Quantitative Fluorescence Microscopy Measures Vascular Pore Size in Primary and Metastatic Brain Tumors Free

The incidence of brain metastasis has substantially increased and represents approximately 50% of intracranial lesions observed in humans (1, 2). Yet the literature on permeability of the blood–tumor barrier stands in opposition with >90% of the work being performed in primary tumor models. This discrepancy must be addressed as it has recently been shown that permeability of brain metastases of breast cancer in two experimental metastasis model systems is highly heterogeneous, ranging nearly 30-fold between lesions and does not directly reflect what is observed in primary tumors (3). Therefore, this study evaluates two critical permeability parameters in metastatic lesions; (i) size-dependent permeability, (ii) size of pores in tumor vasculature, as both are critical to the understanding of the size of chemotherapeutics, which can diffuse into metastatic lesions and potentially have therapeutic benefit.

Within the normal brain vasculature, endothelial cells are sealed together by tight-junction protein complexes. This unique structure coupled with pericytes, astrocytes, and neuronal input creates a barrier that significantly restricts para-cellular diffusion of polar and or large molecules between the blood and brain (blood-brain barrier; BBB). However, the vasculature within a brain tumor is both anatomically and functionally different. Anatomically, tumors as small as 0.5 mm in diameter can promote the formation of new vessels that lack the classic BBB tight junction structure (13), have lost astrocytic contact, and have both increased vesicular density (14) and fenestrations or pores, which allow the free para-cellular passage of molecules from blood into brain (15). These anatomic changes functionally result in the brain tumor vasculature (blood–tumor barrier, BTB) having an increased permeability (16–18), which is size dependent (19, 20). For example, small molecules (<500 Da) have substantially increased diffusion rates (up to 30-fold) and larger molecules (>5,000 Da) have only an approximately 2- to 3-fold increase (21–23).

Most nonbiological chemotherapeutics (e.g., paclitaxel and doxorubicin) are typically less than 1,500 Da in size and accordingly tend to penetrate tumors through angiogenic vascular BTB pores to a significantly greater degree than in normal brain (12). Recently, numerous mAbs have been approved for the treatment of breast, colon, squamous cell carcinomas, glioma, and lymphomas/leukemias (24–26). These antibodies are approximately 100-fold larger in size than chemotherapeutics and, if they are able to cross the BTB, may have efficacy in treating both primary and metastatic tumors. Determining the size of molecules that can diffuse across the BTB is technically difficult but has been shown preclinically in models that have homogenous permeability from tumor to tumor (e.g., gliomas). In this report, we describe novel imaging methodology that allows for the simultaneous measurement of size-dependent permeability and prediction of vascular pore sizes in metastases that have heterogeneous permeability. These data are then directly compared with a primary tumor model as control, because, to the best of our knowledge, this is the first report that calculates pore size in metastatic lesions and/or any tumor in the brain that has substantial variability in permeability among different tumors.

Sulpho-rhodamine 101 (MW 625 Da), TX Red 3 kDa dextran lysine fixable, TX Red 70 kDa dextran lysine fixable were purchased from Molecular Probes and Invitrogen. ¹⁴C-AIB (specific activity: 55 mCi/mmole) was purchased from American Radiolabeled Chemicals. All other chemicals are of analytic grade and were purchased from Sigma.

All animal experiments were conducted in accordance with Institutional Animal Care and Use Committee and NIH guidelines. Intraperitoneal injections of ketamine (100–200 mg/kg) and xylazine (5–10 mg/kg) were used for anesthesia in all experiments. Human glioblastoma multiforme does not have a circumscribed border and as such we utilized the rat glioma (RG-2; obtained from ATCC in 2010 and in passage less than 6 months at the time of the experiment) tumor cell line as this model also displays an infiltrative growth pattern and is representative of human glioblastoma multiforme (27). Briefly, male Fischer-344 rats (Charles River Laboratories) were implanted stereotactically with approximately 100,000 stably eGFP-transfected RG-2 cells in 5 μL (28) at coordinates of 4.5 mm lateral to bregma and 5 mm deep.

Seven days after implantation, one of the three fluorescent markers (6 mg/kg body weight) was injected intravenously; TX Red 625 Da (0–5 minute circulation), TX Red 3 kDa dextran (0- to 20-minute circulation), or TX Red 70 kDa dextran (0- to 60-minute circulation). Six blood samples were collected throughout each experimental time period (see Fig. 1) and after collecting the final blood sample, animals were sacrificed and brain tissue was removed rapidly (<60 seconds) from the skull and flash frozen in isopentane (−50°C). Brains were then sectioned (20 μm) using a cryostat (Shandon Cryotome) at −23°C and sections were mounted on glass slides and air dried. Concentrations of the fluorophores in blood and brain were determined using fluorescent microscopy using matching tissue standards (as described previously; ref. 12). In a separate set of tumor-bearing animals, we injected ¹⁴C-AIB (100 μCi/Kg) along with a fluorophore to compare permeability differences of these tracers. After tissue preparation, fluorescence images were taken to analyze the TX Red 3-kDa dextran permeability, and autoradiograms were generated by coexposing the sections on Fujifilm FLA 7000 with tissue-calibrated ¹⁴C-standards for 4 days. The regional radioactivity was measured in tumor, contralateral regions. Quantitative analysis of the regional radioactivity was performed using MCID program (Interfocus Imaging).

Immunocompromised female NuNu mice (Charles River Laboratories) were inoculated with 175,000 eGFP-transfected MDA-MB-231BR-Her2 cells (obtained as a gift from Pat Steeg, NIH, Womens Malignancies Branch, Bethesda, MD, in 2007 and passaged less than 6 months at the time of the experiments; were authenticated with STR profiling) via left cardiac ventricle in serum-free media. Tumors seeded the brain and were allowed to grow for 4–6 weeks. After tumor development, animals were anesthetized and administered and ¹⁴C-AIB and TX Red 625 Da or TX Red 70 kDa dextran into femoral vein and allowed to circulate for 5 to 60 minutes. At the end of the circulation period, the animals were euthanized by severing the cardiac ventricles and the brain was rapidly removed from the skull and flash frozen in isopentane (−50°C), cut coronally (20 μm), placed on slides and air dried.

Fluorescence was observed with an Olympus MVX-10 stereo microscope using a 2× objective (numerical aperture; 0.5) and an optical zoom of 0.63 to 6.3×. The excitation and emission of TX Red and or the TX Red dextrans was accomplished using a chroma filter (Chroma Technology Corp.) with an excitation filter of 560 ± 55 nm and an emission filter of 645 ± 75 nm. The dichromatic mirror inhibited the capture of emission wavelengths of light below 595 nm (Olympus America Inc.). Tumor-bearing brain slices were analyzed using a binary mask methodology (i.e., voxel-defined regions of interest) based upon eGFP fluorescence to define the tumor (Slidebook 4.2). Briefly, binary masking consisted of defining the tumor based upon the presence of eGFP fluorescence on a voxel-by-voxel basis. If eGFP fluorescence was present (approximately >3-fold above background), it was considered to be tumor, which was further confirmed by cresyl violet staining.

K_in was determined with linear regression using least squares analysis. One-way ANOVA analysis followed by a Bonferroni multiple comparison test was used for the evaluation of regional brain uptake data. All data represent mean ± SD unless otherwise indicated. Differences were considered significant at the P < 0.05 level. (Graphpad Prism version 5.00 for Windows, GraphPad Software).

To determine permeability changes, we developed a quantitative fluorescence microscopy technique coupled to autoradiography that provides approximately 1-μm spatial resolution to evaluate permeability. Briefly, in the first set of experiments, we ascertained the terminal concentrations of the permeability markers in blood, brain, and RG-2 tumors using fluorescent standards to confirm previous results obtained with autoradiography and chromatography methods. As concentration of the marker changed with time, we determined the blood–concentration time profiles at varying times from one to 60 minutes after injection (Fig. 1A). A significant correlation (r² = 0.99) was observed between the molecular weight of the fluorescent marker and clearance (Fig. 1B), agreeing with previous literature showing dextran conjugates undergo renal elimination with filtration rates being dependent on size (29).

Tumor-bearing brain slices were analyzed using a binary mask region-of-interest methodology based upon eGFP fluorescence to define the tumor (Fig. 2A–D). Binary masking consisted of defining the tumor based upon the presence of eGFP fluorescence on a voxel-by-voxel basis. If eGFP fluorescence was present (approximately >3-fold above background), it was considered to be tumor. Once blood and tumor concentrations of the permeability markers were determined, we utilized an intravenous multiple uptake time approach to determine the transfer coefficient (K_in) in normal brain and tumor (Fig. 2E and F; refs. 1, 2). The control K_in obtained from similar plots for all the molecules are shown in the first column of Fig. 2G. The ordinate intercept for the TX Red 625 Da and AIB was approximately 0.003 ± 0.008 mL/g (Fig. 2E) and approximately 0.004 ± 0.009 mL/g (Fig. 2F), respectively, which represents the vascular space (V_v) within brain. Similarly, the V_v was determined using similar plots for TX Red 3-kDa dextran (0.0054 ± 0.0014 mL/g) and TX Red 70-kDa dextran (0.0013 ± 0.0004). These data are consistent with the previous work (30, 31).

One significant advantage of this method is that simultaneous permeability measurements of different sized markers can be done in the same tumor, where in previous studies multiple tumors in different animals were needed to determine whether there were differences in size-dependent permeability (3, 17). As an example in this set of experiments, we injected both ¹⁴C-AIB and TX Red 625 Da in the same animal (Fig. 2). At the time of sacrifice, both brain and tumor (Fig. 2A) concentrations of the fluorophores were determined using microscopy (representative raw image data shown in Fig. 2B and C) and ¹⁴C-AIB concentrations were determined using QAR (representative raw image data shown in Fig. 2D). The transfer coefficient K_in for ¹⁴C-AIB and TX Red 625 Da in the tumor were 42 ± 1 × 10⁻⁵ mL/s/g and approximately 12 ± 0.2 × 10⁻⁵ mL/s/g. The magnitude of permeability changes (Fig. 2E) compared with normal brain (¹⁴C-AIB: 2.9 ± 1.1 × 10⁻⁵ mL/s/g; TX Red 625 Da: ∼1.3 ± 0.1 × 10⁻⁴ mL/s/g) agreed with previously reported values. Similar to the control permeability data, linkage of the dye to a progressively larger dextran reduced vascular permeability changes in the tumor. For example, the K_in for TX Red 3 kDa in tumor was only increased 5-fold in the tumor and the larger TX Red 70 kDa had only a 2-fold increase (control: 0.63 ± 0.26 × 10⁻⁵ mL/s/g; tumor: 1.8 ± 0.3 × 10⁻⁵ mL/s/g; Fig. 2E). Importantly, the ratio of unidirectional uptake of TX Red 625 Da, TX Red 70 kDa, and ¹⁴C-AIB were nearly equal to the ratio of the tracers respective aqueous diffusion constant; D_w (Table 1). These data suggest all three sized molecules can freely diffuse from blood to the glioma across vascular pores without steric restriction suggesting a pore size in the glioma model of >140 nm in diameter. These data, while obtained simultaneously in a single tumor, are consistent with the previous literature evaluating pore sizes in the vasculature of multiple gliomas (5).

Furthermore, using data from simultaneous injections, and from all three fluorescent tracers, we were able to calculate the surface area of the pores (Eq. H) present in the glioma vasculature. Briefly, the total surface area of the pores, which allow water soluble tracers to diffuse into tumor tissue, was approximately 4.3 × 10⁻⁵ cm²/g. These data are consistent with previous work in RG-2 tumor models showing a calculated pore surface area of approximately 6.2 × 10⁻⁵ cm²/g (5). While our value is slightly lower, it may represent the difference between tumor stages; where 10 days of RG-2 tumor growth can increase vascular pore permeability by approximately 20%–30% (32), ours were implanted for 7 days to minimize peripheral pathophysiologic changes and to keep tumor and tumor vasculature normalized between animals.

Once we had determined that our method was in general agreement for primary gliomas both in terms of regional permeability changes and in pore sizes/surface area, we applied the method to study brain metastases of breast cancer. We observed that changes in permeability from normal brain differed broadly among individual metastasis; representative images are shown in Fig. 3A–D. As can be seen qualitatively in Fig. 3, lesions one and two have an increased permeability compared with normal brain, yet are substantially less than lesion five. Of interest the mid-line lesion in Fig. 3 had the highest permeability in this sample. We have recently shown that increased expression of a desmin+ subpopulation of pericytes and reduced CD13⁺ pericytes and laminin α2 correlates with increased permeability (33). The heterogeneous brain metastases of breast cancer model data are in sharp contrast to the relatively homogenous glioma to glioma permeability observed in different animals. On the basis of historical reports (18), it is important to note that the BTB permeability had only a weak correlation with metastasis size (Fig. 3E and F). Quantitatively, of 174 MDA-MB-231BR-Her2 metastases, approximately 9% showed no statistical difference in permeability form normal brain, whereas approximately 75% showed relatively modest permeability increases. Only approximately 15% of the metastases demonstrated TX Red 625 Da permeability changes that were greater than 5-fold.

While there were significant permeability changes in most metastatic lesions in both models, the changes were much less on average compared with the glioma model. Therefore, we set out to determine whether there were differences in the size of the pores in the metastatic vasculature that would limit accumulation of the tracer. In our analysis of the data, we observed a highly significant correlation (r² = 0.81) between the uptake of ¹⁴C-AIB and TX Red 625 Da measured in the same metastases (Fig. 3E). However, the observed diffusion rate for TX Red 625 Da was only approximately 50% of what would be predicted (Table 2, Fig. 3E red line) if the molecule was able to freely diffuse across vascular pores without steric hindrance. To verify this in a second set of experiments, we simultaneously injected the larger TX Red 70-kDa dextran with ¹⁴C-AIB in animal that had metastatic lesions. Evaluation of this data showed no correlation of uptake with ¹⁴C-AIB (Fig. 3F) and the uptake of the larger dextran was also not consistent with pore diffusivity parameters predicted by ¹⁴C-AIB (Table 2; Fig. 3F, red line). Using these data, we then calculated the pore size present in the metastatic lesions to be approximately 10-fold smaller than what was observed in the glioma vasculature. These data strongly suggest that the metastatic vasculature is not near as porous or leaky as what is found in the primary glioma model. The implication is that chemotherapeutic uptake in metastases will be significantly restricted in comparison with gliomas.

Surprisingly, while metastases are the most common intracranial lesion (34), there has been very little quantitative data on permeability changes in metastases, especially compared with the primary tumor literature. Most information on brain metastasis permeability in humans is qualitatively based upon CT or MRI enhancement (34) and there are only a few preclinical datasets that have used a physiologic model of brain metastasis where the hematologic transfer, implantation, and lesion growth in brain is preserved (4, 18, 35, 36). The permeability data presented herein is consistent with our previous report showing there is substantial variability between lesions (12), which differs from primary tumors (see Supplementary Figs. S1–S3 for images showing variability in tumor permeability to the different dyes).

In this work, we have examined the tumor vascular permeability using quantitative fluorescence microscopy. Crucial to the development of this methodology was the ability to quantify fluorescent dye concentration in tissue similar to what has been documented for radioisotopes in QAR studies. Briefly, this was accomplished by summing the fluorescence intensity of each fluorophore in each pixel in a defined region of interest (e.g., brain homogenates, brain, tumor, etc…). Then all images (standards and experimental) were captured using the same settings for gain, exposure, optical zoom, on naïve samples (focus was obtained in the eGFP channel) to maintain consistency of fluorescence emission (37). Using this method, the brain homogenates and blood standards had a range of linearity between fluorescence intensity and dye concentrations of approximately 3 log units, which is consistent with most QAR standards (12). This range was large enough for us to give peripheral doses of dyes that produced, at the time of sacrifice, brain and tumor dye concentrations that fell within the midpoint of the standards.

One limitation in these studies that had to be overcome was that the standard curves of fluorescence intensity versus concentration had different slopes in blood and brain. These data, similar to previous work (38), suggested that tissue matrices can alter fluorescence emission. Attenuation of fluorescence signal by blood is a well-documented phenomenon where the emitted photons of fluorescence can be transferred to the heme moieties present in the red blood cell (39). The consequences of quenching the TX Red signal in our study had the possibility of altering two pieces of critical data. First, quenching could change the accuracy of our peripheral pharmacokinetics, by not allowing an accurate determination of dye concentration in blood. However, our pharmacokinetic parameters (Fig. 1A; Table 3) are in close agreement with the two previous studies that used radiolabeled dextrans and scintillation counting (31). Furthermore, the relationship between clearance and the molecular weight of the dextran shown in Fig. 1B is consistent with the clearance of fluorescein-labeled dextrans in rats using high-performance liquid chromatography for analysis (29, 40).

The second concern that fluorescence quenching in blood created in this study was a potential misrepresentation of dye concentration for the end distribution volume of in brain. This could be compounded by differences in tissue accumulation in the active growing regions of the tumor versus a necrotic core. While some necrosis is evident in an RG-2 model, consistent with our previous work, the brain metastases of breast cancer model shows little evidence of necrosis (12, 21). While this cannot be easily rectified, we believe that the small volume of blood per mass of tissue (assumed to be ∼1% in normal brain and possibly as high as 4% in tumor or tumor core) did not significantly affect the calculated permeability value. For example, TX Red 3-kDa dextran has a molecular radius in fluid of approximately 15.9 A° (41, 42), for which we obtained a mean K_in value of approximately 4.0 × 10⁻⁶ mL/s/g in normal brain (Fig. 3). In comparison, ¹⁴C-inulin (MW 5.5 kDa) has a similar radius of approximately 15 A° (31), for which a permeability value of approximately 2.2 × 10⁻⁶ mL/s/g was observed using radiolabeled isotopes and similar mathematical modeling (43). The observed permeability changes (fold increase) in tumor, which is based upon size, is consistent with the magnitude of change observed in prior literature.

Once we had determined that our method was general validity in measuring the permeability alterations in tumor vasculature, we determined the size of pores in the tumors. Our data suggest that there is substantial discrepancy between the size of the molecule that can freely diffuse into gliomas and metastatic lesions. To make this comparison mathematically, we calculated the size of the vascular pores based upon theories of pore diffusivity. The pore theory model postulates water-soluble tracers that are trapped to a large degree in the normal brain vasculature will enter into tumor tissue through “cylindrical pores” or channels in the newly formed vasculature. Furthermore, if the pores are large enough, the molecules should enter into the tumor tissue at a transfer rate that is proportionately similar to their diffusion rates in water. For the glioma model, the largest Dextran tested (70 kDa) diffused across the tumor vasculature at approximately the same ratio rate (compared with ¹⁴C-AIB) as it would in water. This suggests that the 70-kDa dextran (∼12 nm in diameter) is able to freely diffuse from blood through angiogenic vascular pores into tumor. In order for this free diffusion to occur, the diameter of the vascular pore must be at least 12 times greater than the diameter of the largest molecule for which the diffusion constant remained proportionately constant (5), and accordingly the diameter of the vascular pore in the RG-2 tumor is ≥ 140 nm. These calculations assume that molecules with similar hydrodynamic radii have similar rates of permeability across a vascular pore. However, there are data demonstrating that nanorods have increased permeability rates across a tumor vessel compared with a similar hydrodynamic-sized nanosphere. It is thought that the nanorod may have reduced steric hindrance and viscous drag during vascular flow and as it inters the vascular pore (44).

While tertiary and quaternary structured proteins are different in shape to an aqueous dextran, in theory, the pore size of the glioblastoma vasculature is sufficiently large to allow the 149-kDa mAb (∼10 nm in size (45) to freely accumulate into tumor tissue, and may be a reason bevacizumab and other mAbs are showing promise in the treatment of gliomas (46). It can also be reasoned that VEGF (∼3 nm) can exit from the tumor into the circulation through these vascular pores. As the VEGR-2 is expressed approximately symmetrically on the luminal and abluminal side of the blood (47), we propose bevacizumab may pharmacologically act in the circulation as well to reduce effects of VEGF on the vasculature. This hypothesis is consistent and builds upon previous work that suggests normalization of the vasculature by blockade of the VEGFR-2 allows the permeability of smaller nanometer molecules (<12 nm) across the tumor vasculature (48). This may be a primary reason there are concerns that bevacizumab may normalize the BTB and limit chemotherapeutic efficacy and allow tumor growth (49, 50).

In contrast to the glioma model in which the pores were large enough to allow unrestricted passive diffusion of TX Red 70 kDa, in the brain metastases of breast cancer model, passive diffusion rates of the solutes across the BTB were observed to be approximately 40%–60% less than predicted unrestricted values based off of proportionate diffusion rates in water. This degree of restriction suggests that there is steric hindrance in the pore for free diffusion of large molecules (5). Using similar calculations from the glioma model the pore size may be only approximately 9 nm in diameter and may be as small as approximately 5 nm based upon the size of free TX Red. Although these are average measurements and the spread of the actual pore sizes in the vasculature is most likely very heterogeneous. Importantly, if the free dye and the 70-kDa dextran demonstrated significantly restricted permeability then trastuzumab at an approximate size of 5.5–6 nm will be restricted from diffusing across vascular pores into the metastatic lesion as well. However, we propose similar to gliomas, VEGF may be able to cross the vascular pores (though in a more restricted fashion) from the tumor into the circulation and this may be a factor in how bevacizumab is able to demonstrate antiangiogenic effects in brain metastases.

In summary, the data presented herein have shown in preclinical models that: (i) permeability of brain metastases is highly heterogenous and significantly reduced compared with gliomas; (ii) there are pore size restrictions that may limit the use of mAbs in metastatic lesions; and (iii) the location for the action for bevacizumab in brain metastases may be limited to the luminal side of the vasculature.

No potential conflicts of interest were disclosed.

Conception and design: R.K. Mittapalli, P.R. Lockman

Development of methodology: R.K. Mittapalli, C.E. Adkins, P.R. Lockman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.E. Adkins, A.S. Mohammad, J.A. Lockman, P.R. Lockman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.K. Mittapalli, C.E. Adkins, K.A. Bohn, A.S. Mohammad, P.R. Lockman

Writing, review, and/or revision of the manuscript: R.K. Mittapalli, C.E. Adkins, K.A. Bohn, A.S. Mohammad, J.A. Lockman, P.R. Lockman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.E. Adkins, A.S. Mohammad, J.A. Lockman, P.R. Lockman

Study supervision: A.S. Mohammad, P.R. Lockman

We thank Robert Thorne, PhD, for his valuable input into the manuscript.

This research was supported by grants from the National Cancer Institute (R01CA166067-01A1) and Department of Defense Breast Cancer Research Program (W81XWH-062-0033; awarded to P.R. Lockman). Additional support for this research was provided by WVCTSI through the National Institute of General Medical Sciences of the NIH (U54GM104942).

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.