Imaging the 3D neurovascular architecture with μ
MRI has many potential applications, a few of which will be discussed here. Traditionally, the neurovasculature has been characterized at two different spatial scales: the ‘cellular' or submicron scale using optical microscopy methods, and the ‘systemic' or submillimeter scale using methods such as in vivo
MRI. However, coregistering histology with in vivo
imaging data is challenging because of their vastly different spatial scales. Micro-MRI can potentially be used to bridge this resolution gap between optical imaging and in vivo
MRI, facilitating coregistration of cellular factors (e.g., distribution of vascular endothelial growth factor) with in vivo
biomarkers of angiogenesis, such as cerebral blood volume and vessel size index (Pathak et al, 2008a
). Both cerebral blood volume and vessel size index measurements are derived from susceptibility-contrast MRI, but the relationship between brain tumor angiogenesis and susceptibility-induced contrast is not well understood. It has been shown that abnormal tumor vessel morphology can profoundly affect susceptibility-induced contrast (Pathak et al, 2003
), and that computational models incorporating the actual vascular structure are required to elucidate this complex relationship (Kiselev, 2001
; Pathak et al, 2003
). This is now possible with the μ
MRI data acquired in this work and the recent development of a computational model of MRI contrast known as the finite perturber method (Pathak et al, 2008c
). Finally, recent evidence suggests that angiogenesis inhibition in brain tumors may promote a shift to a more invasive phenotype, providing brain tumor cells an avenue for evading antiangiogenic therapy (Chi et al, 2007
; de Groot et al, 2010
). Methods that are sensitive to both angiogenesis and brain tumor invasion, such as the μ
MRI method presented here, could prove indispensable for answering critical questions about this phenotypic shift in preclinical brain tumor models.
To the best of our knowledge, this is the first study to characterize the vascular phenotype of a mouse brain tumor model with μMRI. We showed that several commonly used measures of vascular geometry could be obtained from μMRI data and used to characterize vascular phenotypes in a mouse brain tumor model. The data presented here show the feasibility of this approach for differentiating the tumor vascular architecture from that of the contralateral brain, and for characterizing global and zonal changes in brain tumor vascular morphology with tumor progression.
The shorter vessel branch length and elevated vessel radius, MVD, LV
, FV, and tortuosity in tumor compared with contralateral ROIs are hallmarks of brain tumor angiogenesis (Jain et al, 2007
; Vajkoczy and Menger, 2000
). Our results also show characteristic differences between the vasculatures of D12 and D17 tumors. The decreases in MVD, LV
, and FV from D12 to D17 tumors are consistent with previous observations that larger tumors have lower vascular density in central and less angiogenic regions than do smaller tumors and highly angiogenic tumor peripheries (Vajkoczy and Menger, 2000
). With 3D vascular data, we performed a zonal analysis, which suggested that D17 tumors consisted of well-vascularized rims and less vascularized cores, as expected of later-stage gliomas, which are characterized by higher angiogenic activity in their peripheries (Vajkoczy et al, 1998
). In contrast, although MVD, LV
, and FV were all higher in D12 versus D17 tumors, indicative of higher overall angiogenic activity, D12 tumors were less vascularized in the rim than in the core. This may be attributable to the build-up of interstitial fluid in the cores of D17 tumors with tumor growth, which in turn, creates an outward interstitial fluid pressure gradient that leads to redistribution of proangiogenic growth factors toward the tumor rim (Vajkoczy et al, 1998
). It is also possible that larger D17 tumors contained vessels that were intermittently or poorly perfused because of abnormal hemodynamics or elevated interstitial fluid pressure, which could affect their degree of microfilling.
Furthermore, combining vascular data with other MR-contrast mechanisms provides a powerful tool for examining the interactions between vascular and neuronal structures. For example, our findings of elevated ADC in tumor compared with contralateral ROIs, and the independence of tumor ADC from tumor volume and growth are in agreement with previous DTI studies of 9L tumors in rat brains (Kim et al, 2008
; Zhang et al, 2007
). The decrease in contralateral ADC from D12 to D17 is probably attributable to compression of the contralateral brain caused by extensive tumor growth. The lateral compression from the tumor may also preferentially restrict diffusion in this direction, thereby contributing to the increase in contralateral FA between D12 and D17. The increase in contralateral FA with tumor progression may also be attributed to the inclusion of white matter tracts in the larger contralateral D17 ROIs as seen in .
Zonal analysis showed lower ADC in the rim than in the rest of the tumor for both D12 and D17 groups; D17 tumors also exhibited higher FA in their rims than in their cores. This may be indicative of higher cell density (Chenevert et al, 2000
) and, for D17 tumors, greater anisotropic cellular organization in the tumor periphery (Zhang et al, 2007
). As discussed previously, the vascular morphometric parameters we measured show that D12 tumors have less vascularized rims relative to the other zones, whereas D17 tumors have relatively higher vascularized rims. Thus, for this tumor model and the spatial resolution used in this study, increased tumor cellularity did not always spatially correlate with elevated vascularity. In addition, vascularization decreased from D12 to D17 tumors as a whole, whereas ADC and FA did not change significantly with tumor progression.
It must be mentioned that, as this is an ex vivo
study using fixed specimens, one cannot preclude the effects of aldehyde fixatives on ADC (Shepherd et al, 2009
), and it is possible that blood vessels filled with polymerized Microfil affect ADC and FA differently than do perfused blood vessels in vivo
. These ex vivo
-specific factors may have different effects on tumor and normal tissue, causing the elevated tumor ADC seen in this and previous ex vivo
The main technical challenge of using μ
MRI to image the vasculature of the whole mouse brain is achieving sufficient spatial resolution to resolve the vascular tree. The acquisition resolution of our μ
MRI scans ranged from 62 to 65μ
m, whereas the diameter of cerebral capillaries is in the range of 3 to 5μ
m. Thus, the capillaries were undetectable by μ
MRI, and the images acquired represent a subset of the total vessel population (radius
m). This is apparent when comparing the MVD visible in histology (, Supplementary Figures S1A and S1B) with μ
MRI (, Supplementary Figures S1C and S1D). When considering just the μ
MRI-visible tumor vessel population (Supplementary Figures S1E and S1F), one can see that the trends in μ
MRI-measured and histologically assessed radii between D12 and D17 tumors were similar. However, there is a slight rightward shift in the μ
MRI radius distribution relative to that assessed histologically, which may be attributed to spatial discretization caused by partial volume effects. With the increasing availability of high field magnets (
T) and more powerful gradient hardware, it is possible to achieve higher spatial resolution, e.g., 30μ
m. However, there will be a concomitant increase in acquisition time, which decreases the throughput, and the increased size of the data also poses a challenge for analysis. The μ
MRI resolution used in this study enabled us to quantify relative changes in six morphologic vascular parameters with brain tumor progression, and between the tumor and the contralateral brain. Moreover, we were able to image the whole mouse brain vasculature in 3D. Acquiring the 3D structure of the intact neurovasculature is a significant advantage of this technique and, as we discuss below, can impact the calculation of vascular morphologic parameters.
Direct comparisons between parameters obtained from μ
MRI and those obtained using μ
CT or optical techniques are challenging because of differences in spatial resolution. For example, the median tumor vessel branch length is significantly shorter than that of the contralateral ROI for both D12 and D17 groups, which is consistent with observations reported in other studies (Heinzer et al, 2008
; Zhang et al, 2009
). However, the vessel lengths computed in this study are longer than those computed using other methods such as μ
CT and optical microscopy. This may be owing to the inability of μ
MRI to resolve vessel lengths below its resolution limit. Consequently, the loss of smaller branches may cause visible parent vessels to appear as longer branches instead of comprising shorter-length vessel segments. This effect is likely compounded by differences in ROI volumes used by different studies. For example, previous studies obtained their measurements from SRμ
CT images of 1
ROIs (Heinzer et al, 2008
) or from laser scanning confocal microscopy images of 100-μ
m-thick sections (Zhang et al, 2009
), whereas we performed our analysis on ROIs that encompass whole tumors with volumes ranging from ~2.5 to 40
. It is noteworthy that imaging the whole brain with μ
MRI permits us to measure the complete lengths of longer vessels without truncation as in other techniques.
The constraint on spatial resolution may also lead to underestimation of MVD and LV
. Microvessel density is traditionally calculated from two-dimensional histologic sections and is reported in units of vessels/area instead of vessels/volume, making it difficult to directly compare our results with other studies. Boero et al (1999)
values for normal mouse brains ranging from ~430 to 1,300
, which is two orders of magnitude greater than the values measured here (1.99 to 9.15
Surprisingly, there is little consensus in the literature about the fractional blood volume of a normal mouse brain because each study uses different measuring techniques and mouse strains. Reported values range from ~0.5% to 6% (Boero et al, 1999
; Chugh et al, 2009
; Heinzer et al, 2008
; Tsai et al, 2009
; Verant et al, 2007
). The mean contralateral FV of 2.83%±0.78% calculated from our μ
MRI data falls within this range. Literature values of the fractional blood volume of orthotopic 9L tumors in the murine brain are scarce. Bremer et al (2003)
, using both in vivo
MRI and nuclear imaging, reported a mean FV of 2% for D16 9L tumors implanted in the gluteal region of nude mice. Nomura et al (1994)
also used nuclear imaging to measure blood volume and reported mean values of 12.4μ
L/g for D8 9L tumors and 1.79μ
L/g for the normal brain in Wistar rats. Assuming a brain tissue density of 1
g/mL, these values translate to 1.24% and 0.02% for tumor and normal brain, respectively. Pathak et al (2001)
measured the FV of D10-D30 9L tumors implanted in Fisher rat brains using stereological techniques and obtained a mean value of 5.29% versus 1.89% for the normal brain. The mean tumor FV over all tumors in our study was 10.81%±3.04%. Although it is well established that 9L tumors exhibit increased blood volume relative to the normal brain, the results of these studies underscore the sensitivity of such measurements to the methodology used and the subsequent difficulty in comparing results derived from different studies.
As expected, tumor vessels are significantly more tortuous than contralateral vessels in the D12 group, but there was no significant difference for the D17 group. Heinzer et al (2008
) reported median tortuosity values between 1.2 and 1.25 for normal vessels >7.5μ
m in diameter, whereas the median contralateral tortuosity calculated in this study was 1.12. Again, it is likely that partial volume effects led to lower tortuosity values because directional variations of blood vessels on a scale comparable with or smaller than the image resolution were undetectable.
On the basis of all μMRI-measured parameters discussed above, we characterized the phenotypic changes of the brain microenvironment that accompany tumor progression. This is evident from the double dendrogram in , which shows that unsupervised hierarchical clustering sorted tumor and contralateral ROIs into two well-separated clusters. It then sorted D12 and D17 contralateral ROIs into two separate, smaller clusters. Within the tumor cluster, four D12 tumor ROIs were further distinguished as one subcluster, and the fifth D12 tumor ROI was assigned to a second subcluster with D17 tumor ROIs. This lone D12 tumor ROI was the closest ROI within that subcluster to the other D12 tumor ROIs in parameter space. We also found that removing ADC and FA from the cluster analysis did not appreciably affect the clustering of tumor ROIs, but did negatively impact the clustering of contralateral ROIs (data not shown). This indicates that the ‘vascular phenotypes' of the D12 and D17 tumors were unique, and that tumor growth from D12 to D17 caused a substantial mass effect in the contralateral brain.
In conclusion, μMRI has the potential to characterize the vascular phenotype of preclinical brain tumor models. Our method could differentiate between tumor and contralateral vasculatures, as well as between the vascular phenotype of D12 and D17 tumors. Although μMRI does not have submicron imaging capability, it is capable of ‘whole-brain' mapping, offering superior coverage to optical techniques. In addition, although μMRI cannot outperform μCT in terms of spatial resolution, it has the advantage of complementary contrast mechanisms such as DW imaging, which allows us to simultaneously measure changes in the brain tumor microenvironment. The work presented here shows the feasibility of using μMRI to study the relationship between angiogenesis and other components of the brain microenvironment in a range of pathologies involving the neurovasculature.