Vascular Profile Characterization of Liver Tumors by Magnetic Resonance Imaging Using Hemodynamic Response Imaging in Mice^,

Abstract

Recently, we have demonstrated the feasibility of using hemodynamic response imaging (HRI), a functional magnetic resonance imaging (MRI) method combined with hypercapnia and hyperoxia, for monitoring vascular changes during liver pathologies without the need of contrast material. In this study, we evaluated HRI ability to assess changes in liver tumor vasculature during tumor establishment, progression, and antiangiogenic therapy. Colorectal adenocarcinoma cells were injected intrasplenically to model colorectal liver metastasis (CRLM) and the Mdr2 knockout mice were used to model primary hepatic tumors. Hepatic perfusion parameters were evaluated using the HRI protocol and were compared with contrast-enhanced (CE) MRI. The hypovascularity and the increased arterial blood supply in well-defined CRLM were demonstrated by HRI. In CRLM-bearing mice, the entire liver perfusion was attenuated as the HRI maps were significantly reduced by 35%. This study demonstrates that the HRI method showed enhanced sensitivity for small CRLM (1–2 mm) detection compared with CE-MRI (82% versus 38%, respectively). In addition, HRI could demonstrate the vasculature alteration during CRLM progression (arborized vessels), which was further confirmed by histology. Moreover, HRI revealed the vascular changes induced by rapamycin treatment. Finally, HRI facilitates primary hepatic tumor characterization with good correlation to the pathologic differentiation. The HRI method is highly sensitive to subtle hemodynamic changes induced by CRLM and, hence, can function as an imaging tool for understanding the hemodynamic changes occurring during CRLM establishment, progression, and antiangiogenic treatment. In addition, this method facilitated the differentiation between different types of hepatic lesions based on their vascular profile noninvasively.

Introduction

Hepatic tumors represent a major burden in the medical practice, considering their high incidence, the diagnostic difficulties, and the high rate of morbidity and mortality. Hepatocellular carcinoma (HCC) is the fifth most common cancer worldwide and the third most common cause of cancer-related mortality [1]. Nowadays, early stage HCC diagnosis is feasible in 30% to 60% of cases, thus enabling the application of curative strategies; however, timely and accurate diagnosis is of paramount importance [2]. The liver is also one of the most commonly involved organs in metastatic disease [3]. Colorectal cancer is the third most common cancer in the United States [4]. The development of colorectal liver metastases (CRLMs) is the main cause of death in these patients [5].

Early diagnosis is critical for successful resection and for refined treatment selection criteria in both liver pathologies [6,7]. Only a minority of CRLM patients (<20%) are amenable to resection [8–10]. Thus, alternative treatments are under investigation. Recently, considerable research has focused on the search for antiangiogenic therapies for CRLM. The immunosuppressant rapamycin (RAPA) is a highly specific inhibitor of mTOR that can further inhibit tumor growth [11], tumor progression [12], and metastasis [13] both through its antiangiogenic activity (by impairing the production of vascular endothelial growth factor [VEGF]) and direct inhibition of cell proliferation and cell cycle progression [14,15]. Tumor response to therapy is usually assessed by measurements of tumor size using morphologic imaging techniques. Because antiangiogenic therapies may not lead to substantial tumor mass reduction, their effect is better imaged using techniques designed to assess vascular function rather than the conventional measurements of tumor size [16–18]. Dynamic contrast-enhanced magnetic resonance imaging (CE-MRI) has been used previously to examine the acute and chronic effects of VEGF signaling inhibitors, both preclinically [19–21] and clinically [22,23]. However, the physiologic significance of these parameters can be complex, and there is currently no consensus as to the best parameters to be used or the most appropriate measurement and analysis method [24,25].

It is well known that, whereas the normal liver is supplied predominantly by the portal vein, in patients with overt hepatic tumor, a higher proportion of liver blood flow is derived from the hepatic artery [26,27]. Moreover, even small or occult lesions may lead to subtle changes in liver blood flow [28,29]. Therefore, monitoring hemodynamical changes using perfusion imaging may facilitate the characterization of the vascular profile, which can lead to earlier and more accurate detection of hepatic tumors. This strategy was recently endorsed by the European [30] and American Associations for the Study of the Liver [31].

Today, to acquire perfusion images, intravenous administration of a contrast agent is necessary. Multiphasic CE-MRI is currently accepted as a reliable method for detecting and characterizing liver tumors [32]. A good separation of arterial from portal venous phases requires short acquisition time, which leads to low spatial resolution or partial volume coverage. Recently, we demonstrated the feasibility of hemodynamic response imaging (HRI), a functional MRI (fMRI) method combined with hypercapnia and hyperoxia, for monitoring changes in liver perfusion and hemodynamics without the need of contrast agent administration [33,34]. We established its ability to image the hemodynamic changes occurring under different pathologic states such us liver fibrosis, acute bleeding, and during liver regeneration [34,35].

In the present study, we aimed to assess the use of HRI for the detection and characterization of the early vascular and perfusion changes occurring during CRLM establishment, for the study of the vascular changes in CRLM during antiangiogenic therapy, and for the assistance to characterize the vascular profile of primary hepatic tumors. We used the CT-26 murine colon carcinoma liver metastatic mouse model [36] and the Mdr2 knockout mice [37] as the inflammationinduced primary liver tumor model. HRI utility for liver tumor diagnosis and vascular characterization was assessed in these animal models. The HRI results were compared with CE-MRI andwere further confirmed by histology. HRI utility for antiangiogenic effect assessment was evaluated on CRLM-bearing mice that were treated with RAPA at a relevant dose. In this study, the HRI method showed high sensitivity for the vascular changes occurring in liver tumors in experimental animal models.

Materials and Methods

Animals

For the CRLM model, 7- to 8-week-old male CB6F1 mice were used (31 mice). For the HCC model, 12- to 18-month-old Mdr2 knockout mice [37] were analyzed (21 lesions in 12 mice). All experiments were performed in accordance with the guidelines and approval of the Animal Care and Use Committee of the Hebrew University, which holds National Institutes of Health approval (OPRR-A01-5011).

Mouse Model of CRLM

CT-26 murine colorectal adenocarcinoma cells [36] were injected intrasplenically to anesthetized CB6F1 mice (10⁴ cells in 300 εl per mouse). After 5 minutes, the spleens were removed, allowing the cells to enter the portal circulation and to initiate liver metastases. In this model, one to five hepatic nodules per mouse were detected by MRI 13 to 17 days after cell inoculation. In this animal model, the tumors appeared hyperintense in T₂-weighted (T₂W) images (Figure 1A). The presence of CRLM lesions was further verified by repeated MRI scans and histologic analysis. For HRI assessment, 47 lesions (1–5 mm in diameter) were examined, and for CE-MRI assessment, 42 lesions were analyzed. For antiangiogenic treatment surveillance, eight CRLM-bearing mice were treated with RAPA (Fermentek Ltd, Jerusalem, Israel) at a dosage of 2 mg/kg per day intraperitoneally. The treatment started on the day of tumor appearance as detected by MRI.

Figure 1.

CRLM vasculature profile characterization by HRI: Representatives axial T₂W images (top) and ΔSo₂ (hyperoxia effect—center row) and ΔSco₂ (hypercapnia effect—bottom) maps of healthy liver (A), small CRLM acquired during early phase (B, red arrow), and advanced multifocal CRLM (C). Bar = 1 cm. Color scales for ΔS maps are located on the right. The HRI method clearly highlighted CRLM from healthy liver tissue. (D) Mean ΔSo₂ and ΔSco₂ values ± SD of healthy livers (n = 10) versus livers of CRLM-bearing mice (n = 10) and CRLM (n = 30). *P < .001.

Magnetic Resonance Imaging

MRI experiments were performed using a horizontal 4-7-T Biospec spectrometer (Bruker Medical, Ettlingen, Germany) with a 3.5-cm birdcage coil. Mice were anesthetized with pentobarbital (30 mg/kg, intraperitoneally) and placed supine. Coronal and axial T₁-weighted (T₁W) spin-echo images were acquired for liver segmentation purposes (repetition time = 360 milliseconds, echo time = 18 milliseconds). Tumor assessment was done using T₂W fast spin-echo images (repetition time = 2000 milliseconds, echo time = 37 milliseconds, in-plane resolution = 117 εm, slice thickness = 1 mm).

Hemodynamic Response Imaging. Changes in hepatic hemodynamics were evaluated using the HRI protocol as previously described [33]. In brief, the images were acquired using T₂*-weighted gradient echo images (repetition time = 147 milliseconds, echo time = 10 milliseconds, field of view = 3 cm, in-plane resolution = 117 εm, slice thickness = 1 mm, 2 averages, 37 seconds per image) combined with breathing of air, air-CO₂ (5%CO₂, 4 L/min), and carbogen (95% O₂ + 5% CO₂, 4 L/min) through a homemade mask.

Contrast-Enhanced MRI. To compare HRI results to a standard method, CE-MRI was performed with T₁W FLASH sequence (repetition time = 34 milliseconds, echo time = 5 milliseconds, field of view = 3 cm, in-plane resolution = 117 εm, slice thickness = 1 mm, 1 average, 80 measurements), resulting in a temporal resolution of 4 seconds. After the 10th image, gadolinium-diethylene-triaminepenta-acetate (Gd-DTPA, Magnetol; Soreq Radiopharmaceuticals, Yavne, Israel; 0.5 M, 100 εl) was administered through the tail vein at a dose of 0.1 mmol Gd/kg. Signal intensity-time curves and multiphasic CE images were calculated using an in-house program written in IDL (ITT Visual Information Solutions, Boulder, CO) where hepatic arterial phase begins 4 seconds after injection and the portal venous phase begins 16 seconds after injection.

Ultrasound

Ultrasound measurements were acquired with a 14-MHz linear transducer (15L8s) (Sequoia-512; Acuson, Mountain View, CA) on anesthetized mice (pentobarbital 30 mg/kg, intraperitoneally). Tumor perfusion was assessed by CE ultrasound, by intravenous injection of 150 εl of saline through the tail vein and, 5 minutes later, by intravenous injection of 150 εl of contrast medium bolus (15 mg/mouse, Definity Perflutren Lipid Microsphere; Bristol-Myers Squibb Medical Imaging, Inc, Billerica, MA). Contrast enhancement was measured for 15 seconds at a rate of 20 frames/second. For CE ultrasound analysis, the enhancement of each region of interest (ROI) was calculated from time series images using dedicated functional molecular image analysis software UIA (I-Labs, Petah-Tikva, Israel).

Image Analysis and Statistics

The number of tumors per liver and their volume assessment was performed by using Analyze 7.0 (BIR, Mayo Clinic, Rochester, MN) from the T₂W images. HRI maps were generated as reported previously [33,34] using IDL (Interactive Data Language of ITT Visual Information Solutions). For healthy mice, the selected liver ROIs covered the entire liver. For CRLM-bearing mice, tumor ROI included the entire lesion and liver ROI included representative liver tissue far away from the detected tumors as defined on the T₂W images by using the Analyze 7.0 software. Mean ΔS values were calculated from these ROI, and results are expressed as means ± SD. The difference between groups was analyzed by one-sided exact paired Student's t test for n > 30 data points and with one-sided exact Wilcoxon signed-rank test for a smaller sample size. Statistical analyses were performed with the Instat Biostatistics software (GraphPad Software, Inc, San Diego, CA). P < .05 was considered statistically significant.

Histology and Immunostaining

Mice were killed immediately after the final MRI scan, and their livers with the surrounding tissues were fixed in 4% formaldehyde for a week; the entire liver was further embedded in paraffin to preserve good correlation to MRI orientation. Finally, the livers were sliced at 1.1-mm intervals in the same axial plane as the MRI sections. The nodules that precisely corresponded to HRI (using the transverse T₂W images as a reference) were determined. These were then cut (5 µm) and stained with hematoxylin-eosin (H&E) or subjected to immunohistochemistry using anti-PECAM-1 antibodies (CD31; Biocare Medical, Concord, CA) and antibody against α-smooth muscle actin (α-SMA) (dilution 1:300; Sigma, St Louis, MO). Histologic examination was conducted by an expert hepatopathologist with more than 10 years of experience in liver pathology. The number of blood vessels was counted in 10 randomly selected high-power microscopic fields (HPF, magnification x400) for each tumor, and the mean value ± SD was determined.

Results

CRLM Vasculature Profile Characterization by HRI

Initially, we assessed the ability of HRI to detect the vascular and perfusion changes occurring in mice with well-defined CRLM. In healthy mice, the mean liver ΔSo₂ values (the change induced by hyperoxia) were extremely positive (95% ± 18%, n = 10 livers), whereas the mean ΔSco₂ values (the change induced by hypercapnia) were negative (-40% ± 4%, n = 10 livers; Figure 1, A and D), in agreement with the results obtained previously in rats [33]. In all the CRLM-bearing mice, the entire liver perfusion was reduced as the HRI reactivity maps for both hyperoxia and hypercapnia were significantly attenuated (62% ± 35% and -29% ± 9%, respectively, P < .001) compared with healthy liver mean values (n = 10 livers; Figure 1, B–D). Moreover, the ΔSo₂ and ΔSco₂ mean values of well-defined CRLM were significantly reduced (21% ± 13% and -10% ± 7%, respectively, P < .001) compared with those of the adjacent liver (n = 30 tumors; Figure 1, B–D). These results may reflect hypovascularity of the CRLM nodules, with increased arterial blood supply.

CRLM Vasculature Profile Verification

Perfusion assessment methods and immunohistochemical staining were applied to evaluate vessel density and blood supply distribution of CRLM in this animal model. The hypovascularity of CRLM was clearly evident by their pale appearance compared with the reddish color of the adjacent liver parenchyma (Figure 2A). Furthermore, blood vessel quantization was achieved by immunohistochemical staining, with anti-PECAM-1 antibody (CD31), which confirmed the lower vessel density in CRLM compared with the dense vasculature of the liver parenchyma (Figure 2, C and D). The actual blood vessel count per HPF in healthy livers was significantly higher (35 ± 7 vessels/HPF, n = 8 livers) compared with CRLM (8 ± 3 vessels/HPF; n = 8 tumors, P < .0005). The increased arterial blood supply to CRLM in this model was assessed by using CE ultrasound (n = 3 mice). The maximum signal intensity of CRLM was lower and observed earlier than the maximum signal intensity of liver parenchyma (Figure 2E). Finally, by stopping Evans Blue perfusion during the arterial phase (n = 4 mice), we observed that most of the CRLM (>1 mm) turned blue, indicating their increased arterial blood supply (Figure 2B). These results indicate that CRLMs (>1 mm) in this animal model are indeed hypovascular and derive their vascular supply predominantly from hepatic arteries as suggested by HRI. These results from the animal model are in good agreement with the clinical knowledge regarding liver metastasis vascular properties in humans [26,27].

Figure 2.

The hypovascularity and increased arterial blood supply of CRLM. (A) A whole liver photograph demonstrating CRLM hypovascularity (arrow). (B) Evan's Blue (EB) perfusion was stopped during the arterial phase demonstrating the increased arterial blood supply to CRLM (arrows). Endothelial immunohistochemical staining with PECAM-1 (CD31; brown stain) of healthy liver section (C) and CRLM section (D) confirmed the lower vessel density of CRLM (original magnification, x400). (E) CE ultrasound enhancement curves from representative CRLM-bearing mouse (liver—black, CRLM—red and aorta—green; injection time is indicated on the graph). The time-intensity curves of each ROI were normalized by dividing each time point by the mean of the baseline block.

HRI and CE-MRI Comparison

The HRI results were further compared withmultiphasic CE-MRI in this model. For CRLM with diameters ranging between 2 and 5 mm, the sensitivity of both imaging methods was high. However, the HRI method enabled detection of smaller lesions (1–2 mm) compared with the CE-MRI method. Figure 3 demonstrates an example of an image with three small foci. According to the arterial phase image and to the signal intensity-time course of CRLM versus liver, only one of them (marked with red arrow; 2. 2 mm in diameter) was enhanced, whereas the smaller lesions in this image (marked with green arrows; 1.6 mm in diameter) were not enhanced with Gd-DTPA (Figure 3, E and G). The HRI results of the same slice clearly delineated all three lesions (Figure 3, B and C). These results were repeated in an additional seven mice in which approximately 60% of the tumors smaller than 2 mm were not enhanced with CE-MRI.

Figure 3.

CRLM detectability by HRI and CE-MRI: Results from a representative mouse with several small CRLM that were analyzed with both HRI and CE-MRI. (A) Axial T₂W image with three visible CRLM (marked with arrows; bar = 1 cm). The corresponding HRI maps of the same slice (B—ΔSo₂ map; C—ΔSco₂ map). Color scales for ΔS maps are located on the left. All three lesions were clearly visible with HRI. The multiphasic CE T₁W images that were obtained before (D) and 13.2 seconds (E; arterial phase) and 26.4 seconds (F; portal phase) after Gd-DTPA injection. Only the largest CRLM (marked with red box; 2.2 mm in diameter) was enhanced during the arterial phase, whereas the smaller lesions (marked with green boxes; 1.6 mm) were not enhanced during the arterial phase. (G) CE-MRI signal intensity-time curves measured from the same mouse of liver parenchyma (black), the enhanced CRLM (red), and the unenhanced CRLM (green). The injection time is indicated on the graph.

We further extended the sample size by including additional mice that were scanned with only one of the perfusion methods (either HRI or CE-MRI). The sensitivity results of lesion-by-lesion analysis for both HRI and CE-MRI are given in Table 1. In this animal model, the sensitivity for small CRLM detection (1–2 mm in diameter) of CE-MRI (n = 21 lesions) was 38%, whereas the sensitivity of the HRI method was higher 82% (n = 17 lesions). For larger CRLM lesions (between 2 and 5 mm in diameter), the sensitivity improved considerably for both methods, whereas CE-MRI showed a sensitivity of 81% (17/21); the HRI sensitivity was 97% (29/30).

Table 1.

Sensitivity Results of Lesion-by-Lesion Analysis.

CRLM Diameter (mm)	No. Detected (True Positive)	No. Missed (False Negative)	Sensitivity (%)
(A) HRI results
1 ≤ t < 2 mm	14	3	82
2 ≤ t < 5 mm	29	1	97
(B) CE-MRI results
1 ≤ t < 2 mm	8	13	38
2 ≤ t < 5 mm	17	4	81

Assessment of the Vascular Changes during Advanced CRLM Progression

To characterize the changes in tumor perfusion during CRLM progression, advanced CRLM (≥7 mm in diameter) were assessed by HRI (n = 6 tumors). Along with tumor growth, the center of these tumors appeared hyperintense in T₂W images, an appearance that could reflect development of necrosis (Figure 4A). When we analyzed the corresponding HRI maps, ΔSo₂ values were significantly positive at the center of these tumors (37% ± 13%, P < .01) compared with the outer CRLM region values (Figure 4, B and D), suggesting increased vascular density and blood content. Moreover, the ΔSco₂ values became significantly negative (-27% ± 6%, P < .01) compared with the outer CRLM region values (Figure 4, C and D). In addition, the HRI values at the center of these tumors were also significantly different compared with those of the adjacent liver parenchyma HRI (P < .05; Figure 4D). Indeed, H&E staining and immunohistochemical staining with anti-PECAM-1 antibody (CD31) and α-SMA antibody revealed highly arborized and widened matured blood vessels at the center of these tumors, reinforcing the HRI findings regarding increased vascularity at the center of these advanced tumors (Figure 4, E–G).

Figure 4.

HRI analysis of advanced CRLM. (A) Axial T₂W image of a mouse with advanced CRLM (encircled, 25 days after cell injection; bar = 1 cm) and the corresponding HRI maps (B—ΔSo₂ map; C—ΔSco₂ map) of the same slice. Color scales for ΔS maps are located on the left. The center of the tumor showed atypical HRI reactivity maps. (D) (top) Mean ΔSo₂ and ΔSco₂ values ± SD of healthy liver parenchyma (black), of central regions in advanced CRLM (pink), and of the outer regions in advanced CRLM (red) (n = 6 tumors): *P < .01 and **P < .05. (bottom) Representative HRI time courses obtained from this mouse, of liver tissue (black), from the center of advanced CRLM (pink), and of the outer CRLM region (red). Relevant histologic slides were stained with H&E staining (E), and immunohistochemical staining with PECAM-1 (F; brown stain) for endothelial cells and with α-SMA (G; red stain) for smooth muscle cells revealed highly arborized matured blood vessels (original magnification, x200) at the center of these tumors.

Assessment of the Vascular Changes during RAPA Therapy

To assess the use of HRI for antiangiogenic treatment evaluation, we treated CRLM-bearing mice with RAPA. CRLM growth was delayed by 8 days in average in the RAPA-treated mice. The HRI response to both hyperoxia and hypercapnia of CRLM in RAPA-treated mice was significantly attenuated compared with control-treated CRLM (Figure 5, A–C; n = 13 tumors; P < .05), whereas the liver HRI values were similar to those measured from untreated mice. Surprisingly, at the center of RAPA-treated CRLM, areas with positive ΔSco₂ were detected (Figure 5B). Blood vessel quantization (by immunohistochemical staining, with anti-PECAM-1 antibody) showed reduced vessel density in RAPA-treated CRLM (6 ± 1.3 vessels/HPF n = 9 tumors) compared with control-treated CRLM (9 ± 1.9 vessels/HPF n = 8 tumors, P < .005), in agreement with the HRI map attenuation. Furthermore, most of the vessels at the center of the RAPA-treated tumors were widened (Figure 5, G and H), clarifying the positive ΔSco₂ values detected in these tumors. When analyzing vessel maturation status in these CRLMs, we noticed that most of the vessels were stained with α-SMA both in the RAPA-treated and untreated tumors (Figure 5, F and I). In addition, there were areas with reduced tumor cell density at the center of RAPA-treated CRLM (Figure 5H).

Figure 5.

The vascular changes during RAPA therapy. (A and B) Representatives axial T₂W images (top) and enlarged HRI maps of the marked tumors (bottom) of control-treated mouse (A) and RAPA-treated mouse (B). Bar = 1 cm (A). Color scales for ΔS maps are located on both sides. (C) Mean ΔSo₂ and ΔSco₂ values ± SD of control-treated CRLM (red; n = 30 tumors) and RAPA-treated CRLM (green; n = 13 tumors): *P < .05. The ΔSco₂ maps of RAPA-treated tumors showed pixels with positive reactivity to hypercapnia differing from control-treated CRLM. Relevant histologic slides were immunohistochemically stained with PECAM-1 (CD31—red stain; D, E, G, H) for endothelial cells or with α-SMA (F, I; red stain) for smooth muscle cells (original magnification, x200 [D, G], x40 [E, F, H, I]). CD31 staining demonstrated widened blood vessels at the center of the RAPA-treated tumors. Anti-α-SMA staining confirmed that most of the tumor vessels were covered with smooth muscle cells, explaining the moderate antiangiogenic effect of RAPA on these CRLMs.

Vascular Profile Assessment of Primary Hepatic Lesions

To evaluate the HRI utility to study the vascular profile of primary hepatic lesions, we used Mdr2 knockout mice as the animal model [37]. We compared the HRI findings with the routinely used T₁W, T₂W, CE-MRI, and histologic evaluation (n = 21 lesions). On T₂W images, the detected lesions were either hyperintense or isointense. Most of these lesions were also isointense on precontrast T₁W images (see examples in Figure 6A). With CE-MRI, only some of the suspected lesions were enhanced either with central enhanced foci (Figure 6, B and C) or at the periphery of the lesion (Figure 6D). The HRI method revealed three distinct patterns of responses: 1) heterogeneous HRI response with regions of high response and others with a slightly reduced response (see example in Figure 6B), 2) elevated HRI response covering the entire lesion (see example in Figure 6C), and 3) significantly reduced HRI response compared with the adjacent inflamed liver with high ΔS values at the periphery (see example in Figure 6D). The pathologic identification, obtained from the corresponding histologic sections, could distinguish between poorly differentiated HCC (pattern I, n = 11 lesions; Figure 6B), well-differentiated HCC (pattern II, n = 6 lesions; Figure 6C), and necrotic/cystic foci (pattern III, n = 7 lesions; Figure 6D). There was a good correlation between the HRI classification and the pathologic differentiation, which may imply a beneficial usage of HRI as a complementary method for the assessment of the vascular profile of suspected primary hepatic lesions.

Figure 6.

Primary hepatic lesions vascular characterization by HRI: Three distinct patterns of HRI responses were observed in hepatic lesions identified in elderly Mdr2 knockout mice. (A) Representative T₂W image of a 17-month-old Mdr2-knockout mouse on which three different lesions are marked with white rectangles (bar = 1 cm). (B–D) Each panel represents one pattern and contains the corresponding enlarged T₁W images obtained before and after Gd injection (Top), and the corresponding enlarged HRI maps (bottom, left—ΔSo₂ map; right—ΔSco₂ map; color scale for ΔS maps is shown on B). (B) Representative sample for pattern I—with heterogeneous HRI response that was found to be characteristic for poorly differentiated HCC. (C) Representative example for pattern II—with homogeneously elevated HRI response that was found to be distinctive for well-differentiated HCC. (D) Representative sample of pattern III—with reduced HRI response inside the lesion with high ΔS values at the periphery that was found to be distinctive for necrotic foci.

Discussion

The widespread use of modern imaging techniques increases the detection of liver tumors. Early detection of liver malignancies, new therapeutic options, and new monitoring methods may improve treatment outcome. Reliable noninvasive characterization and differentiation of these lesions are of utmost importance for clinical practice [38]. In primary and metastatic liver malignancies, there is a relative increase in arterial blood supply to the tumor [26,27]. Perfusion imaging has been suggested to improve the sensitivity and specificity of liver tumor diagnostics [27]. It is well accepted that CE imaging techniques can improve the diagnosis of liver lesions larger than 1 cm [38,39] by showing increased arterial blood supply to the tumor with venous washout. Nevertheless, even the optimized imaging techniques remain relatively insensitive for the detection and vascular characterization studies of smaller nodules. In this research, we demonstrated the applicability of HRI, an fMRI method combined with hypercapnia and hyperoxia, for liver tumor vascular characterization. By using HRI, the hemodynamic changes occurring during CRLM establishment were detected, thus enabling classification of suspected foci with high sensitivity. In addition, the HRI method demonstrated the vascular changes induced by RAPA treatment and facilitated primary hepatic tumor characterization with good correlation to the pathologic differentiation.

Recently, we reported the applicability of HRI for monitoring changes in liver perfusion and hemodynamics during liver regeneration, fibrosis, and acute bleeding in rat models, without the need of contrast agent administration [34,35]. We demonstrated that during CO₂ enrichment, there is an increase in portal blood flow to the liver [33]. The resultant higher deoxyhemoglobin levels produced a decrease in fMRI signal intensity, which is illustrated by negative ΔSco₂ values. Thus, liver ΔSco₂ values are sensitive to changes in the ratio between the portal and arterial blood supplies. The signal change induced by hyperoxia signifies vascular density and tissue perfusion [33,40].

In this study, we demonstrated that, in mice, CRLM have distinctive HRI reactivity maps compared with those obtained from healthy livers. The reduction of negative ΔSco₂ values in response to hypercapnia is extremely sensitive to both tumor hypovascularity and to the increased arterial blood supply occurring during CRLM progression. The decreased ΔSo₂ values in response to hyperoxia emphasize the hypovascularity of CRLM nodules. Moreover, HRI values of healthy livers were significantly higher compared with those from livers of CRLM-bearing mice. This phenomenon may be explained by the notion that, in the presence of overt CRLM, a higher proportion of liver blood flow is derived from the hepatic artery, thus changing the entire liver hemodynamic in CRLM-bearing mice [28,41]. The CRLM mouse model used in this study revealed vascular alterations and pathologic appearance similar to those detected in human CRLM [42]. However, the rate of tumor progression in this model was faster than the kinetics known in CRLM patients.

In the described CRLM animal model, the sensitivity to detect small metastases (1–2 mm) by using HRI was significantly higher (82%) compared with multiphasic CE-MRI (38%). Previous studies showed that small CRLM (<520 µm) are hypovascular [42], and only advanced CRLM (>2000 µm) showed an exclusively arterial blood supply [42,43]. This observation may explain the enhanced detectability of HRI compared with multiphasic CE-MRI. Whereas multiphasic CE-MRI detects only the increased arterial blood supply of CRLM, HRI could also demonstrate the reduced vessel density occurring during the early phases of CRLM progression. Furthermore, the use of contrast agents, in both CT and MRI, could be associated with nephrotoxicity, especially in patients with risk factors such as renal failure, vascular disease, and diabetes [44]. The HRI method may provide a safer alternative for these patients.

The high sensitivity of HRI for small CRLM and the spatial information derived from these maps were further used to classify suspected CRLM lesions in their early growth phase by using a machine learning approach [45] (Figure W1). We showed that the recall (sensitivity) and precision (equal to positive predictive value) of HRI for suspected CRLM confirmation (≤1.6 mm) were 77% and 88%, respectively. The ability to detect subtle hemodynamic changes during CRLM establishment can assist to understand the metastasis development and progression mechanism and to develop drugs directed to the early growth phase.

Identification of new noninvasive monitoring techniques for assessing early tumor response to therapy is a major need and could facilitate decisions regarding therapy continuation or replacement. The assessment of RAPA treatment effect by HRI revealed alterations in CRLM vasculature growth pattern that were further confirmed by histologic findings. Whereas the RAPA therapy reduced tumor vascularity, it also caused the swelling of the remaining vessels. By immunostaining for α-SMA, we confirmed that most of these vessels were covered with smooth muscle cells, thus explaining the moderate antiangiogenic effect of RAPA on these CRLMs. The HRI results revealed differences between the vasculature properties at the center of RAPA-treated CRLM and the center of advanced CRLM lesions (≥7 mm). These differences were further confirmed by histologic evaluation. All of these findings emphasized the potency of HRI for understanding the underling mechanism of antiangiogenic drugs with the goal of developing strategies that could lead to earlier evaluation of the therapeutic efficacy noninvasively.

The detection and characterization of primary hepatic nodules in the multistage development of HCC in a cirrhotic liver remain an important challenge for clinicians [30,31]. We assessed the potential of HRI for HCC diagnosis based on the functional vascular profile of a variety of hepatic lesions in the Mdr2 knockout mouse model [37]. By using HRI, we could distinguish three distinct patterns of responses to the inhaled gases, which were in good correlation to the pathologic diagnosis. These results imply that HRI may facilitate the differentiation between different types of hepatic lesions based on their vascular profile noninvasively.

There are several limitations to the current study. First, liver fMRI in humans can be limited by respiratory motion artifact, which may reduce measurement accuracy and is very critical for achieving significant ΔS maps. Respiratory motion artifacts can be suppressed by navigator-gated methods [46] or corrected by postacquisition processing [47]. Because our experiments were performed on anesthetized mice, most of the motion artifacts were negligible, and there was no need for additional data processing. Second, our HRI method is based on T₂*-weighted images that are very sensitive to in-homogeneity and could be very susceptible to intrasubject and intersubject variability. To avoid this variability and to improve HRI reproducibility, we assessed the potential usage of R₂* maps during hyperoxia and hypercapnia challenges for CRLM detection in mice. Our preliminary results illustrate that the HRI and ΔR₂* maps yield similar qualitative results (Figure W2), suggesting that dynamic measurement may not be required, thus permitting improved spatial resolution. Indeed, the use of R₂* maps in 3-T clinical MR machine for liver fibrosis assessment with carbogen challenge was recently demonstrated [48]. Finally, our experiments were performed on animal models, and although the models resemble the human disease, promising results of HRI for liver tumor detection should be further investigated and reproduced in patients with liver tumors.

In summary, HRI offers a new technique to monitor changes of vessel density and perfusion ratio in liver tumors noninvasively without the need for contrast agent administration. Our experimental data from mice provide comprehensive evidence for the use of this method for earlier and more accurate hepatic tumor diagnosis in both primary and metastatic tumors. In addition, this method can serve as an imaging tool for the study of the underling mechanism of antiangiogenic drugs based on the results regarding vascular alterations during RAPA treatment. We believe that, by combining HRI with the well-established perfusion imaging methods (CE-MRI, etc), hepatic tumor detection and therapy monitoring could be improved.

Supplementary Methods

ΔR₂* Maps

Multigradient echo (MGRE) was used to quantify the transverse relaxation rate R₂*, with a repetition time of 200 milliseconds, an initial echo time of 5 milliseconds, an echo time spacing of 5 milliseconds, 5 echo times, flip angle α = 40 degrees, and 2 averages. The total imaging time was approximately 3.5 minutes. Three sets of MGRE images were acquired: 1) during breathing air, 2) after 3 minutes of 5% CO₂ breathing (4 L/min) through a mask, and 3) after a 3-minute carbogen breathing (4 L/min).

Liver R₂* (=1/T₂*) maps for each slice were generated from the MGRE images on a pixel-by-pixel basis by standard least square fitting of the natural log of the signal intensity versus TE [1], for the three inhaled gases. ΔR*co₂ was calculated by subtracting R₂*-co₂ map from R₂*-air map, and ΔR*o₂ was calculated by subtracting R₂*-co₂ map from R₂*-o₂ map [2].

Abbreviations

CE-MRI

contrast-enhanced MRI

CRLM

colorectal liver metastasis

fMRI

functional MRI

Gd-DTPA

gadolinium-diethylene-triaminepenta-acetate

HCC

hepatocellular carcinoma

HPF

high-power microscopic fields

HRI

hemodynamic response imaging

MRI

magnetic resonance imaging

RAPA

rapamycin

ROI

region of interest

SMA

smooth muscle actin

T₁W

T₁-weighted

T₂W

T₂-weighted

ΔSco₂

signal intensity change due to hypercapnia

ΔSo₂

signal intensity change due to hyperoxia

VEGF

vascular endothelial growth factor