Monitoring Tumor Response to Antivascular Therapy Using Non-Contrast Intravoxel Incoherent Motion Diffusion-Weighted MRI Free

Despite recent advances in treatment, advanced non–small cell lung cancer (NSCLC) remains one of the most lethal cancers. The efficacy of currently available chemotherapy for advanced NSCLC, including platinum-based doublet, crizotinib, gefitinib, erlotinib, afatinib, bevacizumab, and pemetrexed, is still limited (1), suggesting an urgent need to develop effective treatments that can significantly improve the overall survival rates. Vascular-disrupting agents (VDA) are a class of anticancer agents that can selectively attack the tumor vasculature and cause catastrophic vascular shutdown, tumor ischemia, and necrosis (2, 3). Several VDAs, including combretastatin A4 phosphate (CA4P; ref. 4), are now being investigated in phase II/III clinical trials for advanced ovarian, NSCLC, and anaplastic thyroid cancer (5, 6). Differing from antiangiogenic agents, another major class of vascular-targeting therapeutics that act by inhibiting the formation of a neovasculature (angiogenesis), VDAs act on the existing tumor vasculature within the first several hours (up to 24 hours) after administration. However, this acute effect on tumor perfusion is reversible, the survival tumor cells can quickly regain vascular perfusion, and tumors regrow soon after (4). Consequently, VDAs are often used in combination with other therapies, including radiotherapy (7), hyperthermia (8), chemotherapy (9), and antiangiogenic therapy (10). Antiangiogenic agents directly or indirectly inhibit proangiogenic signaling pathways to completely or partially block the formation of new blood vessels (2, 11). For example, bevacizumab (Avastint), a recombinant humanized monoclonal antibody, inhibits VEGF-A and is in clinical use for metastatic colorectal cancer and NSCLC (3, 11).

Noninvasive MRI plays an important role in the evaluation of antivascular therapies (12). As antivascular therapy mainly acts on the tumor vasculature, a reduction in tumor size would occur at some time after therapy administration. A number of MRI methods have been used to evaluate tumor responses during the course of antivascular therapies. For example, dynamic contrast-enhanced MRI (DCE-MRI) is widely used to assess tumor perfusion (10, 13); diffusion-weighted imaging (DWI) is used to monitor the changes in tumor cellularity (14, 15); and ¹⁹F MRI (16) is used to map oxygenation by imaging lipid relaxation enhancement (MOBILE; ref. 17). Among these methods, DCE-MRI has been shown to have a superior ability to render the changes in perfusion and permeability and is the most commonly used MRI method in antivascular therapies (18, 19). However, there is a rising concern about Gd deposition (20, 21), which, although rare, prevents DCE-MRI from being used frequently and repetitively for clinical applications.

We hypothesized that intravoxel incoherent motion (IVIM) DWI can be used as a non–contrast-based alternative to traditional DCE-MRI for assessing the tumor response to an antivascular therapy, which can be used repetitively in a short time interval. The IVIM method was originally developed by Le Bihan in 1986 using a biexponential model to separate the diffusion-weighted signal into a pure diffusion fraction and a perfusion dominated “pseudo-diffusion” fraction (22), permitting the simultaneous assessment of tissue diffusion and microperfusion (22–24). Recently, in both preclinical and clinical studies, there has been renewed interest in utilizing the perfusion-related parameters of IVIM DWI, including the blood pseudo-diffusion coefficient (D*) and the perfusion fraction (f), to quantify microvessel perfusion in solid tumors (25–31). The IVIM method is considered an attractive approach because it can assess microperfusion in the tumor without the need for exogenous contrast agents, making it possible to be used within a short time interval for the evaluation of therapeutic response, even in patients with renal insufficiency, or in patients with contraindications to contrast agents (24, 32).

In the present study, we aimed to investigate whether IVIM DWI can be used as to characterize early changes in tumor perfusion and diffusion as a result of antivascular therapies, including a CA4P-based vascular-disrupting therapy and retreatment using a combination of CA4P and bevacizumab. Furthermore, we examined whether IVIM parameters could be used to predict the therapeutic responses of NSCLC tumors. As DCE-MRI is the most widely used MRI technology, we also studied the correlation between the perfusion-related IVIM-DWI parameters and DCE-MRI parameters. To the best of our knowledge, this is the first study to report on the use of IVIM DWI to evaluate the therapeutic effects of antivascular therapies in an experimental NSCLC model.

The human lung carcinoma cell line A549 was obtained from the American Type Culture Collection (ATCC, lot# CCL-185, purchased 2015) and passaged less than 15 times prior to implantation. The cells were cultured at 37°C in a 5% CO₂ atmosphere in DMEM (Gibco), supplemented with 10% FBS, and 1% penicillin/streptomycin. Immediately prior to in vivo implantation, the A549 cells were assessed by PCR to provide STR profiles, from which cell line authenticity was confirmed, and the absence of mycoplasma was confirmed using an e-Myco PCR detection kit (Boca Scientific).

All animal experiments were performed in accordance with protocols approved by our institutional Animal Care and Use Committee. Male BALB/c nude mice (4–5 weeks of age, body weight ∼20 g) were purchased from the laboratory animal center of Sun Yat-sen University (Guangzhou, China) and maintained in a specific pathogen-free environment. The mice were subcutaneously injected with 2 × 10⁶ A549 cells in 0.2 mL serum-free media into the left flank to develop NSCLC xenografts. Tumors were allowed to grow for 14 days to reach a size of approximately 10 mm in the longest dimension. This tumor size was sufficient for the development of relatively high vascularity, allowing MRI assessment of the effect of antivascular therapies. Tumor volumes were assessed using an MRI-based volumetric measurement method. In brief, for each animal, observer-defined regions of interest (ROI) were placed on the tumor throughout a series of axial fast Spin Echo (FSE) T2w images, and the number of pixels in each ROI was counted. The tumor volume was then calculated by the multiplication of the sum of pixel number and voxel volume (defined by the product of pixel area and slice thickness).

A total of 57 mice with NSCLC xenograft tumors were randomly allocated to three groups: group A received saline as control (n = 21); group B received CA4P treatment (n = 18); and group C received CA4P plus bevacizumab treatment (n = 18). Saline, CA4P (100 mg/kg; OXiGENE, Inc.), or bevacizumab (5 mg/kg, the Avastin, Roche) were administrated via tail vein injection immediately after baseline MRI scans. In the group C (CA4P + bevacizumab), the CA4P was administered 1 hour after the bevacizumab on the first day (33). Then bevacizumab was administrated daily. The schedule for treatment, MRI, and histologic assessments is shown in Fig. 1. In brief, MRI was conducted before treatment, and 2 hours, 2 days, 4 days, and 8 days after the treatment of CA4P or vehicle. Six mice from each group were randomly selected for longitudinal MRI assessment from day 0 to day 8. At each time point, 3 mice from the non-MRI subgroup in each group were randomly selected and sacrificed for histologic analysis. Three mice were randomly selected from a control group as the baseline for histologic analyses.

All MRI scans were conducted using a 1.5-T Signa HDxt superconductor clinical MR system (GE Medical System) equipped with a human eight-channel wrist coil. Animals were anesthetized by intraperitoneal injection of 0.3% pentobarbital and imaged in the supine position. Transverse T₂-weighted images were acquired using an FSE sequence [TR = 2,280 ms, TE = 77.6 ms, matrix size = 256 × 192, field of view (FOV) = 7.0 × 5.6 cm², slice thickness = 2.0 mm, slice gap = 0.2 mm, and NEX = 2]. DCE-MRI was acquired using a 3D fast spoiled gradient-recalled echo (SPGR) sequence (TR = 40 ms, TE = 2.4 ms, flip angle = 35°, slice thickness = 2.0 mm, slice gap = 0.2 mm, matrix size = 128 × 96, and FOV = 7.0 × 5.6 cm²). A total of 35 T1w images were obtained at a temporal resolution of 3 seconds before and after the injection of Gd-DOTA (Magnevist, Bayer Schering Pharma, 0.1 mmol/kg body weight followed by 0.3 mL of a 0.9% saline flush). The baseline T₁ map was assessed using a variable flip angle T1 mapping method, with the flip angle = 3°, 6°, 9°, 12°, and 15°. IVIM-DWI MRI was acquired using a free-breathing, single-shot, echo-planar imaging pulse sequence (TR = 4,000 ms, TE = 91.8 ms, slice thickness = 2.0 mm, slice gap = 0.2 mm, matrix size = 128 × 96, and FOV = 10 × 7 cm²) with diffusion gradients applied in three orthogonal directions (13 b values = 0, 25, 50, 75,100, 150, 200, 400, 600, 800, 1,000, 1,200, 1,500 s/mm²). The chemical shift-selective saturation technique was used for fat suppression.

After each MRI acquisition session, MR images were transferred to a dedicated post-processing workstation (AW4.5, GE Healthcare) for quantitative analysis. The IVIM-DWI data were analyzed by the Functool MADC program, using a biexponential model defined by

where SI₀ is the mean signal intensity of the ROI for a b value of 0, and SI is the signal intensity for higher b values. D represents the true diffusion coefficient, D* is the pseudo-diffusion coefficient, and f is the perfusion fraction. ROIs were manually drawn by outlining the tumor border on the T2w image that showed the largest cross-sectional tumor area.

DCE-MRI data were analyzed using the Cinetools program and the Tofts model, and the volume transfer coefficient (K^trans) was calculated. A total of 90 IVIM-DWI and 90 DCE-MRI datasets from 18 mice, each at 5 different time points (i.e., baseline, 2 hours, 2 days, 4 days, and 8 days), were analyzed by a radiologist with 6-year experience in MRI.

Excised tumors were fixed in 4% paraformaldehyde, embedded in paraffin, sliced into sections of 5 μm thickness, and stained using a hematoxylin and eosin stain. TUNEL staining was performed to assess apoptotic cells using an in situ Cell Death Detection kit (Roche Diagnostics) according to the manufacturer's instruction. Ki67 staining was performed to assess cell proliferation using an anti-Ki67 antibody (1:200; Santa Cruz Biotechnology). HIF1α staining was performed to assess tumor hypoxia using a monoclonal anti–HIF-1α antibody (1:25; Novus Biologicals). Finally, CD31 staining was performed to measure microvascular density (MVD) using an anti-CD31 antibody (1:200; BD Biosciences). For TUNEL, Ki67, HIF1α, and CD31 analyses, ten fields per section were randomly selected and analyzed. The MVD of the tumor was evaluated using the “hot spot” method described by Weidner and colleagues (34). In brief, the three most vascularized areas on each section were selected as the hot spots under low magnification (×40), and vessels were manually counted on the images at high magnification (×200). The MVD was calculated as the mean of three hot spot areas. The TUNEL, Ki67, HIF1α, and CD31 were expressed as the ratio (%) of positive cells to all tumor cells using Image-Pro Plus 6.0 software (Media Cybernetics).

All numeric data were presented as the mean ± SD. The normal distribution of the acquired data was evaluated using the Kolmogorov–Smirnov test. The computed IVIM-DWI and DCE-MRI parameters in all three groups were compared using a one-way ANOVA with least significant difference (LSD) as a post hoc test. Pearson correlation analysis was used to assess the correlations between IVIM-DWI and DCE-MRI parameters at each time point, and the correlations between the relative changes in IVIM-DWI parameters at early time points (i.e., 2 hours), as well as treatment responses quantified by the change in tumor volume on day 8. P < 0.05 was considered statistically significant. All statistical analyses were performed using the SPSS software package (Version 19.0, SPSS Inc.).

Our studies showed that CA4P treatment resulted in a striking inhibitory effect on tumor growth as early as 4 days after the treatment. As shown in Fig. 2A, treatment groups (both CA4P and CA4P + bevacizumab) exhibited a much slower increase rate in tumor volumes compared with the control group. As shown in Fig. 2B, the relative increase in tumor volume on day 8 for the CA4P and CA4P + bevacizumab treatment groups were 95.3% ± 13.5% and 73.2% ± 22.4%, respectively. In comparison, the tumor volumes in the control group increased for more than 2-fold, i.e., 201.4% ± 31.1%. The doubling time for tumor volumes was estimated using a single exponential growth equation to be 5.5 days (control), 8.6 days (CA4P), and 11.3 day (CA4P + bevacizumab), which is consistent with previous findings (33). An ANOVA with post hoc LSD tests revealed significant differences between the tumor sizes of the treatment groups and that of the control groups (all P values were less than 0.001), but not between the two treatment groups (P > 0.05, Fig. 2B). However, no group showed a decrease in tumor size, indicating that a single injection of CA4P or CA4P + bevacizumab, while inhibiting tumor growth effectively, cannot eradicate the tumor completely.

To evaluate the effect of antivascular treatments on NSCLC A549 tumors, we performed CD31, Ki67, TUNEL, and HIF1α at different time points after treatment with CA4P or CA4P + bevacizumab to assess MVD, cell proliferation, cell apoptosis, and tumor hypoxia, respectively. Representative immunohistochemical sections of CD31, Ki67, TUNEL, and HIF1α staining in the CA4P + bevacizumab group at different time points are shown in Fig. 3A, and the quantitative analyses for each staining are shown in Fig. 3B–E. CD31 staining showed a sharp drop at 2 hours after the treatment in the two treatment groups, which was attributed to the vessel-disrupting effects of CA4P. This was followed by a gradual recovery after day 2. Moreover, CD31 staining also revealed a markedly lower MVD in the CA4P + bevacizumab treatment group than in the CA4P treatment group. Ki67 staining showed distinctive patterns of changes in cell proliferation for the different groups: the control group showed a slow but continuous increase in Ki67 expression, indicating growth and proliferation of tumor cells; the CA4P treatment group demonstrated a negligible change in Ki67 expression throughout the experiment, indicating tumor cells were in a nonproliferating stage due to the inhibition of CA4P on tumor growth. In contrast, the CA4P + bevacizumab treatment group showed a substantial decrease in Ki67 expression as early as 2 hours after the treatment, indicating a much higher therapeutic effect on the proliferation of tumor cells. TUNEL staining showed that the maximal degree of cell apoptosis occurred on day 2 after treatment. HIF1α staining revealed a noticeable drop in tumor oxygenation, which was most profound at 2 hours after treatment in both treatment groups, and this was maintained at a higher level in the CA4P + bevacizumab group but oscillated greatly in the CA4P group.

To evaluate the feasibility of using IVIM-DWI as a noninvasive means by which to assess tumor responses to antivascular therapies, we performed longitudinal IVIM-DWI MRI on NSCLC A549 tumors in the control group, the CA4P group, and the CA4P + bevacizumab group for up to 8 days (Fig. 4). We also used DCE-MRI as a standard method with which to validate the quantitative measurement of tumor perfusion using the IVIM-DWI method. The longitudinal measurements from IVIM-DWI and DCE-MRI parameters in each group at each time point are shown in Fig. 5, and the relative changes in each parameter in the three groups are summarized in Table 1. In the control group (n = 6), both IVIM-DWI and DCE-MRI measurements showed good intragroup reproducibility, i.e., the CVs of D* and f for the three groups were 15.9%, 16.8%, and 14.8%, respectively, and the CV of K^trans was 9.8%.

A catastrophic effect of CA4P (with or without bevacizumab) on the tumor perfusion was revealed by the perfusion-related IVIM parameters, including D*, f, and fD* (Fig. 5A–C and Table 1). For the CA4P treatment group, the average D*, f, and fD* were first decreased dramatically to 40.8%, 35.5%, and 61.6% of their baseline values, respectively, at 2 hours after the treatment. This effect was then reversed slowly and, as early as day 2, no significant difference could be observed in these perfusion-related parameters between the CA4P-treated groups and the control group (P > 0.05). On day 8, only a net decrease of 10.7%, 18.1%, and 27.3% was observed for D*, f, and fD*, respectively, compared with the baseline values. In contrast, the decrease in perfusion was much stronger and longer in the CA4P + bevacizumab treatment group. While all the perfusion-related parameters were also reversed slowly on day 2, they were still significantly lower than those of the control group and the CA4P treatment group (P < 0.05). The difference in the perfusion-related parameters between the CA4P groups and the other groups became insignificant after day 4.

As shown in Fig. 5D, a similar trend in the tumor perfusion could be seen using DCE-MRI. For example, the K^trans in the treated tumors decreased significantly by 43.5% and 55.3% in the CA4P and CA4P + bevacizumab groups, respectively, at 2 hours after the treatment. The decrease in K^trans was then reversed gradually. Compared with the control group, both the CA4P-treated and (CA4P + bevacizumab)-treated groups showed a significantly different K^trans till day 2 (P = 0.003 and P < 0.001, respectively). However, no significant difference in K^trans was observed after day 4, except on day 8, where a significantly high K^trans in the CA4P groups was observed. On day 8, the average K^trans in the CA4P + bevacizumab group was still 5% lower than their baseline value, indicating a great synergy between the vascular-disrupting effect of CA4P and the inhibitive effect on the recovery of collapsed tumor vessels and formation of neovasculature by bevacizumab.

We also assessed the changes in the diffusion parameter, D, in the IVIM model, in the treatment groups. As shown in Fig. 4E, the D values in both groups B and C were slightly decreased at 2 hours (P = 0.009 for group B and P = 0.007 for group C), and then markedly increased after day 2 (P = 0.004 for group B and P < 0.001 for group C) and remained at a high level on day 8 (P = 0.131 for group B and P = 0.005 for group C) compared with the control group. Interestingly, although the differences were not statistically significant, the average increases in D in the CA4P + bevacizumab treatment group were always higher than those in the CA4P group at each time point.

Heterogeneous distribution can be seen among all metrics in the obtained MRI maps. Therefore, we also investigated the changes on these parameters in the rim and core of each tumor using a simple segmentation method as shown in Supplementary Fig. S1. As expected, the pronounced changes in diffusion parameters occurred in the tumor core (Supplementary Fig. S2), whereas all the perfusion-related parameters showed much stronger changes in the tumor rim (Supplementary Figs. S3–S6).

To validate the usefulness of IVIM-DWI parameters in monitoring the changes in tumor perfusion, we performed analyses to study of the correlation between the perfusion measured using IVIM and that by DCE-MRI. As shown in Fig. 6, moderate but significant correlations were observed between the three perfusion-related IVIM parameters and K^trans: D*: r = 0.6934 (P < 0.0001), f: r = 0.6453 (P < 0.0001), and fD*: r = 0.7127 (P < 0.0001), respectively. Moreover, the analyses based on segmented tumor regions (Supplementary Figs. S4–S6) clearly showed an improved correlation between the changes in the perfusion-related IVIM metrics with K^trans measured using DCE-MRI in the tumor rim than those in the whole tumor.

To investigate the usefulness of IVIM-DWI parameters in early predicting the tumor response to an antivascular treatment, we also carried out analyses to study the correlation between the IVIM parameters at early time points and the tumor responses, quantified by the changes in tumor volume, on day 8. As shown in Fig. 6D, the changes in the perfusion-related parameter fD* at 2 hours after treatment showed statistically significant correlations with the relative changes in tumor volume on day 8, with a Pearson correlation coefficient of 0.8663 (P = 0.0003). When compared with the changes of all four immunohistochemical markers in the time course of the study (Fig. 6E), fD* at 2 hours after the treatment can definitively separate both treatment groups from the control group, reflecting the changes in tissue perfusion. It should be noted that the measured perfusion changes in the tumor rim had an improved correlation with the changes in tumor volume (i.e., r = 0.93, P < 0.0001, Supplementary Fig. S7G), indicating that further improvements can be achieved using appropriate image segmentation methods to take the tumor heterogeneity into account.

As shown in Fig. 6F, the changes in the diffusion parameter D on day 2 also had a strong negative correlation with the changes in tumor volume on day 8, with r = −0.8853 (P = 0.0001). Figure 6G shows that D on day 2 can effectively separate the CA4P + bevacizumab group from the control group in a similar way of immunohistochemical markers. However, D alone cannot be used to discriminate CA4P group and control group effectively. It should be noted that the measured diffusion changes in the tumor core had an improved correlation with the changes in tumor volume (i.e., r = −0.8592, P = 0.0003, Supplementary Fig. S7H), indicating that, again, further improvements can be achieved using appropriate image segmentation methods to take the tumor heterogeneity into account.

The purpose of the present study was to investigate the feasibility of using IVIM to assess the changes in tumor perfusion and diffusion in response to antivascular therapies. It should be noted that the current study used an A549 NSCLC mouse model. It is because that, even orthotopic models such as Lewis lung adenocarcinoma (3LL) can be used, xenografts of human tumor cells such as A549 and H460 indeed have been used more commonly to investigate vascular disruption in NSCLC xenografts in preclinical studies (35, 36), which have led to a number of clinical trials. Moreover, as exemplified by several recent clinical studies (37, 38), the IVIM technique used in the current study indeed is ready for imaging lung cancers in clinical settings. We used the most widely studied vascular-disrupting agent CA4P that would result in a significant inhibition of tumor growth (4, 7–9). Our result was in a good agreement with previous studies that, when combined with the antiangiogenic agent bevacizumab, the antitumor effect can be much stronger than the CA4P alone, by the synergy of the vascular-disrupting effect and inhibition of new vessels (10, 33). The change in tumor perfusion was confirmed by DCE-MRI, a tracer-based MRI method that is being widely used in antivascular therapies as an imaging biomarker (13, 18). The inhibition of tumor growth correlated well with the strong decrease in tumor perfusion at 2 hours after treatment. The changes in the tumor vasculature and those on a cellular level were confirmed by immunohistochemistry. Using this well-established and validated animal model, we showed that IVIM is capable of assessing the changes in both tumor perfusion and diffusion in response to antivascular treatments, without the need for contrast media injection.

The biexponential IVIM model can be used to estimate perfusion-related parameters, including the blood pseudo-diffusion coefficient (D*), indicative of blood flow, and the perfusion fraction (f), indicative of the fractional volume of active capillaries in the tumor, by fitting to fit DWI data acquired at multiple b values (22), which provides a simple and practical way to quantitatively assess the perfusion in small vessels and capillaries, namely microperfusion (23–31). In contrast, DCE-MRI measures classic perfusion, which is the pharmacokinetics of the injected tracer, and the volume transfer constant, K^trans, represents tissue perfusion and permeability (39). Therefore, the perfusion studied by IVIM and that by DCE-MRI are intrinsically different, which can explain the discrepancy reported in some studies (40, 41). However, many other studies have shown that there is, indeed, a good correlation between IVIM parameters and the K^trans measured by DCE-MRI (26). Our results also suggested that IVIM parameters were well correlated with DCE-MRI with Pearson correlation coefficients on the order of 0.6–0.7. Among the parameters studied, fD* showed the strongest correlation (r = 0.7127). Because tumor perfusion assessed by DCE-MRI is the most widely used biomarker in antivascular therapies (13, 39), our current study implies that IVIM DWI, as a non–contrast-enhanced alterative to DCE-MRI, is a promising approach for monitoring the tumor response to antivascular therapies, and can be used repetitively at short time intervals, without concerns for Gd deposition safety, which is particularly helpful to patients with impaired renal function (6).

It should be noted that there was a non-negligible heterogeneous spatial distribution in all MRI metrics (Figs. 4 and 5; and Table 1). The nature of this heterogeneity is the tumor heterogeneity within an individual tumor and the response of different parts to an antivascular treatment. Therefore, our MRI metrics should be considered as a gross assessment for the changes in tissue perfusion and diffusion globally. Not only the mean values but also standard deviations should be considered in order to interpret the data accurately. Ideally, a careful mechanistic validation using a high-resolution small animal MRI scanner can guarantee the spatial correlation between the pathologic changes and MRI findings, which is however beyond the scale of current study.

As early as 2 hours after the administration of CA4P, all perfusion-related parameters, including D*, f, fD*, and K^trans, showed a sharp decrease of about 30% to 60%. The destruction of the tumor vasculature was confirmed by immunohistochemical findings using CD31 staining. A strong hypoxia due to the loss of blood supply was also confirmed by HIF1α staining. As a result of the shutdown of blood perfusion, tumor cells began to develop apoptosis and the portion of proliferating cells decreased. Both the measures of IVIM DWI and DCE-MRI could reflect the change in perfusion, indicative of the acute effect of CA4P. There was no significant difference in the perfusion parameters between the CA4P group and CA4P + bevacizumab group, indicating the effect of CA4P is dominant, whereas the effect bevacizumab is negligible in the early stage. In this stage, D obtained from IVIM model strongly decreased, likely attributable to the decrease of extracellular space (reduced overall water diffusivity) as a result of ischemic swelling cells (31), which is consistent with our histologic findings and in good agreement with previous studies on rodent models of rhabdomyosarcoma (42) and liver tumors (31).

On day 2, both immunohistochemistry (CD31 and HIF1α staining) and MRI results showed the reversion of tumor perfusion. There was no significant difference in IVIM parameters (D*, f, and fD*) between the CA4P treatment group and the control group, whereas there was a significant difference in K^trans between them. This observation can be explained by either the recovery of collapsed capillaries or the development of new capillaries, which led to an increase in microperfusion, but not classic perfusion, as large vessels were still not fully opened. In contrast to the CA4P treatment alone, the differences in the IVIM parameters (D*, f, and fD*) between the CA4P + bevacizumab treatment group and the control group were still significant, confirming that the function of bevacizumab was to inhibit the formation of new blood vessels (11). Interestingly, only IVIM parameters, but not K^trans, showed a significant difference between the CA4P and CA4P + bevacizumab groups. At this time point, histologic analysis showed a recovered CD31 and HIF1α expression, the greatest degree of cell apoptosis (TUNEL staining), and strongly inhibited cell proliferation (Ki67 staining). All the immunohistochemical analyses showed a noticeable difference between the two treatment groups, suggesting a significant difference on the cellular level. The IVIM measurement also revealed a markedly increased D caused by cell death (both necrosis and apoptosis), edema, and further tumor repopulation (2, 5, 31, 42, 43). For example, Loveless and colleagues reported that diffusion parameters but not DCE-MRI parameter K^trans served as a better biomarker for antiangiogenic therapies (44). Our result also proved the strong correlation between the diffusion parameter D with the changes in tumor volume on day 8, suggesting that the D diffusion parameter is a complementary biomarker to the perfusion parameters.

After day 4, both immunohistochemistry and MRI results showed the regrowth of blood vessels and the recovery of blood flow in the treatment groups. Among all the perfusion-related parameters, only the relative changes in f (and fD*) in the CA4P + bevacizumab group showed a significant difference compared with the control group. Higher D values were still observed in the two treatment groups, which is consistent with the TUNEL staining and Ki67 staining.

The present study suggests that fD* (both relative change and absolute value) at 2 hours can be used as early predictors of the efficacy of antivascular treatments. This finding is consistent with previous studies on the antiangiogenic agent, sorafenib (45), and the vascular disrupting agent, CKD-516 (31), on liver cancers. Our study also indicated that the relative changes in f and fD* on day 2 could predict tumor resistance to vascular-disrupting agents. When combined with antiangiogenic agents, such as bevacizumab, the antivascular effects of CA4P were greatly augmented, which is in good agreement with previous reports (33, 46). In our study, only the relative changes in f and fD* on day 2 could effectively predict the different efficacies of CA4P and CA4P + bevacizumab treatment.

Moreover, our studies demonstrated that the early changes in microperfusion (fD* at 2 hours) and diffusion (D on day 2) measured by IVIM DWI correlated well with immunohistochemical changes. Specifically, Fig. 6E and G shows that fD*(2 hours) and D (2 days) correlated negatively with CD31 and Ki67 but positively with TUNNEL and HIF1α, which are consistent with the previously reports (31, 42, 44, 47). It is well known that D and fD* reflect different aspects of the tumor response to antivascular treatments, which are complementary to each other: D (2 days) represents the early tumor responses at cellular level on the first 2 days (25, 47) and fD* (2 hours) represents the acute tumor vascular responses in first 2 hours. Compared with D (2 days), fD* (2 hours) of control group and those of treatment groups were more separable, indicating treatments caused much stronger effects on tumor microperfusion (i.e., MVD and microcirculation) than on diffusion characteristics (tissue cellularity, extracellular space tortuosity, and cell membrane integrity) in the early stage (47). Nevertheless, D (2 days) and fD* (2 hours) might be used combinedly for the early and accurate prediction of the tumor response to antivascular treatments.

In conclusion, our results suggest that IVIM DWI is a promising alternative to DCE-MRI for assessing the change in tumor perfusion by antivascular agents. In addition, the diffusion parameter measured using IVIM DWI provides insights into the changes in tumor cellularity, which is complementary to perfusion measurements, thus providing a better understanding of tumor response to antivascular treatment. The relative changes in f and fD* at early time points can be used to predict the efficacy of antivascular therapies without the need for contrast media injection.

No potential conflicts of interest were disclosed.

Conception and design: C. Shi, Z. Xiao, H. Chen, L. Luo

Development of methodology: H. Chen, L. Luo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Liu, Z. Xiao, D. Zhang, H. Chen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Shi, D. Liu, D. Zhang, G. Liu, G. Liu

Writing, review, and/or revision of the manuscript: Z. Xiao, D. Zhang, G. Liu, G. Liu, H. Chen, L. Luo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Zhang, L. Luo

Study supervision: H. Chen, L. Luo

C. Shi, Z. Xiao, and L. Luo received Science and Technology Planning Project of Guangdong Province, China (2014A020212673) and Characteristic Innovative Project of Guangdong Province Department of Education (2014KTSCX021).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.