The Blood-Brain Barrier at Sites of Metastasis
The Blood-Brain Barrier at Sites of Metastasis
We have described a new approach to permeabilize the vasculature of brain metastases that is based specifically on the stromal vascular phenotype of these tumors, with a view to enhancing detection and improving the delivery of chemotherapeutic agents. The 4T1-GFP mouse model of brain metastasis was used to generate micrometastases in the brain, which, at the age used here, rarely present with a permeable BBB. To ensure that all metastases within the brain were targeted, the cytokines LT or TNF were injected intravenously and were found to induce permeabilization of the BBB specifically at the sites of brain metastases, as revealed both ex vivo through histology (HRP) and in vivo through contrast-enhanced MRI (Gd-DTPA). We further replicated these findings in a second model of human-derived brain metastasis and again found that systemically administered TNF increased permeability focally at tumor sites. With this method, the commonly used treatment for breast cancer, trastuzumab (145kDa), which is excluded from the brain under normal conditions, was successfully delivered selectively to metastatic colonies in the brain as revealed by SPECT/CT imaging. We suggest that this permeabilization works primarily though the activation of TNFR1, which was found specifically on the vascular endothelium of vessels closely associated with brain metastases in both mouse models studied and also in human brain metastasis tissue. Our findings suggest that the degree of permeabilization would be sufficient for diagnostically and therapeutically relevant molecules to gain access to the tumor site.
Brain metastases in mice showed increased expression of both receptors for LT and TNF (TNFR1 and TNFR2) before any cytokine treatment. TNFR1 expression was exclusively found on blood vessels within the metastases and colocalized with the endothelial marker Glut-1. These findings suggest the potential for systemic administration of either TNF or LT to alter BBB function that is likely to be through activation of TNFR1. The downstream effects of endothelial TNFR1 stimulation include alterations to cytoskeletal proteins, cell–cell adhesion molecules, and changes to paracellular permeability of the BBB. Moreover, in vitro studies have shown that TNF can increase permeability of endothelial cells.
At sites of brain metastasis, expression of TNFR2 was also clearly increased but restricted to what appear to be recruited intravascular leukocytes and parenchymal microglia. This receptor expression pattern is in accord with previous studies reporting TNFR2 expression by immune system cells. Given this and the fact that soluble TNF cannot induce full activation of TNFR2, it appears likely that this receptor plays a lesser or secondary role in the cytokine-induced permeabilization observed in this work.
After treatment with either cytokine, the intravenous tracer HRP was found within metastases in the brain, indicating compromise of the BBB. Increasing cytokine dose was accompanied by an increase in the frequency of metastases exhibiting a permeable BBB. Breakdown of the BBB was specific and local to the sites of brain metastases, and no other permeabilization of the BBB was observed in normal brain tissue. These findings support the concept that upregulation of TNFR1 on tumor-associated vessels alone enables specific permeabilization of the tumor vasculature in response to systemic LT or TNF. This effect was maintained for all of the time points studied here. The greatest number of HRP-positive metastases was evident at 6 hours, whereas the lower number of positive metastases at the 24-hour time point suggests that this is an extended but transient opening of the BBB. The concentration of the administered cytokine rapidly returns to baseline levels, which supports the concept that endothelial TNFR1 signaling pathways, subsequent to receptor activation, give rise to reorganization of the junctional complexes beyond the clearance time of the cytokine bolus. The length of time of the increase in permeability suggests that the vasculature may be structurally compromised by the treatment, and TNF is known to be cytotoxic to endothelial cells. However, we found no evidence of hemorrhage, suggesting that frank vascular injury is not the mechanism for permeabilization (Supplementary Figure 1, B–D, available online).
LT elicited a less marked response than TNF at each dose and time point, which is in agreement with previous studies that describe a less stable complex formation of lymphotoxin with TNFR1 compared with TNF, thus inducing a lower signaling capability.
Numerous local hyperintensities were seen on post-Gd-DTPA T1-weighted images after treatment with LT or TNF. Although these regions were associated with sites of metastasis shown histologically, other sites of metastasis were not detectable with MRI. The apparent lack of permeability may be because of the limited resolution of our in vivo scans (approximately 160 μm), giving a higher size detection threshold than the histological method (approximately 1 μm). Alternatively, it is possible that the micrometastases must show a particular growth pattern or cell recruitment profile before TNFR1 expression is sufficient to allow permeabilization. Nevertheless, even metastases of approximately 50 μm diameter showed sensitivity to cytokine-induced permeabilization, which is considerably smaller than those currently detectable clinically owing to natural BBB permeability (0.5–1.0cm diameter).
Trastuzumab is a monoclonal antibody that is active against HER2-overexpressing breast cancer, leading to reduced breast tumor burden and increased patient survival. Extravasation of circulating antibodies and binding to the Her2 receptor leads to a number of antitumor actions [reviewed in]. However, trastuzumab is ineffective in treating brain metastases, in part because of restricted access to the Her2 receptor once the metastatic tumor cells are sequestered on the brain side of the BBB, because it does not effectively cross the BBB. Similarly, in the 4T1-GFP model used here, trastuzumab was excluded from brain metastases in saline-treated controls. However, in mice treated systemically with LT or TNF, the antibody was found to permeate the BBB and accumulate within the brain to a level detectable by SPECT imaging. Metastases were present at all sites of intracerebral SPECT signal but were much smaller than the volume of signal presenting on the SPECT/CT image. Thus, although accurately displaying the amount of radioactivity, the partial volume effect and potential spread of antibody from point of delivery may over-represent the metastasis size. As with our MRI studies, some metastases were present in the brain that did not appear to accumulate radiolabelled antibody. Again, this may reflect the low spatial resolution of SPECT detection (approximately 1mm) precluding detection of small micrometastases. Once again the metastases that were detected and thus exposed to the therapeutic compound were well below the detection threshold currently possible clinically and clearly represent metastases that would not otherwise be accessible to the therapeutic agent. Thus, this approach confers a substantial advantage for treatment of early brain metastases when efficacy may be greatly enhanced.
Six cases of human brain metastasis were analyzed, with different primary tumor origins. Because these metastases had been clinically detected and excised, they were substantially larger than those in the mouse model. Some degree of necrosis and mucus deposition was present, yet critically a similar expression profile for TNFR1 and TNFR2 was seen. In particular, TNFR1 was heterogeneously expressed on metastasis-associated vessels and was not found in nonmetastasis brain tissue. These findings suggest that a similar selectivity in response to a TNF/LT dose could be elicited in human brain metastases. Additionally, TNFR1 and TNFR2 expression was apparent on other cell types in the close vicinity of the metastasis. However, it is less likely that the nonendothelial TNF receptors would contribute substantially to the endothelial changes in the BBB integrity.
Previous work involving bradykinin analogs in glioma models demonstrated a peak of increased drug delivery after 15 minutes of RMP-7 infusion, a size limitation for access to the brain of 1 kDa–sized molecules and dose-limiting side-effects. Interestingly, it has been suggested that a possible mechanism of action of bradykinin involves the accelerated release of TNF. The approach described here of a direct systemic administration of TNF or its endogenous analog LT may have three major advantages over the use of bradykinin. First, the window of permeability shown here appears to peak 6 hours after cytokine administration and to be maintained to at least 24 hours. This extended window of permeability would increase the opportunity for intravenous drugs to bind to their targets. Second, entry of a range of molecules was facilitated with this approach, from gadolinium-DTPA (590Da) to HRP (44kDa) and up to the therapeutic monoclonal antibody, trastuzumab (approximately 148kDa), suggesting that drug size may not be prohibitive. Finally, although conscious of the potential concerns over TNF toxicity, the human equivalent of the range of doses used in the mouse model here would elicit the desired response within the maximum tolerable dose [150 μg/m]. The toxicity profile of lymphotoxin has been reported to be substantially below that of TNF at similar levels of antitumor activity. Further, at the dose of TNF used here, no statistically significant differences were found between any of the groups for markers of the acute phase response and hepatoxicity (Supplementary Methods and Supplementary Figure 4, A–E, available online). It is also worth noting the very short half-life of TNF within the circulation [2.8min], and we would, therefore, expect any effects of the cytokine bolus on the biology of the tumor to be minimal. However, to test this possibility we used an adenovirus to induce prolonged expression of systemic TNF and found no statistically significant difference in the number or volume of metastases within the brain compared with animals injected with either the equivalent null adenovirus or no virus (Supplementary Methods and Supplementary Figure 5, A and B, available online). There are also known interactions between HER2 and TNF signaling pathways but, once again, we would expect the short half-life of the cytokine bolus to minimize any unpredictable interactions. Moreover, given the proposed mechanism of TNF action through TNFR1, the use of selective TNFR1 agonists [eg, htr-9–specific TNFR1 antibody] or selective TNF muteins [eg, LK805] may provide a more targeted approach with fewer potential side effects.
This study also had some limitations. For example, although we have shown that TNF will increase local permeability in two mouse models of metastasis and the pattern of receptor expression is similar in human metastatic cancer, species-specific alterations in TNFR1 and TNFR2 signaling pathways may give rise to different effects in human. Our findings now need to be validated in a limited clinical trial. Additionally, in our models there is no primary tumor, which may alter the systemic inflammatory phenotype. However, it is often the case that the primary tumor has been successfully removed in individuals with brain metastases; hence our models may actually have good clinical relevance in terms of disease burden.
The work presented here shows a novel approach to facilitating the delivery of therapeutic and diagnostic agents to cerebral metastases by exploiting a previously unknown phenotype of the vasculature of brain metastases. Critically, even small metastases (approximately 200-fold smaller than those currently detectable in the clinic) are targeted with this approach. The identification of a similar vascular phenotype in human brain metastasis tissue indicates the potential for clinical translation. This work has demonstrated that cytokine-enhanced drug delivery to brain metastases is possible and that this strategy may be critically important for the detection and treatment of brain metastases clinically.
Discussion
We have described a new approach to permeabilize the vasculature of brain metastases that is based specifically on the stromal vascular phenotype of these tumors, with a view to enhancing detection and improving the delivery of chemotherapeutic agents. The 4T1-GFP mouse model of brain metastasis was used to generate micrometastases in the brain, which, at the age used here, rarely present with a permeable BBB. To ensure that all metastases within the brain were targeted, the cytokines LT or TNF were injected intravenously and were found to induce permeabilization of the BBB specifically at the sites of brain metastases, as revealed both ex vivo through histology (HRP) and in vivo through contrast-enhanced MRI (Gd-DTPA). We further replicated these findings in a second model of human-derived brain metastasis and again found that systemically administered TNF increased permeability focally at tumor sites. With this method, the commonly used treatment for breast cancer, trastuzumab (145kDa), which is excluded from the brain under normal conditions, was successfully delivered selectively to metastatic colonies in the brain as revealed by SPECT/CT imaging. We suggest that this permeabilization works primarily though the activation of TNFR1, which was found specifically on the vascular endothelium of vessels closely associated with brain metastases in both mouse models studied and also in human brain metastasis tissue. Our findings suggest that the degree of permeabilization would be sufficient for diagnostically and therapeutically relevant molecules to gain access to the tumor site.
Brain metastases in mice showed increased expression of both receptors for LT and TNF (TNFR1 and TNFR2) before any cytokine treatment. TNFR1 expression was exclusively found on blood vessels within the metastases and colocalized with the endothelial marker Glut-1. These findings suggest the potential for systemic administration of either TNF or LT to alter BBB function that is likely to be through activation of TNFR1. The downstream effects of endothelial TNFR1 stimulation include alterations to cytoskeletal proteins, cell–cell adhesion molecules, and changes to paracellular permeability of the BBB. Moreover, in vitro studies have shown that TNF can increase permeability of endothelial cells.
At sites of brain metastasis, expression of TNFR2 was also clearly increased but restricted to what appear to be recruited intravascular leukocytes and parenchymal microglia. This receptor expression pattern is in accord with previous studies reporting TNFR2 expression by immune system cells. Given this and the fact that soluble TNF cannot induce full activation of TNFR2, it appears likely that this receptor plays a lesser or secondary role in the cytokine-induced permeabilization observed in this work.
After treatment with either cytokine, the intravenous tracer HRP was found within metastases in the brain, indicating compromise of the BBB. Increasing cytokine dose was accompanied by an increase in the frequency of metastases exhibiting a permeable BBB. Breakdown of the BBB was specific and local to the sites of brain metastases, and no other permeabilization of the BBB was observed in normal brain tissue. These findings support the concept that upregulation of TNFR1 on tumor-associated vessels alone enables specific permeabilization of the tumor vasculature in response to systemic LT or TNF. This effect was maintained for all of the time points studied here. The greatest number of HRP-positive metastases was evident at 6 hours, whereas the lower number of positive metastases at the 24-hour time point suggests that this is an extended but transient opening of the BBB. The concentration of the administered cytokine rapidly returns to baseline levels, which supports the concept that endothelial TNFR1 signaling pathways, subsequent to receptor activation, give rise to reorganization of the junctional complexes beyond the clearance time of the cytokine bolus. The length of time of the increase in permeability suggests that the vasculature may be structurally compromised by the treatment, and TNF is known to be cytotoxic to endothelial cells. However, we found no evidence of hemorrhage, suggesting that frank vascular injury is not the mechanism for permeabilization (Supplementary Figure 1, B–D, available online).
LT elicited a less marked response than TNF at each dose and time point, which is in agreement with previous studies that describe a less stable complex formation of lymphotoxin with TNFR1 compared with TNF, thus inducing a lower signaling capability.
Numerous local hyperintensities were seen on post-Gd-DTPA T1-weighted images after treatment with LT or TNF. Although these regions were associated with sites of metastasis shown histologically, other sites of metastasis were not detectable with MRI. The apparent lack of permeability may be because of the limited resolution of our in vivo scans (approximately 160 μm), giving a higher size detection threshold than the histological method (approximately 1 μm). Alternatively, it is possible that the micrometastases must show a particular growth pattern or cell recruitment profile before TNFR1 expression is sufficient to allow permeabilization. Nevertheless, even metastases of approximately 50 μm diameter showed sensitivity to cytokine-induced permeabilization, which is considerably smaller than those currently detectable clinically owing to natural BBB permeability (0.5–1.0cm diameter).
Trastuzumab is a monoclonal antibody that is active against HER2-overexpressing breast cancer, leading to reduced breast tumor burden and increased patient survival. Extravasation of circulating antibodies and binding to the Her2 receptor leads to a number of antitumor actions [reviewed in]. However, trastuzumab is ineffective in treating brain metastases, in part because of restricted access to the Her2 receptor once the metastatic tumor cells are sequestered on the brain side of the BBB, because it does not effectively cross the BBB. Similarly, in the 4T1-GFP model used here, trastuzumab was excluded from brain metastases in saline-treated controls. However, in mice treated systemically with LT or TNF, the antibody was found to permeate the BBB and accumulate within the brain to a level detectable by SPECT imaging. Metastases were present at all sites of intracerebral SPECT signal but were much smaller than the volume of signal presenting on the SPECT/CT image. Thus, although accurately displaying the amount of radioactivity, the partial volume effect and potential spread of antibody from point of delivery may over-represent the metastasis size. As with our MRI studies, some metastases were present in the brain that did not appear to accumulate radiolabelled antibody. Again, this may reflect the low spatial resolution of SPECT detection (approximately 1mm) precluding detection of small micrometastases. Once again the metastases that were detected and thus exposed to the therapeutic compound were well below the detection threshold currently possible clinically and clearly represent metastases that would not otherwise be accessible to the therapeutic agent. Thus, this approach confers a substantial advantage for treatment of early brain metastases when efficacy may be greatly enhanced.
Six cases of human brain metastasis were analyzed, with different primary tumor origins. Because these metastases had been clinically detected and excised, they were substantially larger than those in the mouse model. Some degree of necrosis and mucus deposition was present, yet critically a similar expression profile for TNFR1 and TNFR2 was seen. In particular, TNFR1 was heterogeneously expressed on metastasis-associated vessels and was not found in nonmetastasis brain tissue. These findings suggest that a similar selectivity in response to a TNF/LT dose could be elicited in human brain metastases. Additionally, TNFR1 and TNFR2 expression was apparent on other cell types in the close vicinity of the metastasis. However, it is less likely that the nonendothelial TNF receptors would contribute substantially to the endothelial changes in the BBB integrity.
Previous work involving bradykinin analogs in glioma models demonstrated a peak of increased drug delivery after 15 minutes of RMP-7 infusion, a size limitation for access to the brain of 1 kDa–sized molecules and dose-limiting side-effects. Interestingly, it has been suggested that a possible mechanism of action of bradykinin involves the accelerated release of TNF. The approach described here of a direct systemic administration of TNF or its endogenous analog LT may have three major advantages over the use of bradykinin. First, the window of permeability shown here appears to peak 6 hours after cytokine administration and to be maintained to at least 24 hours. This extended window of permeability would increase the opportunity for intravenous drugs to bind to their targets. Second, entry of a range of molecules was facilitated with this approach, from gadolinium-DTPA (590Da) to HRP (44kDa) and up to the therapeutic monoclonal antibody, trastuzumab (approximately 148kDa), suggesting that drug size may not be prohibitive. Finally, although conscious of the potential concerns over TNF toxicity, the human equivalent of the range of doses used in the mouse model here would elicit the desired response within the maximum tolerable dose [150 μg/m]. The toxicity profile of lymphotoxin has been reported to be substantially below that of TNF at similar levels of antitumor activity. Further, at the dose of TNF used here, no statistically significant differences were found between any of the groups for markers of the acute phase response and hepatoxicity (Supplementary Methods and Supplementary Figure 4, A–E, available online). It is also worth noting the very short half-life of TNF within the circulation [2.8min], and we would, therefore, expect any effects of the cytokine bolus on the biology of the tumor to be minimal. However, to test this possibility we used an adenovirus to induce prolonged expression of systemic TNF and found no statistically significant difference in the number or volume of metastases within the brain compared with animals injected with either the equivalent null adenovirus or no virus (Supplementary Methods and Supplementary Figure 5, A and B, available online). There are also known interactions between HER2 and TNF signaling pathways but, once again, we would expect the short half-life of the cytokine bolus to minimize any unpredictable interactions. Moreover, given the proposed mechanism of TNF action through TNFR1, the use of selective TNFR1 agonists [eg, htr-9–specific TNFR1 antibody] or selective TNF muteins [eg, LK805] may provide a more targeted approach with fewer potential side effects.
This study also had some limitations. For example, although we have shown that TNF will increase local permeability in two mouse models of metastasis and the pattern of receptor expression is similar in human metastatic cancer, species-specific alterations in TNFR1 and TNFR2 signaling pathways may give rise to different effects in human. Our findings now need to be validated in a limited clinical trial. Additionally, in our models there is no primary tumor, which may alter the systemic inflammatory phenotype. However, it is often the case that the primary tumor has been successfully removed in individuals with brain metastases; hence our models may actually have good clinical relevance in terms of disease burden.
The work presented here shows a novel approach to facilitating the delivery of therapeutic and diagnostic agents to cerebral metastases by exploiting a previously unknown phenotype of the vasculature of brain metastases. Critically, even small metastases (approximately 200-fold smaller than those currently detectable in the clinic) are targeted with this approach. The identification of a similar vascular phenotype in human brain metastasis tissue indicates the potential for clinical translation. This work has demonstrated that cytokine-enhanced drug delivery to brain metastases is possible and that this strategy may be critically important for the detection and treatment of brain metastases clinically.