Pharmacokinetic Modeling of Targeted Ultrasound Contrast Agents for Quantitative Assessment of Anti-Angiogenic Therapy: a Longitudinal Case-Control Study in Colon Cancer

Cancer is one of the leading causes of death worldwide and represents a major health challenge, accounting for 8.8 million deaths in 2015—about 16 % global mortality [1]. Among all cancers, colorectal cancer is particularly relevant, representing the third most prevalent and second most lethal cancer in man and women combined in the USA [2].

Based on the established link between cancer growth and the process of angiogenesis [3, 4], i.e., the formation of a vascular network supporting tumor development, novel therapeutic strategies aim at blocking or disrupting specific angiogenic pathways [5‐7]. Due to its ubiquitous overexpression, the vascular endothelial growth factor (VEGF) is the dominant target of antiangiogenic drugs. Currently, several VEGF inhibitors are approved for first and second lines of treatment of different types of cancers in the USA and Europe. In colon cancer, VEGF pathways represent one of the main targets for treatment of metastatic disease [6‐9].

Early evaluation of the therapeutic response is crucial to identify potential non-responders, allowing for better therapy tailoring and patient management. Current therapy assessment criteria based on survival time and tumor dimension are not suitable for evaluation of early response, especially in the case of novel antiangiogenic therapies, which act by interfering with angiogenic processes and may not lead to any change in tumor size [10]. Biomarkers for early assessment of antiangiogenic therapies are thus needed to improve therapeutic decision-making [11].

Activation of intratumoral angiogenesis through the VEGF pathway involves the phosphorylation of VEGF receptors overexpressed on tumor vasculature [12, 13]. Among these, the VEGF receptor 2 (VEGFR2) has been established as the main receptor in human malignancies [12], showing up to five times higher expression in tumor compared to normal vasculature [13]. Under the hypothesis that the overexpression of VEGF may mediate the upregulation of VEGFR2, imaging probes targeting VEGFR2 may represent a promising option for assessment of cancer therapies inhibiting the VEGF/VEGFR2 pathways [12‐15].

Portability, low cost, availability, and absence of ionizing radiations make ultrasound imaging a promising option for antiangiogenic therapy monitoring, whereby several repeated exams are needed. Ultrasound molecular imaging (USMI) of angiogenesis has become possible by the introduction of novel targeted ultrasound contrast agents (tUCAs) [16]. These are composed of microbubbles which can flow through the vasculature and attach to the vessel walls where the target molecule is overexpressed, thus causing selective enhancement in areas of active angiogenesis. In this context, the clinical grade tUCA BR55 targeting VEGFR2 was recently developed and tested for human use [17‐19].

Evaluation of the degree of microbubble binding has shown to be a promising biomarker of angiogenesis [14, 20‐22]. Semi-quantitative assessment is typically achieved by evaluating the late enhancement, several minutes after injection. Especially in preclinical studies, a more quantitative evaluation has been achieved by application of a high-pressure ultrasound burst to calculate the differential targeted enhancement (dTE), i.e., the difference in the image intensity before and after microbubble destruction [14, 16, 21, 23‐26]. Semi-quantitative USMI, however, is user- and machine-dependent, it requires lengthy procedures (~ 5–10 min in animals), and when a destructive burst is applied, it raises concerns for damages to the endothelial tissue [27].

Quantitative assessment may overcome these limitations. Several mathematical models have been proposed to describe the tUCA kinetics, which are based either on purely empirical models [28, 29] or on the combination of physiological and empirical models [30] or on pharmacokinetic modeling [31, 32]. Fitting these models to time-intensity curves (TICs) measured with USMI enables the estimation of quantitative parameters related to cancer angiogenesis. In this context, the first-pass binding (FPB) model enables characterization of microbubble binding by the estimation of the binding rate (k_b) [32, 33]. By focusing only on the first pass of the contrast bolus, this method is not affected by potential inaccuracies due to contrast recirculation and enables reducing the required USMI acquisition time to about 1 min.

The purpose of this study was to evaluate semi-quantitative and quantitative USMI for assessment of the early response to antiangiogenic treatment on colon cancer-bearing mice monitored during therapy.

In this study, we evaluated semi-quantitative and quantitative USMI for assessment of the response to antiangiogenic treatment in two colon cancer mouse models, simulating clinical responders and non-responders. Our results show the ability of USMI biomarkers to assess the response to treatment earlier than by assessment of tumor volume. A significant decrease in the proposed quantitative parameter k_b was observed as early as 1 day after treatment, earlier than by all other methods. This suggests USMI to represent a better option for assessment of the early response to therapy than traditional dimension-based criteria, and possibly also suited for assessment of therapies with cytostatic action, whereby no changes in tumor size are expected.

As tumor growth is supported by angiogenesis, stable or increased values of the USMI parameters can be expected in non-responder mice and in responders treated with saline. This is confirmed by our results, showing no significant changes in the USMI parameters, except for the responders treated with bevacizumab. However, non-monotonic variations were observed for the semi-quantitative parameters LE and dTE, suggesting lower reproducibility with respect to quantitative assessment by k_b.

Other imaging modalities are under investigation for monitoring the response to antiangiogenic therapies, including computed tomography (CT) [38], positron emission tomography (PET) [39], and magnetic resonance imaging (MRI) [40]. Although DCE-CT, PET/CT, and MRI have shown to be useful in detection and staging of colorectal cancer and metastases, especially in the preoperative settings [41‐43], the use of these modalities for therapy monitoring is hampered by the inherent radiation risk (CT, PET) and by the high cost and low availability (MRI, PET). Combining low cost, widespread availability, portability, and absence of ionizing radiation, USMI is advantageous for therapy monitoring, whereby repeated exams are performed. However, clinical translation of USMI requires more extensive clinical validation, evidencing the need for quantitative biomarkers of therapy response and for a standard clinical protocol to enable reliable evaluation and comparison of findings.

In this study, quantitative analysis was performed by fitting the FPB model to USMI-derived TICs. This model is the solution of a bi-compartmental model describing the concentration of free microbubbles as resulting from a convective dispersion process, and the concentration of bound microbubbles as resulting from a well-mixed compartment, where binding occurs at a rate given by k_b. Since only the first pass of the contrast bolus is considered, the assumption of negligible unbinding can be made, and any effect due to tUCA recirculation can be disregarded. Moreover, fitting the tUCA first pass avoids the need for long acquisition times and for the application of a high-pressure destructive burst, which are instead required by current semi-quantitative methods. Although more complex modeling would permit relaxing these assumptions and fitting the whole curve, the increased number of free parameters may result in inaccurate parameter estimation [3].

In-vitro studies have shown very low microbubble detachment also in high-flow and high-shear stress conditions [44], supporting the assumption of negligible unbinding. Moreover, (un)binding kinetics in the order of min⁻¹ were reported for the surface size of interest (~ 0.002 mm²) [45], confirming the validity of the adiabatic approximation. Regarding the first-pass assumption, although 1-min recirculation time is appropriate in humans, this might not apply to small animals, for which a recirculation time of about 15–20 s can be expected [46, 47]. However, considering that the actual appearance time in the measured TICs was typically larger than 20 s (Fig. 6) and that the tUCA concentration in the second and later bolus passes is typically much lower [35], the first-pass assumption might still represent a valid approximation. A preliminary in-vivo validation of the method performed in rats, comparing k_b estimates obtained for targeted and non-targeted UCA, suggests the model to accurately describe the kinetics of both [48]. However, simulation studies should be performed in the future to evaluate the accuracy, precision, and repeatability of parameter estimation, and its sensitivity to the abovementioned underlying assumptions; also, the signal-to-noise and temporal sampling requirements of the method could be tested in silico, possibly proposing an optimized acquisition protocol for improved performance.

Although angiogenesis inhibitors have been used successfully as first- and second-line treatment in some tumor types [7, 8, 49, 50], resistance has been reported in some cases [49, 51, 52]. Early distinction of clinical non-responders is therefore crucial to permit timely adjustments in the therapeutic strategy, possibly improving treatment efficacy, sparing patients the morbidity and severe side effects associated with antiangiogenic therapies [53] and potentially cutting clinical costs due to unnecessary treatment. In our study, comparison of treated responders and non-responders showed USMI parameters to be able to predict the response to therapy already 1 day after treatment initiation. However, only one tumor model responding to treatment and one tumor model showing resistance to the investigated antiangiogenic inhibitor were here compared. Moreover, as the investigated drug and imaging probe targeted the same angiogenic expressions and only short-term monitoring up to 10 days after treatment was here performed, late resistance due to the development of alternative angiogenic pathways could not be investigated [49, 51, 52]. Further preclinical validation, involving different organs, tumor models, and antiangiogenic drugs, and comparison with perfusion assessment and long-term survival criteria are thus necessary to clarify the role of USMI for early prediction of the therapeutic response.

The ability of the proposed in-vivo USMI parameters to reflect angiogenesis was validated by comparison with the ex-vivo immunohistological assessment of VEGFR2 expression levels and the percentage blood vessel area, performed on excised tumors. Excluding the rank correlation between LE and the percentage blood vessel area (ρ_s = 0.43, p-value = 0.09), significant linear and rank correlation was found between all in-vivo and ex-vivo angiogenesis biomarkers. In general, higher correlation was found with the VEGFR2 expression levels compared to the percentage blood vessel area. This may be expected considering that the adopted in-vivo biomarkers all reflect the binding levels of VEGFR2-targeted microbubbles and thus are more directly related to VEGFR2 expression levels than the percentage blood vessels area, which reflects more structural features of angiogenic vasculature. Moreover, the lower linear correlation found between VEGFR2 expression for k_b, compared to LE and dTE, may be due to the fact that k_b, representing the microbubble binding rate, does not reflect only the degree but also the kinetics of microbubble binding. Further validation in different tumor models may provide greater insight on whether the binding kinetics vary in different tumor types.

Similar conclusions can be drawn from the correlation analysis within the USMI parameters, which showed higher correlation between the semi-quantitative parameters LE and dTE, than between k_b and LE/dTE. This suggests that k_b may provide different insights into microbubble binding, while the information given by LE and dTE may be overlapping; this might also explain the surprising result by the FPB-fitting approach, performed only on the first minute and yet providing earlier prediction than late enhancement. In fact, by reflecting the binding kinetics, the quantitative parameter k_b may represent a more accurate biomarker of angiogenesis as compared to LE and dTE, whose accuracy may be affected by the risk of late bubble detachment and limited by the technical difficulty to obtain a stable concentration plateau before and after the destructive burst. Since k_b showed lower correlation with LE than dTE, in future work, it might be interesting to investigate whether improved performance can be obtained by the combination of k_b and LE, which would provide quantitative analysis with an easier and safer acquisition protocol, not requiring the application of a destructive US burst. Information about the LE, for instance, could be exploited for a smarter initialization of k_b, possibly improving estimation convergence and accuracy.

There are some limitations in this study. Although lacking a clear physiological link with the underlying angiogenic processes, other empirical and pharmacokinetic models are available to fit USMI-derived TICs [28‐31]. A comparative study should be performed in future work to assess which model and parameters are most suitable for quantitative USMI of cancer angiogenesis. Also, no comparison was here performed with other US methods such as perfusion assessment by conventional contrast-enhanced ultrasound, which has shown promise for early assessment of the response to therapy and good correlation with dimension-based criteria and overall survival [54]. In previous studies, USMI showed ability to detect treatment response much earlier than by assessment of perfusion and tumor volume [15]; moreover, USMI showed higher correlation with VEGFR2 expression levels [14], an established prognostic biomarker of cancer aggressiveness [55‐58]. However, since antiangiogenic therapy is mostly used as second-line treatment in conjunction with chemotherapy [6, 8], future research should clarify on the different information provided by molecular and perfusion assessment in relation to the response to combined therapies. Moreover, although the adopted FPB model has shown promise for antiangiogenic therapy monitoring by the quantitative parameter k_b, this study was performed on two mouse xenograft models, for which tumor biology is inherently different than that of humans, and on a limited dataset, with the largest group including five mice. More extensive preclinical validation and feasibility studies in humans are thus necessary to confirm the promising results.

Our findings contribute to the cohort of preclinical studies showing the promise of VEGFR2-targeted microbubbles in the context of angiogenesis imaging and therapy monitoring [14, 19‐21, 24]. Currently, no molecularly targeted contrast agent has been approved for clinical use. Initial studies in humans have demonstrated the feasibility and clinical safety of VEGRF2-targeted microbubbles for USMI of prostate [18], breast, and ovarian [19] cancer. However, more extensive validation and the implementation of multi-center studies are required to allow clinical translation. In this context, a standardized quantitative protocol may be useful to improve reproducibility and to facilitate the comparison of findings between different studies and centers, especially important for therapy monitoring, whereby several longitudinal measurements need to be compared. Based on the results of our study, here we suggest the combination of the (semi)quantitative parameters LE and k_b for a standardized quantification protocol feasible for clinical USMI in the context of antiangiogenic therapy monitoring.