and overlayed T1 concentration maps (right) 12 minutes after MH administration on day 15 post-tumor implantation (i.e., one day post-treatment termination). B) Bar graph of MH concentration over all tumor voxels at 12 minutes post MH injection for IgG- and DC101-treated tumors. C) Fold-change of day 15 post FUS concentration over day 14 baseline concentration (hatched bars) for IgG- (n=3) and DC101-treated (n=5) tumors. Welch’s t-tests (one-tailed). Means ± SEM.

Chronic VEGFR2 inhibition yields more stable MB oscillation

As we anticipated changes in the tumor vasculature to impact the MB dynamics during the FUS treatment, we also assessed acoustic emissions and how these emissions affected the passive cavitation detection (PCD) feedback control system. The FUS treatment scheme (Figure 3A) consisted of four sonication treatment spots placed over the tumor in the striatum. A PCD feedback control mechanism was used to tune the peak negative pressure (PNP) on a per-spot, per-mouse basis. The average pressures for both groups fell within the expected 0.2 MPa (starting pressure) and 0.4 MPa (maximum pressure) range. However, IgG-treated tumors had a lower average treatment PNP over FUS treatment time (Figure 3B), with a trend towards a decrease in final average PNP (Figure 3C, p = 0.0601), when compared to the VEGFR2 inhibition group. This decrease in average PNP was due to a significantly higher number of PCD feedback control loop thresholds in the IgG-treated group, with a total average of 16 thresholds met in comparison to an average of 5.2 met in the DC101-treated group (Figure 3D, p = 0.0067). The reduced number of thresholds in the DC101-treated group suggested that chronic VEGFR2 inhibition was diminishing subharmonic and ultraharmonic emissions, as they trigger the PCD feedback control system to reduce PNP. Indeed, when normalized to the average baseline sonication emissions for each target, the average fold-changes in subharmonic and ultraharmonic signatures were lower in the DC101-treated group (Figure 3E–G, p = 0.0087, p = 0368, p = 0509). There was no significant difference in the fold-change in broadband emission signatures (Figure 3H, p = 0.1055). We also found that DC101 increased the average 2^nd harmonic signature (p = 0.0343), but did not affect the 3^rd harmonic signature (p = 0.1856) (Figure 3I–J).

Figure 3. Chronic VEGFR2 inhibition before FUS toggles microbubble (MB) oscillation toward a more stable regime.

A) Schematic of treatment planning with four sonication focal spots overlayed over the tumor in the striatum. B) Average peak negative pressure (PNP) over treatment time. 20 baseline bursts are applied at 0.2 MPa to establish PCD thresholds. C) Bar graph of average final PNP. D) Enumeration of PCD thresholds met during FUS treatments. F-K) Bar graphs of fold-change differences in acoustic emissions signatures: E) subharmonic, F) 1^st ultraharmonic, G) 2^nd ultraharmonic, H) broad band noise over baseline emissions, I) 2^nd harmonic, and J) 3^rd harmonic. IgG, n=3; DC101, n=5). Welch’s t-test (one-tailed). Means ± SEM.

Radiomic feature analysis suggests VEGFR2 blockade improves uniformity of intratumor MH delivery

Examination of the T1 maps shown in Figure 2A suggested that DC101+FUS-treated tumors have relatively constant MH concentration throughout the tumor volume, as opposed to high concentrations around the tumor edge in IgG control-treated groups, indicating potential changes in the uniformity of model drug delivery (Figure 4A). Preliminary quantification of concentration distribution uniformity using skewness and kurtosis revealed that there is a trend towards DC101+FUS decreasing both metrics (Figure 4B–C). This is consistent with chronic VEGFR2 inhibition beneficially increasing the uniformity of low-MW model drug delivery compared to IgG controls.

Figure 4. Statistical and radiomic analyses indicate that chronic VEGFR2 blockade improves the uniformity of intratumor Multihance distribution after FUS.

A) Zoomed in representative post-FUS intratumor concentration maps shown in Figure 2. Scale bars = 1.75 mm. B, C) Bar graphs of intratumor concentration distribution skewness (B) and kurtosis (C). D-H) Radiomic features: GLDM Gray Level Non-Uniformity (D), GLRLM Run Length Non-Uniformity (E), GLSZM Zone Percentage (F), GLDM Small Dependence Emphasis (G), and GLSZM Large Area Emphasis (H). IgG, n=3; DC101, n=5. Welch’s t-tests (one-tailed). Means ± SEM.

To further assess how chronic VEGFR2 inhibition pre-treatment may improve MH distribution after FUS, we assessed intratumor radiomic features from the T1 concentration maps using PyRadiomics[40]. This process extracts features in 8 groups: first-order statistics, 2D and 3D shape-based features, gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zone (GLSZM) matrix, neighboring gray tone difference matrix (NGTDM), and gray level dependence matrix (GLDM). Of the radiomic features extracted, 5 were different between IgG- and DC101-treated tumors: GLSZM large area emphasis (LAE), GLDM gray level non-uniformity (GLN), GLRLM run length non-uniformity (RLN), GLSZM zone percentage (ZP), and GLDM small dependence emphasis (SDE) (Table 1). Chronic VEGFR2 inhibition before FUS reduced GLN (Figure 4D, p = 0.0501) and RLN (Figure 4E, p = 0.0235), indicating more homogenous gray level intensities and run lengths throughout the image. Additionally, treatment increased both ZP (Figure 4F, p = 0.0149) and SDE (Figure 4G, p = 0.0243) and decreased LAE (Figure 4H, p = 0.0163), indicating smaller, connected zones and an overall finer image texture.

Table 1:

Overview of Radiomic Features.

Feature	Description	Change with aVEGF+FUS	Interpretation
Skewness	Measure of the ‘skew’ or symmetry of a distribution about its mean.	↓ (p= 0.0787)	Lower skewness suggests the data has less asymmetry, or fewer extreme values in aVEGFR2 treated tumors.
Kurtosis	Measure of the ‘tailedness’ of a distribution about the mean in comparison to a normal distribution.	↓ (p= 0.0838)	Lower kurtosis suggests the data has fewer extreme, or outlier, values in aVEGFR2 treated tumors.
GLDM Gray Level Non-Uniformity (GLN)	Measure of the intensity of the gray level values within an image.	↓ (p=0.0501)	Lower GLN suggests a more homogenous gray level intensities in aVEGFR2 treated tumors.
GLRLM Run Length Non-Uniformity (RLN)	Measure of run length, or consecutive voxel values within an image.	↓ (p=0.0235)	Lower RLN suggests a more homogenous run lengths in aVEGFR2 treated tumors.
GLSZM Zone Percentage (ZP)	Measure of the coarseness of the image texture.	↑ (p=0.0149)	Higher ZP suggests a finer structure consisting of smaller zones in aVEGFR2 treated tumors.
GLDM Small Dependence Emphasis (SDE)	Measure of the distribution of small dependence, or small areas.	↑ (p=0.0243)	Higher SDE suggests more small areas in aVEGFR2 treated tumors.
GLSZM Large Area Emphasis (LAE)	Measure of the distribution of large area size zones	↓ (p=0.0163)	Lower LAE indicates fewer or smaller large homogenous areas in aVEGFR2 treated tumors.

We also investigated the relationships between these 5 radiomic features and metrics associated with the acoustic emissions feedback control system. As such, we performed a Spearman’s correlation and found that the acoustic emissions metrics associated with triggering of the feedback control mechanism to reduce PNP (i.e., number of thresholds, subharmonic, and 1^st and 2^nd ultraharmonics) were positively correlated with GLN, RLN, and LAE (Figure 5A). This indicates that, as these metrics increase, size zone and non-uniformity also increase. Conversely, SDE and ZP were inversely correlated with these metrics, suggesting that as there are fewer small-sized zones, these metrics increase. We next explored how well the acoustic emissions features predict these radiomic features using linear regression. We found that the 1^st ultraharmonic was strongly linearly predictive of both GLN (r² = 0.904, p = 0.001) and LAE (r² = 0.921, p = 0.000612) (Figure 5B–C). Additionally, the subharmonic was strongly linearly predictive of LAE (r² = 0.915, p = 0.000739) (Figure 5D). For each correlation, the IgG-treated tumors are in the upper right quadrant, with high radiomic and ultra- and subharmonic values.

Figure 5. Acoustic emissions signatures correlate with radiomic features.

A) Spearman’s’s correlation of the 5 significantly different radiomic features identified for IgG+FUS vs DC101+FUS (GLSZM large area emphasis, GLDM gray level non-uniformity, GLRLM run length non-uniformity, GLSZM zone percentage, and GLDM small dependence emphasis) compared to foldchange in PCD emissions over baseline (Final PNP, Total Number of Thresholds, Subharmonic, 1^st and 2^nd Ultraharmonics, Broadband, 2nd, 3^rd, and 4^th Harmonics). Circle size corresponds to the strength of the correlation, with larger circles indicating stronger correlations. Color scale indicates correlation direction, with −1 indicating inverse correlation, and +1 indicating positive correlation. Significant correlations are indicated with white asterisks. *p<0.05; **p< 0.01; ***p< 0.001. B, C) Linear regression of the average 1^st ultraharmonic as a predictor of GLRLM Gray Level Nonuniformity (B) and GLSZM Large Area Emphasis (C). D) Linear regression of average subharmonic as a predictor of GLSZM Large Area Emphasis.

Acute neoadjuvant VEGF inhibition does not enhance total FUS-mediated MH delivery or modulate PCD emission signatures

We reasoned that the improvement in the uniformity of the intratumor distribution of a small molecule model drug after chronic VEGFR2 blockade and FUS may be attributed to either (i) changes in vascular structure and/or (ii) baseline permeability, the latter of which may have reduced interstitial tumor pressure. To better understand the influence of these individual variables on small molecule model drug distribution, we leveraged an acute (24h) neoadjuvant αVEGF administration strategy that decreases permeability, much like chronic αVEGFR2 inhibition, albeit without altering vascular structure. To this end, we first tested how a single dose of aVEGF at two different concentrations, 15 mg/kg and 25 mg/kg, would impact the concentration of MH 24 h post aVEGF I.V. administration (Figure 6A). Representative T1 maps are shown in Figure 6B. T1 mapping showed that, at baseline, the 15 mg/kg aVEGF dose group had a trend towards a decrease in MH concentration, and there was a significant reduction in concentration with a 25 mg/kg dose (Figure 6C, p = 0.0576, p = 0.0181). Overall, a single dose of aVEGF reduced the average baseline concentration of MH by >20% in both treatment groups (Figure 6D). We then assessed the percent CD31+ vessel area within GL261 tumors and confirmed there was no change in vasculature structure between the untreated and the 25mg/kg aVEGF-treated tumors (Figure 6E, p = 0.3980). We next tested how reducing baseline leakiness, without altering vascular structure, impact FUS-mediated delivery of MH (Figure 6A). Quantification of MH concentration (Figure 6F) showed that FUS increased the concentration of MH to similar levels, independent of vessel pre-treatment. This suggests that, under these conditions, vascular pre-treatment does not limit FUS BTB opening, with both FUS and aVEGF+FUS increasing the concentration 2.21- and 2.12-fold, respectively (Figure 6G, p = 0.4225).

Figure 6. Acute neoadjuvant αVEGF administration neither enhances total FUS-mediated Multihance delivery nor alters acoustic emission signatures.

A) Overview of experiment to test the influence of a single dose of αVEGF administration on intratumoral MH concentration, both with and without FUS. B) Left: Representative baseline contrast enhanced images and overlayed T1 maps, with (bottom) and without (top) αVEGF. Right: Representative post-FUS magnitude images and T1 maps, with (bottom) and without (top) αVEGF. C) Bar graphs of MH concentration at baseline and 24 h post 15 (left) and 25 (right) mg/kg αVEGF administration. Paired t-tests (one-tailed). D) Fold changes in MH concentration at 24 h after αVEGF administration. Welch’s t-test (two-tailed). E) Percent tumor area positive for CD31 in untreated and 24 hours-post 25 mg/kg αVEGF treated mice. Welch’s t-test (two-tailed). F) Bar graph of mean intratumor MH concentration after FUS or FUS at 24h after 25 mg/kg αVEGF. Welch’s t-test (one-tailed). G) Post-FUS concentration over baseline. Welch’s t-test (one-tailed). H) Average peak negative pressure (PNP) over treatment time. 20 baseline bursts were applied at 0.2 MPa to establish PCD thresholds. I) Bar graph of average final PNP. Welch’s t-test (one-tailed). J) Subharmonic, 1^st and 2^nd Ultraharmonic, and broad band (BB) noise over baseline emissions. K) 2^nd and 3^rd harmonics over baseline emissions. (FUS, n = 6; αVEGF, n = 6 for concentration and n=5 for PCD). Welch’s t-tests (one-tailed). Means ± SEM.

Upon assessing the corresponding fold-changes in PCD emissions, we found no differences in final PNPs between groups, with an average PNP of about 0.37 MPa for FUS and aVEGF+FUS, respectively (Figure 6H, I). Additionally, we observed no change in subharmonic emissions (p = 0.1235), a slight increase in 1^st ultraharmonic emissions (p = 0.0172), and no change in 2^nd ultraharmonic (p = 0.2522) and broadband (p = 0.9729) emissions (Figure 6J). Although the 2^nd and 3^rd harmonics were higher than baseline, there were no differences between groups (Figure 6K). In addition to limited differences in PCD emission signatures, there were no differences in concentration distribution skewness or kurtosis (Figure 7A–B) nor in the radiomic features identified with DC101 treatment (Figure 7C–G). This suggests that, while MH concentration is reduced to similar levels with both aVEGFR2 pre-treatment and acute αVEGF administration due to augmented vessel permeability, modulation of vessel size may be necessary to achieve enhanced local-MW drug delivery distribution.

Figure 7. Statistical and radiomic analysis indicate that a single dose of αVEGF does not improve the uniformity of intratumor MH distribution after FUS.

A, B) Intratumoral MH concentration distribution skewness (A) and kurtosis (B). C, D) Radiomic features indicative of enhanced uniformity – GLSZM Zone Percentage (C) and GLDM Small Dependence Emphasis (D). E-G) Radiomic features indicative of enhanced non-uniformity – GLDM Gray Level Non-Uniformity (E), GLRLM Run Length Non-Uniformity (F), and GLSZM Large Area Emphasis (G). Welch’s t-tests (one-tailed). Means ± SEM.

Discussion

GBM is a devastating disease with dismal survival outcomes for patients. Rapidly translatable treatment options are sorely needed[41]. FUS BBB and BTB opening represents a powerful tool for GBM therapy as it can safely augment chemo- and immunotherapeutic drug delivery under MRI-guidance. However, dysregulated tumor vasculature may be an unappreciated limiting factor for the efficacy of FUS-mediated drug delivery. Further, few studies have tested how the uniformity of intratumor drug delivery after FUS can be improved. Because FUS facilitates drug delivery across the BTB by transmitting energy from oscillating MBs to tumor vasculature, we postulated that pre-treating tumor vasculature via chronic VEGFR2 inhibition to normalize vascular structure could enhance the uniformity of subsequent drug delivery with FUS. To this end, we employed a small molecule gadolinium-based model drug, MH, which has a molecular weight of 1058 Da (1058 g/mol)[42]. For comparison, many clinically approved small-molecule therapies that can, or partially, cross the BBB fall in the 100–900 Da range[43]. MH sits at the upper end of molecular weights relevant to clinical brain tumor small-molecule therapeutics representing a challenging delivery scenario, making it a useful surrogate for MRI-based monitoring of small-molecule drug delivery and distribution.

As expected, chronic VEGFR2 inhibition significantly reduced baseline GL261 tumor vessel permeability to MH and tumor vessel caliber when compared to IgG control (Figure 1). While the total intratumor delivery of low-MW model drug with FUS did not improve with chronic VEGFR2 inhibition (Figure 2), radiomic feature analysis identified a significant improvement in intratumor drug distribution uniformity (Figure 4). Notably, this improved uniformity was accompanied by a transition toward acoustic signatures indicative of more stable MB oscillation (Figures 3 & 5). To test whether enhanced uniformity was caused by reduced baseline permeability and/or reduced vascular caliber, we performed FUS BTB opening after acute VEGF inhibition, an intervention that mimicked the permeability change elicited by chronic VEGFR2 blockade, albeit without affecting vascular caliber (Figure 6). Acute aVEGF administration before FUS affected neither total delivery, the uniformity of delivery (Figure 7), nor the acoustic signatures. Thus, we conclude that narrowing vascular caliber in glioma via chronic VEGFR2 inhibition improves subsequent FUS-mediated penetration of small molecule drugs into otherwise difficult to access tumor regions. The physical mechanism of this response may entail switching MBs to a more stable oscillatory regime. In all, this power biophysical effect opens new opportunities for generating therapeutic synergy between VEGFR2 inhibition, FUS, and small molecule drug delivery.

To assess how VEGFR2 pre-treatment impacted FUS-mediated delivery of MH, tumor-bearing mice were treated with FUS one day following baseline imaging. There were no differences in overall MH concentration after FUS between DC101 or control-treated tumors (Figure 2B). This may indicate that under these experimental conditions, blood vessel size and permeability are not a limiting factor in overall delivery with FUS. However, there is a trend towards a higher fold change in post-FUS to pre-FUS average concentration, suggesting vascular pre-treatment may enhance the degree of barrier opening and local delivery, without changing the overall final concentration (Figure 2C).

It is of note that the FUS parameters used herein were modest, as we used a PCD feedback control algorithm with a maximum PNP of 0.4 MPa. In a recent study assessing the transcriptional and tissue response to BBB opening with albumin-shelled MBs in naïve brain, we found minimal tissue damage at comparable treatment parameters[44], suggesting an acceptable safety profile. The PCD feedback control mechanism enhances this safety profile by tuning the FUS PNP based on the MB dynamics on a per-sonication spot, per-mouse basis. While PCD feedback is clinically used for enhanced BBB opening safety, studies suggest that higher pressures correspond to an increase in therapeutic drug delivery[21]. Further, we employed Optison^™ albumin-shelled MBs that have different dynamics than other clinically used MBs, such as lipid-shelled, which can show larger amplitudes of oscillation under lower FUS pressures, interacting differently with the surrounding endothelium[45,46]. Moreover, bubble formulation, polydispersity, average size, and administration route (i.e., bolus vs infusion) are likely also of importance, with reports suggesting different emissions and delivery responses that may be differentially susceptible to the changes due to VEGFR2 inhibition[47–50].

Our results indicate that changes in vascular structure and/or permeability induced by VEGFR2 inhibition enhance harmonic acoustic emissions, while reducing subharmonic and ultraharmonic emissions. Other studies have explored how MBs within channels comparable in size to brain and tumor microvasculature have different thresholds for sub- and ultraharmonic emissions generation as well as oscillation behavior[51]. Additionally, one study using Definity MBs in Sinclair swine with naturally occurring melanomas suggests that there may be an inverse linear relationship between tumor interstitial fluid pressure and subharmonic PCD emissions[52]. aVEGFR2 treatment herein results in reduced vessel size and permeability, and studies suggest DC101 can augment tumor interstitial fluid pressure [53]. To begin understanding which aspect of vascular modulation may change emission signatures, we compared results between VEGFR2 and VEGF inhibition. We would predict that the changes in vasculature with aVGEFR2 treatment would lower inertial cavitation signatures, reducing the number of PCD thresholds, resulting in a higher average final PNP. Indeed, aVEGFR2 treatment resulted in a higher average PNP over the treatment time (Figure 3C), with a trend towards a higher final PNP (Figure 3D) and a significantly lower average number of PCD feedback control system thresholds indicating differences in inertial cavitation signatures(Figure 3E). However, with VEGF inhibition, where vessel permeability was reduced without changes to vessel size (Figure 7E), different emission signatures and overall PNP changes were observed (Figure 7J & K). This supports that the PCD emissions may be specifically sensitive to changes in vessel size, as seen with aVEGFR2 treatment.

Both statistical measures and radiomic features indicate that VEGFR2 inhibition improves the uniformity of model small molecule drug distribution in gliomas after FUS-mediated delivery (Figure 4). With the changes in vascular morphology between IgG- and DC101-treated tumors at the therapeutic window, this finer texture may be indicative of changes in vessel structure and subsequent differences in micro-level delivery. This supports that aVEGFR2 pre-treatment augments low-MW model drug distribution throughout the tumor at this therapeutic window. This finding is important, as non-homogeneous delivery of low-MW anti-cancer agents, like chemotherapies, can induce partial responses and secondary resistance mechanism formation, resulting in tumor progression[54,55]. It is of note that this enhanced distribution response may be different for the delivery of larger molecular weight therapeutics, like antibodies or nanoparticles, and future studies should explore how therapeutic size affects delivery and dispersion. We also explored the relationships between radiomic features with the corresponding PCD emission signatures. There were significant positive correlations between GLN, RLN, and LAE with subharmonic and ultraharmonic signatures (Figure 5A). Additionally, we found linearly predictive relationships between sub- and 1^st ultra-harmonic emission signatures. These findings emphasize the importance of multimodal monitoring of treatment outcomes to better understand how microenvironmental changes translate to different imaging and emission biomarkers.

Additionally, we recently published a study exploring how chronic aVEGFR2 treatment impacted baseline contrast agent concentration using T1 mapping to quantify Gadovist^™ delivery in a CT-2A mouse model of GBM[56]. aVEGFR2 significantly decreased baseline tumor permeability to Gadovist^™ at day 17 post tumor implantation, three days post DC101 treatment termination. This indicates that aVEGFR2 has similar effects on baseline permeability at tumor-specific therapeutic windows. This study also explored the immunological consequences of combined aVEGFR2 and FUS, identifying changes in the immune landscape, which can have long-term impacts on tumor permeability and therapeutic effects. This suggests that while VEGFR2 inhibition may not enhance overall survival, as seen here, there may be immunological benefits to vascular pre-treatment due to a more favorable immune microenvironment and more even immune cell exposure to circulating small molecule therapeutics.

These findings are additionally promising as the methods used herein are already used in the clinic, with common use of quantitative MRI techniques for FUS treatment planning and tumor diagnosis and monitoring[57], PCD modulated PNP using feedback control algorithms[41], and radiomic features to compare and predict treatment outcomes[58]. Additionally, by using therapeutics that are analogous to clinically approved and widely used treatment options, we test a treatment paradigm with high clinical potential. While VEGFR2 inhibition is not widely used in clinical GBM treatment, DC101 is analogous to the VEGFR2 antibody, CYRAMZA^®, which is clinically approved for progressed stomach/esophageal cancers, metastatic non-small cell lung cancer, colorectal cancer, and hepatocellular carcinoma[59]. Further, aVEGF is used widely in the clinic, as Avastin^®, in the treatment of recurrent GBM[60]. Overall, our results emphasize a promising strategy employing low-dose neoadjuvant VEGFR2 inhibition for enhanced uniformity of small molecule drug delivery with FUS to treat pockets of tumor that would otherwise remain untreated.

Materials and Methods

Cell Culture

Luciferase-transduced GL261 cells (GL261-luc2) were cultured for up to three passages and maintained in logarithmic growth phase in high glucose 1x Dulbecco modified Eagle medium (DMEM, Gibco) supplemented with 1 mM sodium pyruvate (Gibco), non-essential amino acids (Gibco), 10% fetal bovine serum (Gibco), and 100 μg/mL G418 (GoldBio). Cells were maintained at 37°C and 5% CO₂. Cells tested negative for mycoplasma.

Intracranial Tumor Cell Inoculation

Animal experiments were approved by the Animal Care and Use Committee at the University of Virginia and conformed to the National Institutes of Health guidelines for the use of animals in research. GL261-Luc2 cells (1×10⁵ cells per 2 μL) were resuspended in sterile PBS for intracranial tumor implantation then implanted into the right striatum of 6–10-week-old female C57BL/6 mice (The Jackson Laboratory). Following anesthetization with an intraperitoneal injection of Ketamine (50 mg/kg; Zoetis) and Dexdomitor (0.25mg/kg; Pfizer), their heads were depilated. Mice were placed into a stereotactic head frame and aseptically prepared. GL261-Luc2 cells were implanted using a mechanically controlled rate of 0.5 μL/min with a 10 μL Hamilton syringe and micropump (UltraMicroPump, World Precision Instruments). Cells were injected at ~2 mm lateral from the sagittal suture, 0.5 mm anterior to bregma, and 3 mm below the dura. Mice were housed on a 12/12 h light/dark cycle and supplied food and water ad libitum.

MR Size Matching Imaging

MR imaging was performed on a 9.4T Bruker Biospec 94/20 small animal MRI. Spin-echo images for tumor volume size matching were acquired using a Bruker RARE sequence (repetition time of 2000 ms, echo time of 55 ms, turbo factor of 18, pixel size of 125 μm × 125 μm × 125 μm, 1 average, and 30 min acquisition time) with contrast (Multihance^™). DICOM images were manually segmented in Horos^™ (Horos Project) software to size-match tumors prior to treatment administration.

Antibody Injection

DC101 pretreatment experiments were conducted using 100 μL of anti-VEGFR2 monoclonal antibody (clone: DC101, Bio X Cell, BE0060) and rat IgG1 isotype control (clone: HRPN, Bio X Cell, BE008) injected intraperitoneally at a concentration of 10 mg/kg mouse weight on days 8, 12, and 14 post tumor implantations. Single-dose anti-VEGF-A pretreatment experiments were conducted using 100 μL of anti-mouse VEGF-A (clone: 2G11-2A05, Bio X Cell, BE0399) injected intravenously at a concentration of 15 or 25 mg/kg mouse weight immediately after MR imaging on day 13 post tumor implantation.

T1 Mapping MRI Acquisition

Data for T1 maps were acquired with a 9.4T small-bore MRI (Bruker BioSpec). The DC101 treated mice were imaged using a four-channel phased array coil (Bruker) and the subsequent aVEGF studies were conducted using a circular single channel surface coil (Bruker). A set of interleaved multi-slice 2D spin echo (SE) images were taken at varied repetition times (TR) to generate a saturation recovery curve. The slice gap was set to 100% of the slice thickness to eliminate cross talk. Two sets of 7 images, a total of 14 slices, were acquired prior to FUS and contrast agent administration to obtain saturation recovery curves with a satisfactory dynamic range. The two sets (slice packages) of image series were offset by the slice gap in the slice select plane to ensure 3D coverage of the brain. The parameters for these scans were: TR=790, 1040, 1350, 1750, 2300, 3215, and 7000 ms, TE=6.71 ms, slice thickness=0.6 mm, slice gap=0.6 mm, FOV=35 × 35 mm, matrix size=200 × 200, rare factor=10, and R= 0.175 × 0.175 × 0.6 mm³. After FUS and contrast agent administration, 10 additional sets of spin-echo images were acquired with identical parameters except at a fixed TR=1040 ms to monitor the contrast agent cumulation versus time. These acquisitions alternated between the two slice packages, resulting in 14 slices imaged at 10 different time points post-FUS. Time per acquisition was 1 minute and 19 seconds.

T1 Mapping Data Processing

Voxel-wise T₁ values, both pre- and post-contrast, were obtained by fitting the variable TR acquisitions to a two-parameter saturation-recovery model under the assumption of perfect saturation[61–63]:

$$\left|S\right|=M_0\left(1-e^{\frac{-TS}{T1}}\right)$$ Eqn [1]

In equation 1, |S| is the magnitude of the signal within the voxel, M₀ is the product of the thermal equilibrium magnetization, T2-decay exponential, and coil sensitivity, TS is the saturation recovery time (ms), and T1 is the spin-lattice relaxation (ms). For the fitting, TS was defined as the interval between the last refocusing pulse in the echo train and the next excitation pulse. This adjustment accounts for the fact that repeated refocusing pulses perturb the longitudinal magnetization and limit its recovery, making the effective saturation-recovery period shorter than the nominal TR delay.

A custom-written MATLAB (MathWorks) script fit the signal magnitude data on a voxel-by-voxel basis to equation 1. Each fitting procedure simultaneously fit the data to 6 functions: function 1 incorporated the 7 pre-contrast variable TR scans, while functions 2–6 incorporated the singular scan at a fixed TR but different time points. The fits were constrained to have the same M₀ value but allowed different T1 values at each time point. Pre-contrast and post-contrast T1 values were then used to calculate the contrast agent concentration on a voxel-by-voxel basis at each time point using equation 2.

$$\frac{1}{T_{1\_\mathit{\text{Post}}}}=\frac{1}{T_{1\_Pre}}+r_1{C}_1$$ Eqn [2]

In equation 2, T_{1_Post} is the post-contrast value at a particular time point (ms), T_{1_Pre} is the pre-contrast T1 value (ms), r₁ is the known contrast agent relaxivity (4.9×10⁻³L/mmol/ms for Multihance)[64], and C1 is the contrast agent concentration (mM). The alternating slice-package acquisition introduced temporal gaps in concentration for each package, as the two slice groups were sampled at different time points. To obtain full 3D coverage at each nominal time point, concentration values at the unacquired time points were estimated by linearly interpolating between the nearest measured points for each package. For slice package 2, no measurement was available at minute 0; therefore, a concentration of zero was assigned at that time point to enable interpolation. A second custom MATLAB script was used to calculate average concentrations with manually drawn regions of interest (ROIs) on the concentration maps which corresponded to the area of the tumor. Extracted tumor concentration maps were then overlayed on top of the corresponding T1 contrast image using Fiji/ImageJ[65].

Immunofluorescence

Twelve hours after FUS or time-matched control, mice underwent cardiac perfusion with phosphate-buffered saline (PBS) followed by 1x zinc fixative (BD Bioscience, 550523). A subset of mice were injected via tail vein catheter with DyLight 649 Tomato-Lectin (Vector Laboratories, DL-1178-1) five minutes prior to harvest for immunofluorescence. Brains were then dissected and fixed overnight in zinc fixative, followed by overnight cryopreservation in 30% sucrose. Brain samples were embedded in OCT (Fisher, 23-730-571) and frozen on dry ice. Frozen blocks were stored at −80 °C before sectioning and staining. Cryo coronal sections with the tumor were cut at 20 μm thick. The mounted sections were incubated with blocking solution (1% NDS in 2% BSA and 0.1% Tween 20 PBS) for 1 h at room temperature. Sections were next incubated overnight at 4°C with Alexa Fluor 488 anti-mouse CD31 (1:200, BioLegend, 102514) in antibody solution (2% BSA and 0.1% Tween 20 in PBS). After washing 3x for 10 min in 0.1% Tween 20 PBS, sections were incubated for 20 mins at room temp with DAPI (1:1000, ThermoFisher, 62248). After final washes in PBS, sections were sealed with ProLong Gold antifade reagent (ThermoFisher, P36930) and cover slipped with cover glass for confocal imaging.

Confocal Microscopy

Stained sections were imaged with a Leica Stellaris 5 confocal microscope using sequential scanning mode for DAPI, Alexa 488, and perfused Lectin 649. All images were acquired at 20X magnification. Images were analyzed with Fiji/ImageJ[65]. Final images were adjusted with Fiji/ImageJ and assembled in PowerPoint (Microsoft).

Arterial Spin Labeling

MRI was performed using a 9.4 T small bore 94/20 small animal MRI. Anesthesia was maintained throughout image acquisition using 1.5%–2% isoflurane delivered in a 1 L/min 50:50 mixture of oxygen and air. Respiratory rate was continuously monitored during the scan (SA Instruments Inc., Stony Brook, NY, USA). For radiofrequency transmission, an 86 mm quadrature volume coil was used, while signal reception was achieved via a four-channel phased-array mouse brain coil (Bruker BioSpin GmbH, Germany). Scout images in the transverse plane were acquired to confirm accurate positioning. A single 1 mm-thick, two-dimensional T2-weighted fluid inversion recovery (T2-FLAIR) coronal image (image matrix = 80 × 80, field of view = 18 × 18 mm, 8.250 ms echo spacing, 33 mg echo time, 8 rare factor, 10,000 ms repetition time, 90-degree excitation angle, 180 degree refocusing angle, 1 dummy scan with 10,000 ms duration) slice was placed such to intersect the center of the tumor based off the transverse anatomical images, followed by magnetic field optimization using an automated 3D field mapping routine with automated 3D shimming.

ASL imaging was conducted using a flow-sensitive alternating inversion recovery (FAIR) spin-echo EPI sequence. A single 1 mm-thick slice at the region of interest, identified from the T2-weighted anatomical images, was selected (80 × 80 image matrix, 18 × 18 mm field of view, 11.12 ms echo time, 90-degree flip angle, 10,000 ms recovery time, with variable repetition times). Cerebral blood flow (CBF) maps were generated using Bruker Paravision 360 software V3.4. CBF values were extracted from original, unprocessed voxels.

MRI-Guided Focused Ultrasound

FUS was performed using the RK-300 small-bore FUS device (FUS Instruments). Prior to FUS, the heads of mice were shaved and depilated. Mice were placed supine on the RK-300 MRI-guided focused ultrasound device, and their heads were coupled to the transducer with degassed ultrasound gel. BBB opening was performed using a 1.15 MHz single-element transducer with a 25 mm aperture, 20 mm radius of curvature, 0.8 f number, using a 10 ms burst length and 2000 ms period, for a total of 60 sonications over a 2-min treatment duration. The transducer’s lateral normalized pressure full width at half maximum (FWHM) was 1.4 mm. Passive cavitation detection (PCD)-modulated peak negative pressure (PNP) was employed using the “Blood-brain Barrier” mode on the FUS Instruments Aureus Software. The feedback control parameters used were 0.2 MPa starting pressure, 0.05 MPa pressure increment, 0.4 MPa maximum pressure, 20 sonication baselines without microbubbles, 500 Hz AUC bandwidth, 10 standard deviations AUC threshold, 0.95 pressure drop, and selection of subharmonic, first ultraharmonic, and second ultraharmonic frequencies. Baseline T1 Mapping images were taken, then 100 μL of the contrast agent model therapeutic, Multihance^® (gadobenate dimeglumine; Bracco) was injected intravenously as a bolus dose via tail vein catheter. Immediately after, Optison^™ albumin-shelled microbubbles (GE HealthCare) were intravenously injected at a bolus dose of 10⁵ microbubbles per gram body weight. Microbubble distribution, diameter, and concentration were obtained using a Coulter counter (Multisizer 3; Beckman Coulter, Fullerton, California) prior to use, representative values include 265.6×10⁶ bubbles per mL, mean size of 4.413 μm, median size of 4.266 μm, and mode size of 4.815 μm with a size range of ~2–10 μm. Following microbubble injection, catheters were flushed with 2% heparin saline. Baseline T1 Mapping images were used to guide FUS targets over the tumor volume. All experiments used 4 non-overlapping sonication targets in a diamond shape covering the tumor volume. Directly after the 2-minute FUS treatment period, post-T1 mapping images were acquired.

Passive Cavitation Detection Analysis

Passive cavitation detection (PCD) data was obtained using a hydrophone embedded within the center of the FUS transducer with a 0.5 MHz high pass and 5 MHz low pass filters (FUS Instruments) during FUS treatment with the “Blood brain Barrier” mode selected. This system monitored MB acoustic emissions and modified the PNP based on bubble dynamics on a per-animal, per-target basis. The feedback control parameters included a 0.2 MPa initial pressure, 0.05 MPa pressure increment, 0.4 MPa maximum pressure, 20 sonication baselines without microbubbles, 500 Hz AUC bandwidth, 10 standard deviations AUC threshold, 0.95 pressure drop, and selection of subharmonic (0.5f₀), first Ultraharmonic (1.5f₀), and second Ultraharmonic (2.5f₀) frequencies. These parameters were adapted from Fisher et al. [66] and Sheybani et al. [67].

Baseline emissions were first acquired using 20 sonications at an initial PNP of 0.2 MPa without MBs. After baseline acquisition, MBs were injected, and treatment began at the same initial pressure. For each target, the PNP was then increased by 0.05 MPa for each subsequent sonication until either (1) MB emissions exceeded the algorithm-determined permissible threshold, or (2) the PNP reached a maximum pressure of 0.4 MPa. The permissible threshold was defined as any of the three 0.5f₀, 1.5f₀, and 2.5f₀ emission levels exceeding 10 standard deviations from their corresponding baseline signals obtained without microbubbles. When the threshold was exceeded, the system automatically reduced the PNP by 5% and continued lowering the pressure until the emission spectra fell back below the threshold. After the emissions were below the threshold, the PNP was held at that level until the end of the treatment unless the threshold was exceeded again, triggering another pressure drop.

PCD data were processed with a custom MATLAB script (MathWorks) to extract the PNP and quantify PCD emissions for each FUS+MB sonication. Emissions were quantified as the AUC within a 300 Hz bandwidth centered on each monitored emission frequency (0.5f₀, 1.5f₀, 2.5f₀, 3.5f₀, f₀, 2f₀, 3f₀, 4f₀). Broadband emissions were calculated by summing the emissions after removing all emissions at the fundamental frequency (f₀), harmonics (2f₀, 3f₀, 4f₀), subharmonic (0.5f₀) and ultraharmonics (1.5f₀, 2.5f₀, 3.5f₀). Each mouse had 4 treatment targets, covering the tumor in the right striatum. Emissions recorded during the FUS+MB treatment were normalized to each target’s corresponding average baseline emission, generating a fold-change over baseline. Fold-changes were then averaged across all targets to produce an overall treatment-level fold-change for each emission type. Final foldchange values were plotted using GraphPad Prism.

Radiomic Feature Extraction

For preliminary kurtosis and skewness measurements, the distribution of intratumoral concentrations was assessed using GraphPad Prism. For additional radiomic features, the solid tumor region was segmented using the T1W image for baseline and post-FUS imaging for each mouse in 3D Slicer (slicer.org)[68]. Radiomic features were extracted from the final timepoint concentration maps acquired using the above methods for all segmented regions using the open-source library Pyradiomics[40]. Because some radiomic features require isotropic voxels, prior to feature extraction, all images were resampled to cubic voxels with a side length of the smallest dimension of the pixel spacing using the Pyradiomics feature extractor. After features were extracted, R Studio was used to screen significant differences in post-FUS radiomic features, excluding shape based features, between the two treatment groups. Significant features were then plotted in GraphPad Prism. Next, we used R Studio to perform a Spearman’s correlation to compare the average PCD foldchange values (average total number of thresholds, average foldchange in subharmonic, average foldchange in 1^st and 2^nd ultraharmonics, average foldchange in 2^nd, 3^rd, and 4^th harmonics, and average foldchange in broadband noise) to the 5 significant radiomic features using the cor.test( ) function in base R with method ‘spearman’. Features with correlation p-value < 0.05 were then fit with a linear regression model using the lm( ) function in R with the radiomic features as the dependent and PCD emissions as the independent variables. The linear regressions were then subset to identify and plot those with p-value > 0.05 and R-squared > 0.85 to only focus on strong linear relationships.

Statistical Analysis

All results reported with error bars are means with the standard error of the mean. The number of mice per group is made evident by a statement of “n” value in the figure legend. Statistical significance was assessed at p < 0.05 for all experiments and calculated using GraphPad Prism 9 (San Diego, USA). Statistical significance for radiomic features was determined using a custom R script. Statistical tests are provided in the figure legends.

Data Availability and Sharing.

All data are provided in the manuscript and also available upon reasonable request.