VEGFR2-targeted three-dimensional ultrasound imaging can predict responses to anti-angiogenic therapy in preclinical models of colon cancer

Abstract

Three-dimensional (3D) imaging capabilities to assess responses to anticancer therapies are needed to minimize sampling errors common to two-dimensional approaches as a result of spatial heterogeneity in tumors. Recently, the feasibility and reproducibility of 3D ultrasound molecular imaging (USMI) using contrast agents which target molecular markers has greatly improved, due to the development of clinical 3D matrix array transducers. Here we report preclinical proof of concept studies showing that 3DUSMI of VEGFR2/KDR expression accurately gauge longitudinal treatment responses to anti-angiogenesis therapy, in responding versus non-responding mouse models of colon cancer. Tumors in these models exhibited differential patterns of VEGFR2-targeted 3DUSMI signals during the course of anti-angiogenic treatment with bevacizumab. In responding tumors, the VEGFR2 signal decreased as soon as 24 hours after therapy was started, whereas in non-responding tumors there was no change in signal at any time point. The early decrease in VEGFR2 signal was highly predictive of treatment outcome at the end of therapy. Our results offer a preclinical proof that 3DUSMI can predict responses to anti-angiogenic therapy, warranting further investigation of its clinical translatability to predicting treatment outcomes in patients.

Introduction

Tumor angiogenesis, the spouting of neoangiogenic vessels from pre-existing vasculature, is a crucial process for tumor progression and metastasis (1). Among the signal pathways involving angiogenesis, vascular endothelial growth factor (VEGF)-A has been shown to be one of the key mediators of physiological and pathologic angiogenesis (2). VEGF-A, a signaling protein released by tumor cells and associated stroma, predominantly binds to VEGF receptor type 2 (VEGFR2) and the activation of the VEGF/VEGFR2 pathway stimulates endothelial cell migration, proliferation, and vascular permeability (3).

Given the eminent role of VEGF/VEGFR2 pathway in tumor angiogenesis, growth and progression, targeting VEGF or its receptors for therapeutic purposes has been proposed more than 10 years ago (4), and various types of VEGF inhibitors are currently in different stages of clinical development. Bevacizumab, an anti-VEGF-A humanized monoclonal antibody, has been the first anti-angiogenic drug approved by the Food and Drug Administration in the USA in patients with metastasized colorectal cancer (5). Although multi-center randomized controlled trial showed that the addition of bevacizumab to chemotherapy improved progression-free survival and overall survival in patients with metastatic colorectal cancer, the overall tumor response rate was only 44.8%(6). Inherent or acquired resistance is one of the major challenges associated with anti-angiogenic therapy (7). Therefore, there is a critical need for developing and validating predictive biomarkers for earlier identification of responders from non-responders to minimize unnecessary treatments and associated adverse effects and to allow earlier transition to next-line therapy. However, traditional end points based on anatomical oncological imaging has been shown to be insufficient for evaluation of anti-angiogenic treatments (8).

Molecular imaging allows visualization and quantification of specific molecular markers expressed in diseased tissues, thereby facilitating earlier evaluation of treatment response before overt anatomical changes occur. Ultrasound molecular imaging (USMI) using contrast microbubbles targeted at molecular markers expressed on the tumor neovasculature is emerging as a promising tool to assess expression levels of angiogenic markers such as VEGFR2 for earlier cancer detection (9–12) and monitoring anti-angiogenic therapy (13,14). USMI combines the advantages of ultrasound such as relatively low cost, wide availability and portability, and lack of ionizing radiation with those of molecular imaging to allow quantitative monitoring of disease processes at the molecular level. Recent studies have highlighted the need for three-dimensional (3D) imaging capabilities of USMI which is critically needed for longitudinal monitoring of treatment response in cancer due to the known spatial heterogeneity of molecular markers of neovasculature caused by focal areas of hypoxia, necrosis, or hemorrhage in cancer. This can result in sampling errors on consecutive USMI exams with substantial over- and underestimation of treatment response if only a two-dimensional (2D) imaging approach is used (15,16). Recent advances in ultrasound transducer technology with the introduction of 3D matrix array transducers have led to a marked improvement of 3DUSMI imaging and has made 3D ultrasound molecular imaging (3DUSMI) feasible and reproducible (17). However, it is not known whether 3DUSMI allows prediction of longitudinal treatment response of anti-angiogenic therapy in responding and non-responding tumors.

Therefore, the purpose of our study was to assess whether VEGFR2-targeted 3DUSMI using a clinical matrix array transducer and clinical ultrasound system allows prediction of treatment response to anti-angiogenic therapy in two animal models of colon cancer that simulate clinical responders and non-responders.

Materials and Methods

Murine models of responding and non-responding colon cancer to anti-angiogenic treatment

All experiments involving animals were approved by the Institutional Administrative Panel on Laboratory Animal Care. To simulate clinical responders and non-responders, two colon cancer cell lines were used:1) the human colon cancer cell line LS174T (obtained from ATCC in January 2015, Manassas, VA), which has been shown to be sensitive to the treatment of VEGF pathway blockade (18); and 2) the murine colon cancer cell line CT26.WT (obtained from ATCC in February 2015, Manassas, VA), which has been shown to be resistant to VEGF pathway blockade (19,20). Both cell lines were cultivated in Dulbecco’s Modified Eagle Medium (DMEM; Gibco, Grand Island, NY) supplemented with 10% fetal bovine serum (Gibco), penicillin (50 U/ml), and streptomycin (50 μg/ml) at 37°C in a humidified 5% CO2 atmosphere. Tumor cells were collected following trypsinization, and4×10⁶ LS174T cells or 1×10⁶ CT26 cells suspended in 50 μl phosphate-buffered saline and 50μl Matrigel (BD Biosciences, San Jose, CA) were injected subcutaneously on the right lower hind limb of 6–8 weeks old female nude mice (Charles River, Wilmington, MA).

Anti-angiogenic treatment

Both LS174T and CT26 tumors were allowed to grow for 10 days after cancer cell injection when the tumor reached a maximum diameter of 6–13 mm (mean, 10mm) as measured by using an electronic caliper available on the ultrasound system. Animals with either tumor type were randomized into 1) the treatment group (n=8 from each cell type; total, 16 tumors) and 2) the control group (n=8 each; total, 16 tumors). Animals in the treatment group received bevacizumab (Avastin, 10 mg/kg injected intravenously; Genentech, South San Francisco, CA) on days 0, 3, and 7, and the control group received sterile saline (Figure 1).

Figure 1.

Timeline of longitudinal 3DUSMI scans and anti-angiogenic treatment schedule for both LS174T and CT26 tumors. Ten days after tumor cell injection, 3DUSMI scans were performed prior to treatment to obtain baseline imaging (day 0), and at subsequent days 1, 3, 7, and 10 after treatment initiation using either an anti-angiogenic drug (bevacizumab) or saline. All mice were sacrificed at day 10 and tumors were removed for ex vivo analysis.

Three dimensional ultrasound molecular imaging with VEGFR2-targeted microbubbles

Three dimensional USMI was performed in all animals on days 0, 1, 3, 7, and 10 by using a clinical ultrasound scanner (EPIQ 7; Philips Healthcare, Andover, MA) and a clinical matrix array transducer (X6-1, 6.0–1.0 MHz frequency, 9212 elements; Philips). To reduce artifacts in the near-field zone of the clinical transducer, a 3-cm customized standoff ultrasound gel was placed on the skin of mice. In all tumors, 3DUSMI was performed in power modulation contrast imaging mode using the following imaging settings: center frequency, 3.2 MHz, mechanical index, 0.09; volume rate, 1 Hz; dynamic range, 52 dB; imaging focus, 5 cm. All imaging settings were kept constant during all experiments. The transducer was fixed in stable position with a clamp to minimize motion artifacts. All mice were anesthetized with 2% isoflurane in room air (administered at 2 L/min) during imaging and placed on a heated support in order to maintain constant body temperature for the whole duration of the experiment. The greatest longitudinal, transverse, and anteroposterior dimensions of tumors were measured in grayscale imaging using electronic calipers. Tumor volume was determined using the formula for a prolate ellipsoid (volume = π/6 × L × W × H, where L is length, W is width, and H is height).

Clinical-grade VEGFR2 targeted microbubbles (MB_VEGFR2, BR55, Bracco Suisse, Geneva, Switzerland) (12,21), were used for 3DUSMI. MB_VEGFR2 is a perfluorobutane and nitrogen gas-filled, phospholipid-shelled microbubble with a mean diameter of 1.5 ± 0.1μm (range, 1–3 μm) as assessed by using a cell counter and sizer (Multisizer III Coulter Counter, Beckman Coulter, Fullerton, CA), and functionalized with a VEGFR2-binding heterodimeric peptide (5.5 kDa; dissociation constant, KD = 0.5 nmol/L) (22,23). The mean number of heterodimeric peptides per square micrometer of the microbubble shell was 34,200 ± 1300 (range, 31,800–36,600)(13).

MB_VEGFR2 (5×10⁷ microbubbles;100μl) were injected at a constant injection rate within 5 seconds by using an infusion pump (Kent Scientific, Torrington, CT)through a 27G needle catheter (Vevo Micromarker; VisualSonics, Toronto, Canada) placed in a tail vein. Four minutes after microbubbles injection, which allowed the microbubbles to circulate through the whole tumor volume and attach to VEGFR2, a sequence of two high-power ultrasound pulses (mechanical index, 0.72) was applied for a 2-second duration to destroy all bound and unbound microbubbles in the tumor volume. Sixty seconds after microbubble destruction (the time needed to fully replenish the tumor vessels within the tumor volume in the animal models used, according to our experience), imaging signal was measured a second time and the 3DUSMI imaging signal intensity from molecularly bound MB_VEGFR2 was quantified as the difference in imaging signals before and after the high-power ultrasound pulses as described (17,24,25). The entire 3DUSMI imaging sequence was recorded in real time by using a built-in Digital Navigation Link of the ultrasound system with custom in-house MevisLab modules written in C++ (26).

To confirm binding specificity of MB_VEGFR2 to VEGFR2, a subset of mice with either LS174T (n=6) or CT26 (n=6; both randomly chosen from control mice during the treatment course and from treated mice at day 0 before antiangiogenic treatment), was imaged with non-targeted control microbubbles (MB_Control with the same composition as MB_VEGFR2 except the lack of VEGFR2-specific lipopeptide) after two separate sequential microbubble injections. A minimum of 30 minutes of waiting time between MB_Control and MB_VEGFR2 injections was allowed for clearance of microbubbles from previous injections (27,28).

Image analysis of 3D ultrasound molecular imaging data sets

All 3DUSMI datasets were analyzed by one reader blinded to the treatments status with a custom software which was developed by using the software package of MeVisLab (29). A volume of interest (VOI) was manually contoured covering the whole tumor visualized on sagittal, longitudinal, and coronal planes. Results obtained from the selected VOI represented a linear depiction of the signal intensity from molecularly attached microbubbles, and the intensity is linked to microbubble concentration (30). Tumor volume and 3DUSMI signals intensities after treatment were normalized to the baseline values (Day 0).

Ex vivo analysis of tumors

All mice were sacrificed at day 10 after imaging and tumors were excised for ex vivo analysis. Tumor tissues were fixed in 4% paraformaldehyde and phosphate-buffered saline solution for 24 hours, followed by 30% sucrose and phosphate-buffered saline solution (Sigma Aldrich, St Louis, Mo) for up to 3 days. Tumor tissues were embedded in optimum cutting temperature (OCT, Fisher Scientific, Pittsburgh, Pa), and then sectioned into 10 μm slices for immunofluorescence staining. Standard methods for immunofluorescence were used. Briefly, the sections were rinsed with phosphate buffered saline for 10 minutes to remove remaining OCT and permeabilized in 0.5% Triton-X 100 for 10 minutes. Incubation in a solution containing 3% bovine serum albumin (Sigma, St Louis, MO), 3% goat serum (Sigma, St Louis, MO), and 3% donkey serum (Sigma, St Louis, MO) for up to 5 hours was performed to block nonspecific proteins. Sections were then simultaneously incubated overnight at 4°C with primary antibodies [1:100 rat anti-mouse CD31 antibody (eBioscience, San Jose, CA) and 1:50 rabbit anti-mouse VEGFR2 antibody (Cell Signaling, Danvers, MA), followed by incubation of secondary antibodies of 1:250 AlexaFluor 546 goat anti-rabbit immunoglobulin G (IgG) (Invitrogen, Grand Island, NY) and 1:250 AlexaFluor488 donkey anti-rat IgG (Invitrogen, Grand Island, NY) for 30 min. Samples were mounted by using ProLong Gold (BiogeneX, San Ramon, CA). Fluorescent micrographs were captured by using a LSM510 metaconfocal microscope (Zeiss, Maple Grove,MN) attached to a digital camera (AxioCam MRc, Bernried, Germany) using a ×20 objective. Quantitative immunofluorescence of VEGFR2 expression and the percentage area of blood vessels per field of view was performed by using Image J software (National Institutes of Health, Bethesda, MD) as the average value from 5 randomly selected fields of view (single field of view area, 0.19 mm²).

Statistical analysis

Paired-samples Wilcoxon rank test was used to compare the changes in 3DUSMI signal intensity and tumor volume following treatment. Mann–Whitney U test was used to compare tumor volume, 3DUSMI signal and ex vivo VEGFR2 expression levels in anti-angiogenic treated and control groups. Comparisons of 3DUSMI signal intensities in the same tumor by using targeted versus non-targeted microbubbles were performed by using a paired-samples Wilcoxon rank test. To assess whether relative changes in 3DUSMI signal intensity at day 1 could predict treatment response at a later time point at day 10, imaging signals were compared to the relative change of tumor volume at day 10 with tumor progression defined as a 4-fold or more increase in tumor volume. The proportions of true positives and negatives were estimated and the 95% confidence intervals were constructed with exact method due to small sample sizes. All analyses were performed using SPSS version 16.0 (SPSS, Inc, Chicago, IL). A P value of < 0.05 or less was considered statistically significant.

Results

Effect of anti-angiogenic treatment on tumor growth

Before treatment (day 0), tumor volumes were not significantly different between control and treated mice for both responding (LS174T, P=0.226) and non-responding (CT26, P=0.833) colon cancer. In responding tumors, tumor volumes increased significantly more in saline treated mice (by 537%; P=0.001) compared to bevacizumab-treated mice (by 86%). On days 3, 7, and 10, the relative increase in tumor volume (P=0.001) in bevacizumab-treated mice was significantly smaller compared to saline-treated mice, while there was no significant difference at day 1 (P=0.674). In contrast, in non-responding tumors, tumor volumes in both bevacizumab-treated (by 799%) and saline-treated (by 839%) mice significantly increased (P=0.012) at all four time points. There were no significant differences in relative increase of tumor volumes compared to baseline between bevacizumab-treated and saline-treated at any time point (P from 0.345 to 0.916) in non-responding tumors (Figure 2).

Figure 2.

Longitudinal change of tumor volume in responding and non-responding tumors during the 10-day treatment course. (A) In treatment responders, tumor growth was substantially delayed following anti-angiogenic treatment compared to saline treatment on days 3, 5, 7, and 10, while there was no difference in relative changes of tumor volume at day 1.* P = 0.001 for comparison of control and treated tumors. (B) In non-responders, tumor volumes substantially increased in anti-angiogenic and saline-treated tumors without significant differences in tumor volumes at any time point.

Longitudinal change of VEGFR2-targeted 3D ultrasound molecular imaging signal and prediction of treatment response

In responding tumors, although tumor volume (normalized to baseline) of treated and control mice were not significantly different (P=0.674) at 1 day after initiation of bevacizumab treatment, 3DUSMI signal was significantly lower (by on average 64%; P=0.001) in bevacizumab-treated compared to saline-treated mice at day 1 as well as on all subsequent time points (by 73% at days 3, P=0.001; by 88% at day 7, P=0.001; and by 88% at day 10, P=0.001). 3DUSMI signal did not significantly change in saline treated mice on the first three time points (day 1: P=0.106; day 3: P=1.0; day 7: P=0.139), but then significantly decreased at day 10 by 51% (P=0.018), likely due to tumor necrosis (see discussion).

In contrast, in non-responding tumors, there were no significant differences in 3DUSMI signals in bevacizumab- or saline-treated mice on anytime point (P from 0.752 to 0.916; Table 1, Figure 3 and Figure 4).

Table 1.

Summary of the relative VEGFR2-targeted three-dimensional ultrasound molecular imaging signal compared to baseline (Day 0) obtained at four time points in responding (LS174T) and non-responding (CT26) colon cancers

Cell Line	Treatment	Day 0	Day 1	Day 3	Day 7	Day 10
Responder (LS174T)	Bevacizumab	1.0	0.36±0.15	0.27±0.17	0.12±0.08	0.12±0.08
	Saline	1.0	1.40±0.63*	1.13±0.43*	0.85±0.31*	0.49±0.29*
Non-responder (CT26)	Bevacizumab	1.0	1.36±0.54	1.22±0.74	0.77±0.44	0.57±0.29
	Saline	1.0	1.36±0.62	1.14±0.62	0.75±0.43	0.63±0.41

Note.-Numbers are mean ± standard deviation of VEGFR2-targeted 3DUSMI signal, expressed as relative values compared with baseline 3DUSMI signal

^*P = 0.001 for comparison of control and treated tumors in responding colon cancer (LS174T).

Figure 3.

VEGFR2-targeted 3DUSMI signal in responding and non-responding mice, treated either with bevacizumab or saline. Bar graph summarize relative VEGFR2-targeted 3DUSMI signal compared to baseline (day 0), obtained at four time points in responding (LS174T) (A) and non-responding (CT26) (B) colon cancers. * P = 0.001 for comparison of control and treated tumors.

Figure 4.

Representative volume-rendered VEGFR2-targeted 3DUSMI signals of responding and non-responding tumors imaged over the 10-day treatment course. In responding tumor, one day after initiation of anti-angiogenic therapy, 3DUSMI signal substantially decreased and remained low during the subsequent days in treated (T; lower row) versus control tumors (C; upper row). In non-responding tumors, there was no significant difference in the 3DUSMI signal between tumors treated with anti-angiogenic agent (T, lower row) and saline treated control tumors (C; upper row). C, control saline treated tumors; T, anti-angiogenic treated tumors. Scale bar = 10 mm.

At 30% or more decrease of the 3DUSMI signal at day 1 following initiation of anti-angiogenic therapy, a positive treatment response could be predicted in 100% of responding and non-responding tumors, respectively (95% CI, 68.5%, 100%).

Binding specificity of targeted microbubbles to VEGFR2

In a subset of 6 mice per tumor type, both MB_VEGFR2 and MB_Control were injected during the same imaging session. In both tumor types, the 3DUSMI signal was significantly higher (P=0.028) following MB_VEGFR2 injection (6.57×10⁵±3.88×10⁵ a.u. in LS174T and 5.35×10⁵±2.94×10⁵ a.u. in CT26 tumors) compared to MB_Control injection (2.15×10⁵±1.34×10⁵ a.u. in LS174T and 1.73×10⁵±0.76×10⁵ a.u. in CT26 tumors).

Ex vivo analysis of tumor tissue

In responding tumors, the percentage area of blood vessels was significantly decreased in bevacizumab-treated (0.98%±0.22%) compared to saline-treated tumors (5.41%±1.06%, P=0.001). Similarly, VEGFR2 expression levels were significantly lower in bevacizumab-treated (218±41 a.u. fluorescence signal) compared to saline-treated tumors (1150±227 a.u., P=0.001; Figure 5).

Figure 5.

Summary of quantitative immunofluorescence data and representative tissue micrographs obtained from responding and non-responding tumors following anti-angiogenic and saline treatment. (A) Bar graphs show percentage area of blood vessels and VEGFR2 expression levels in responding and non-responding tumors with and without anti-angiogenic treatment. Error bar = standard deviation. * P = 0.001. (B) Representative micrographs of CD31-stained (outlining tumor vessels), VEGFR2-stained, and merged tissue slices from both responding and non-responding tumors, following anti-angiogenic treatment (T) and control saline treatment (C). Scale bar = 100 μm.

In contrast, in non-responding tumor, there were no significant differences in both the percentage area of blood vessels between bevacizumab-treated and saline-treated tumors. The percentage area of blood vessels was 4.84%±0.85% in bevacizumab-treated and 4.77%±0.98% in saline treated tumors (P=0.798). Likewise, VEGFR2 fluorescence signal was not significantly different (P=0.721) in bevacizumab-treated (1298.40±117.63 a.u.) compared to saline-treated tumors (1331.49±112.40 a.u.; Figure 5).

Discussion

Our results showed that VEGFR2-targeted 3DUSMI signal changes differently over the course of a 10-day anti-angiogenic treatment regimen in two colon cancer types simulating responding and non-responding tumors in mice. In responding tumors, VEGFR2-targeted 3DUSMI signal substantially decreased as early as 24 hours after treatment initiation, while in non-responding tumors there was no change at any time point. Furthermore, early change of VEGFR2-targeted 3DUSMIat day 1 was predictive of treatment response at day 10 while tumor volume was not predictive of treatment response.

Although clinical trials have shown that anti-angiogenic therapy, either alone or in combination with first-line chemotherapy, is beneficial in several cancers including colon cancer (5), lung cancer (31)and hepatocellular carcinoma (32), resistance to angiogenic inhibitors may occur; this can lead to disease recurrence, one of the major challenges associated with anti-angiogenic therapy (7,33). To simulate patients that developed resistance and those that respond to anti-angiogenic therapy in preclinical experiments, two different colon cancer cell lines were used in our study. The LS174T human colon cancer cell line which has been previously shown to be sensitive to the treatment of bevacizumab (18,34), was used to represent responding tumors in our study. As expected, bevacizumab treatment significantly inhibited the growth of LS174T colon cancer compared to saline-treated mice in our study. The CT26 murine colon cancer cell line was used to represent non-responding tumors in our study. Previous studies have shown that CT26 colon tumors are resistant to VEGF pathway blockade since tumor neoangiogenesis can occur via VEGF-independent mechanisms in CT26 tumors (19,20). Furthermore, bevacizumab, which is a humanized VEGF neutralizing monoclonal antibody, is species specific with no interaction with murine VEGF (35,36). A recent study showed no growth inhibition of bevacizumab in CT26 tumors in mice (37), which was confirmed in our study with tumor volumes increasing by 839% and 799%, respectively, at day 10 in both bevacizumab- and saline-treated tumors. The use of the two tumor types to simulate responding and non-responding colon cancers was further corroborated by quantitative immunofluorescence used as reference standard in our study. Treated LS174T tumors showed a substantial decrease in percentage area of blood vessels and fluorescence VEGFR2 signal while control CT26 tumors did not show significant changes following anti-angiogenic or control saline treatment.

Various imaging modalities including contrast-enhanced magnetic resonance imaging (38), computed tomography (39), positron emission tomography (40) and ultrasound (41) are currently being investigated for improved monitoring of therapy response in cancers treated with new drugs, in particularly molecularly targeted drugs, that have either no or only minimal changes in tumor sizes. Among those imaging modalities ultrasound is advantageous due to its wide availability and relatively low cost allowing to perform repetitive exams during the course of treatment, in particular during the early course of treatment, to assess whether early changes of imaging signal may be predictive of treatment response. With the use of molecularly targeted contrast microbubbles, 2D USMI has been shown to be feasible for monitoring tumor anti-angiogenic treatment (13,14). However, one of the major limitations of 2D USMI in monitoring tumor response compared to whole body imaging modalities such as CT, MRI, or PET is the limited field of view only representing a small fraction of the imaged tumor volumes. This can result in substantial sampling errors as shown previously for 2D USMI (15,17) due to the known spatial heterogeneity of the neovasculature in cancer with heterogeneous expression of angiogenic markers such as VEGFR2 across the tumor volume (42,43). This becomes particularly problematic when repetitive imaging is required during longitudinal assessment of treatment response. Therefore, 3DUSMI allowing for volumetric assessment of VEGFR2 expression across the entire tumor volume is critically needed to allow for a more robust longitudinal evaluation of treatment response in cancer using ultrasound.

Recently, 3DUSMI has been shown to be technically feasible and highly reproducible for assessing tumor angiogenesis with the use of a previous generation IU22 clinical US system (17). In that study (17), the importance of 3D capabilities in USMI has been highlighted with an average variability in signal change of on average 27% (ranging between 2% and 73%) between 3D and 2D USMI in tumors treated with anti-angiogenic therapy. In our study, we extend those initial findings and bring them into a more clinical context by investigating whether 3DUSMI could be used to monitor anti-angiogenic treatment response over several treatment cycles up to 10 days and whether differentiation between responding from non-responding tumors is possible based on the magnitude of volumetric VEGFR2-targeted USMI signal intensities using a clinical-grade VEGFR2-targeted contrast microbubble and the latest generation clinical EPIQ7 US system, further optimized for molecular ultrasound imaging. Our results showed that VEGFR2-targeted 3DUSMI signal was significantly decreased compared to baseline values in responding tumor as early as one day after initiation of bevacizumab treatment, while there was no significant changes in VEGFR2-targeted 3DUSMI in non-responding tumors. This early decrease in 3DUSMI signal was predictive of later time point treatment response assessed at day 10. We chose day 10 as treatment response endpoint as the longest time point since beyond that time both tumor types became substantially necrotic. Occurrence of tumor necrosis in large subcutaneous tumor models is a well described phenomenon (44,45). Since we used a clinical transducer primarily used for abdominal imaging in patients with a center frequency of 3.2 MHz, after tumor cell inoculation we waited until a critical tumor diameter of at least 6 mm was reached in mice. By day 10, most tumors reached a critical size beyond 20 mm, which mandated sacrificing animals per our institutional guidelines. Interestingly, at day 3 and at later time points, 3DUSMI signal spontaneously decreased in both tumor types (in LS174T tumors after saline treatment and in CT26 tumors after bevacizumab or saline treatment). This may be due to a combination of down regulation of VEGFR2 and evolving tumor necrosis once tumors grow beyond a certain size, as previously shown for subcutaneous ovarian, pancreatic, and breast cancers (46).

We acknowledge several limitations of our study. First, as mentioned above, we could only monitor treatment response for 10 days due to the fast tumor growth in mice and our model therefore cannot simulate and assess longer term treatment outcome assessment usually performed between 8 and 12 weeks after treatment initiation in cancer patients. Since such a long treatment cycle is challenging to simulate in preclinical animal models, future studies in patients are warranted to assess the prognostic value of early changes in VEGFR2-targeted 3DUSMI signal in terms of longer term treatment outcomes. Second, a cell line which is known to be resistant to VEGF pathway blockade was used to simulate clinical non-responders, which might not fully represent the clinical situation with tumor resistance evolving during anti-angiogenic treatment (7). Therefore, as a next step, clinical translational studies are needed to assess the potential of 3DUSMI to predict early treatment response in cancer patients. The VEGFR2-targeted contrast microbubble used in our study has recently received FDA investigational new drug (IND) approval and a Phase 2 clinical trial is ongoing in the USA to evaluate its safety and efficacy to assess VEGFR2 expression in cancer patients (47). Finally, for the 3DUSMI protocol we chose to wait 60 seconds after the destructive pulse to obtain post-destructive 3DUSMI data since in the colon cancer tumor models used in our study it takes that long for all tumor vessel to fully replenish based on our experience. This is substantially longer than the few seconds commonly used for 2D USMI and might have influences our absolute USMI signal values using targeted and control microbubbles in at least two ways: First, during the waiting time of 60 seconds, the targeted contrast agent had time to bind again to the molecular target; this may have increased the post-destructive pulse 3DUSMI signal, thereby decreasing the calculated signal. Second, since non-attached microbubbles are cleared via the reticular endothelial system, the signal from freely circulating microbubbles may have been further decreased compared to the pre-destructive pulse during the 60-second interval, which may have increased the calculated signal. Alternative strategies of quantifying USMI are currently being explored (48–50) that allow more real-time measurements of the targeted USMI signal without the need for a destructive pulse or a long enough waiting time until non-attached freely circulating microbubbles have fully cleared from the circulation. Such strategies are eventually needed when moving 3DUSMI into the clinics as high powered destructive pulses as used in our study for diagnostic purposes may cause unwarranted biological effects (51,52) and since real-time imaging without the need of post processing improves the workflow of USMI.

In conclusion, our study suggests that 3DUSMI using a clinical US system with a clinical 3D transducer and clinical grade VEGFR2-targeted contrast microbubbles allows longitudinal monitoring of anti-angiogenic treatment response in two colon cancer models simulating clinical responders and non-responders to anti-angiogenic therapy. Since US is widely available, relatively inexpensive, and does not expose patients to radiation, our study lays the foundation to further develop 3DUSMI for clinical translation to allow assessment of early treatment response at the molecular level in cancer patients.