Ultrasound localization microscopy of renal tumor xenografts in chicken embryo is correlated to hypoxia

Abstract

Ultrasound localization microscopy (ULM) permits the reconstruction of super-resolved microvascular images at clinically relevant penetration depths, which can be potentially leveraged to provide non-invasive quantitative measures of tissue hemodynamics and hypoxic status. We demonstrate that ULM microbubble data processing methods, applied to images acquired with a Verasonics Vantage 256 system, can provide a non-invasive imaging surrogate biomarker of tissue oxygenation status. This technique was applied to evaluate the microvascular structure, vascular perfusion, and hypoxia of a renal cell carcinoma xenograft model grown in the chorioallantoic membrane of chicken embryos. Histological microvascular density was significantly correlated to ULM measures of intervessel distance (R = -0.92, CI95 = [-0.99,-0.42], p = 0.01). The Distance Metric, a measure of vascular tortuosity, was found to be significantly correlated to hypoxyprobe quantifications (R = 0.86, CI95 = [0.17, 0.99], p = 0.03). ULM, by providing non-invasive in vivo microvascular structural information, has the potential to be a crucial clinical imaging modality for the diagnosis and therapy monitoring of solid tumors.

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1

Vascular imaging of CAM-engrafted Renca tumors. (A) The ex ovo chicken embryo model has readily accessible surface vasculature, facilitating the injection of precise volumes of microbubble contrast agent. Renca cells engrafted into the CAM surface result in spheroidal tumor masses that are consistently hyper-vascularized. (B) The spheroidal Renca tumors were imaged with ultrasound at five distinct imaging planes to volumetrically sample these renal tumor xenografts. Each imaging plane was imaged using 15 angle plane-wave compounding with a Vantage 256 system (L35-16vX transducer), for a total acquisition length of 3600 frames (7.2 s). (C) B-mode image of example tumor, generated by compounding plane-wave acquisitions, demonstrates a high SNR and well characterized tumor margin. (D) Conventional contrast-enhanced ultrasound image of example tumor confirms the high degree of vascularization typical of CAM-engrafted Renca tumors. Nominal imaging resolution was 50 µm axial and 100 µm lateral. (E) Fluorescent histology section, taken from the example tumor in (D), demonstrates both a high peripheral tumor vascularization (red signal) as well as intratumoral hypoxia (green signal).

Figure 2

Microbubble localization and tracking. (A) IQ data (3600 frames) was represented as a three-dimensional matrix by stacking each B-mode image frame together along a ‘slow-time’ dimension. (B) The resulting imaging volume was reshaped into a 2D Casorati matrix in a column-wise manner, and a singular value decomposition (SVD) was used to remove the highly spatiotemporally coherent tissue background (low singular value orders). (C) An inverse SVD was performed to recover a three-dimensional imaging volume that contained predominantly microbubble signals. A microbubble PSF was convolved with each frame of the dataset shown in (C) to identify the positions of potential microbubble locations. (D) Representative image stack of localized microbubble centroids, after a local maximum search. There are some false positive microbubbles identified in this first pass of localization processing. (E) Bipartite graph-based filtering was used to pair microbubbles along several frames of data. This process reduces the number of false positive microbubble detections by eliminating low-confidence events. (F) Representative microvascular structure map from an example tumor. (G) Blood-flow velocity map for the same representative tumor.

Figure 3

Comparison between diffraction limited CEUS and ULM (A) Contrast-enhanced power images demonstrate that CAM-engraft Renca tumors are highly vascularized through-out the entire tumor volume. (B) Contrast-enhanced color-flow imaging exhibited a mosaic pattern of alternating blood flow velocity directions, with the majority of faster flow typically present in the tumor periphery. (C) Super-resolution microbubble localization maps demonstrated improved resolution of small vessels, while retaining the vascular features of the larger feeding vessels. (D) The microvessel blood-flow velocity images disambiguated some of the velocity information present in the conventional color-flow images, particularly for smaller vessels. The interconnectedness of flow direction is more apparent. (E) Both conventional power images and super-resolution localization maps demonstrated similar trends in their estimate of blood volume for each tumor, with a high significant correlation (R = 0.83, p = 8.7 × 10−7, N = 30, Spearman correlation) (F) Likewise, the correlation between the conventional estimate of FMBV and ULM vascular density was significant, but only moderate in magnitude (R = 0.59, p = 0.0006, N = 30, Spearman correlation) (G) Estimates of mean flow velocity between the two image processing techniques were not found to be significant (R = 0.036, p = 0.85, N = 30, Spearman correlation).

Figure 4

Intratumoral heterogeneity of CAM Renca tumors. (A) Tumor microvessel density maps demonstrate a phenotype of lower microvascular density in the center of the tumor, with most of the large feeding vasculature found in the tumor periphery. (B) ULM blood flow velocity images corroborate these findings, with slower blood flow velocities found in the center of the tumor and a characteristic mosaic patterning of multidirectional blood flow directions. (C) Hypoxyprobe signal demonstrated that the center of this tumor is hypoxic, but showed that the top-most region was well oxygenation (arrow). This may be due to oxygen diffusion from the atmosphere (Supplementary Fig. 1). (D) The tortuosity of the tumor microvasculature was calculated using the Distance Metric and Sum of Angle Metric. (E) Boxplot of the rhodamine MFI, a histological measure of microvascular density. (F) ULM intervessel distance quantifications. (G) Boxplot of hypoxyprobe MFI taken as a quantified measure of intratumoral hypoxia. (H) The Distance Metric for this tumor cohort.

Figure 5

Correlation to fluorescent histology. Correlation matrix of histological measurements in comparison to the vascular metrics derived from ULM imaging (Pearson’s correlation coefficient (R)). Notable findings include a significant negative correlation between rhodamine and ULM intervessel distance (R = −0.92, p = 0.01, N = 6), and a significant positive correlation between hypoxyprobe signal and the Distance Metric (R = 0.86, p = 0.03, N = 6). Significant correlations are denoted with asterisk (*p