STAT3-Mediated Astrocyte Reactivity Associated with Brain Metastasis Contributes to Neurovascular Dysfunction

Neurovascular coupling matches local cerebral blood flow (CBF) to neuronal energy use and, thus, ensures normal functioning of the brain (1). While neuronal-, astrocyte-, and pericyte-mediated pathways have been implicated in neurovascular coupling to varying degrees (2), the significant contribution of astrocytes remains the most widely recognized (1–3).

Astrocyte processes cover most of the cerebrovascular surface and connect neurons with blood vessels (3–5), playing essential roles in recycling ions and neurotransmitters (3), supplying energy substrates to neurons (6), maintaining endothelial tight junctions to support the blood–brain barrier (BBB; ref. 7), and releasing vasoactive molecules that regulate vascular tone (1). However, in response to disease or injury, astrocytes become reactive, displaying hypertrophy, upregulation of intermediate filament proteins (e.g., GFAP; ref. 8), and metabolic changes (9). The consequences of this astrocyte reactivity, however, on the control of CBF remains unknown (3).

Changes in astrocyte function may be particularly pertinent in brain metastasis, where tumor cells interact closely with blood vessels. Such perivascular growth enables tumor cells to use the brain's vasculature for nutrient supply (10–13), but also disrupts astrocyte-vascular coupling and BBB integrity (12). At the same time, brain metastases trigger astrocyte reactivity from the earliest stages (10–13), and gradually expel astrocyte end-feet to the border of the tumor (14). Notably, STAT3-mediated astrocyte reactivity, specifically, has been shown to be important in maintaining the vascular co-option profile of brain metastases (15). Consequently, brain metastases may impact on cerebrovascular function not only through mechanical dissociation of astrocyte end-feet from vessels (12), but also through astrocyte reactivity per se and, as such, may contribute to the neurologic sequelae associated with brain metastases. However, a lack of suitable animal models and imaging techniques that are able to separate these two processes, to date, have limited our ability to fully understand their relative contributions.

The aim of this study, therefore, was to determine the impact of reactive astrocytes, specifically, on cerebrovascular function using animal models of brain metastasis and ciliary neurotrophic factor (CNTF)-induced astrocyte reactivity (16, 17), together with human brain metastasis biopsies. The CNTF model induces astrocyte reactivity, as observed in disease, in the absence of any other confounding pathology. Subsequently, the potential of inhibiting STAT3-mediated astrocyte reactivity to maintain vascular function was determined, as a potential adjuvant therapy for brain metastasis. Cortical vascular and neuronal function were measured using a multimodal imaging strategy involving MRI, laser speckle contrast imaging (LSCI) and local field potential (LFP) measurements, in combination with postmortem IHC analysis.

All animal experiments were approved by the University of Oxford Clinical Medicine Ethics Review Committee and UK Home Office [Animals (Scientific Procedures) Act 1986], and conducted in accordance with the University of Oxford Policy on the Use of Animals in Scientific Research, the ARRIVE Guidelines and Guidelines for the Welfare and Use of Animals in Cancer Research (18). No overt clinical signs or weight loss were evident, in any cohort, over the experimental period. For in vivo experiments, blood gases were within the normal ranges for animals under isoflurane anesthesia (Supplementary Table S1). No effects of WP1066 treatment on blood gases and mean arterial blood pressure were observed (Supplementary Table S1).

For the brain metastasis model, an ENU1564 tumor cell line was used (kind gift from G. Stoica, Texas A&M University, Texas), which originated from an N-ethyl-N-nitrosourea–induced mammary adenocarcinoma in a female Berlin-Druckrey IX (BD-IX) rat, and is highly metastatic to brain and bone tissues. As a rat cell line, there is no publicly available short tandem repeat profile for the cells, thus they have not been externally validated. Cells were tested for Mycoplasma immediately prior to freezing each stock vial using MycoAlert Kit (Lonza). Stock vials were revived, grown for up to 1 week in DMEM supplemented with 10% FBS and 1% 200 mmol/L glutamine, harvested by trypsinization, washed and then injected, as described previously (19).

For the reactive astrocyte model, lentivirus-mediated gene transfer of CNTF (CNTF-Lv) was used to induce sustained production of CNTF by neurons, which is released and then gives rise to stable activation of astrocytes via the STAT3 pathway, as described previously (9, 16, 17). To this end, self-inactivated lentiviruses encoding either the human CNTF gene or the (control) β-galactosidase gene (LacZ-Lv), under the control of the mouse phosphoglycerate kinase I promoter, were used (9, 17). In the CNTF-Lv model, transfected neurons display a normal phenotype in terms of morphology, expression of neuronal proteins and spontaneous electrophysiological activity (9, 17).

For STAT3 inhibition, a potent inhibitor that can cross the BBB, WP1066 (Millipore) was injected intraperitoneal daily (3 mg/kg) using DMSO as a vehicle.

All experimental procedures used in this study were approved by the UK Home Office. Two cohorts of female BD-IX rats (Charles River), 6–12 weeks old, were injected intracortically in the whisker barrel somatosensory cortex (2 mm posterior, 5 mm lateral to Bregma, and 1 mm from the cortical surface). One cohort of rats was injected with either 10³ metastatic ENU1564 cells in PBS (n = 8) or PBS alone (n = 6). The second cohort of rats was injected with either CNTF-Lv (n = 9) or LacZ-Lv (n = 7) diluted in PBS at a concentration of 100 ng p24/μL. The third cohort of rats was injected with 10³ metastatic ENU1564 cells (n = 10). Animals injected with ENU1564 cells were randomly distributed for intraperitoneal injection of either 50 μL WP1066 in DMSO (3 mg/kg, n = 5) or 50 μL DMSO (n = 5).

All animals were transcardially perfusion-fixed under terminal anesthesia at the end of the experiment for histologic analysis. For full details, please see Supplementary Methods.

All MRI experiments were performed on a horizontal bore 9.4T spectrometer driven by an Agilent DirectDrive console (Agilent Technologies). Animals were anesthetized with 1%–2% isoflurane in 70%N₂/30%O₂ and positioned in a 72-mm quadrature volume transmit coil with 4-channel phased-array surface receiver (RAPID Biomedical GmbH). In some animals, a cannula was positioned in the tail vein prior to the MRI experiment for gadolinium-DTPA (Gd-DTPA) injection to identify potential BBB breakdown.

Multimodal acquisitions included T₁- and T₂-weighted MRI and arterial spin labeling (ASL) MRI. For technical details, please visit the Supplementary Methods section.

Through the cranial window, superficial (ascending venules and pial vessels) and deep blood vessels (penetrating arterioles and capillary beds) within the cortex were visible (0.5–1 mm deep). Animals underwent LSCI on the day following MRI measurements. For full details regarding LSCI setup and surgery, see Supplementary Data (20, 21).

In a subset of animals, LFPs were recorded to measure neuronal responses to the same electrical stimulation of the whisker pad (see above). In this case, burr holes were drilled over both barrel cortices and an electrode (0.155 mm diameter, Teflon insulated platinum; Bilaney Consultants Ltd.) inserted to a depth of 0.5 mm under stereotaxic control, as described previously (22). Electrode signals were recorded using a CED 1902 isolated pre-amplifier and signal processor (Cambridge Electronic Design limited), which was controlled using Spike2.7 data acquisition and analysis package software (Cambridge Electronic Design limited). Electrode signals were acquired with a waveform sampling rate of 10 kHz, 2 μs per time unit resolution and without notch filtering. Low-frequency noise was eliminated from the electrode signal using a 50/60 Hz noise eliminator (Hum Bug, Quest scientific).

Signal intensity on T₁- and T₂-weighted images in the ipsilateral and contralateral cortices of animals was quantified using ImageJ software package (NIH; ref. 23). CBF maps were quantified with nonlinear perfusion signal modelling (24) using the BASIL (Bayesian Inference for Arterial Spin Labeling MRI) toolset in FSL version 5.0 (FMRIB Software Library). For more details regarding data analysis, please see Supplementary Data (24).

LSCI datasets were redefined and downsampled to 1 Hz for data analysis. To quantify the time course of the CBF response after electrical stimuli, datasets were averaged over 10 trials and regions of interest (ROI) drawn around areas corresponding to the whisker barrel cortex. The CBF response and its associated time series were extracted from the ROIs in both ipsilateral (left) and contralateral (right) whisker barrel cortices. More information regarding data acquisition can be found in Supplementary Data.

At the end of the LFP recording period, waveform and sampling times were converted into a MATLAB file using Spike 2.7 software. The data were aligned with the stimulation time series and binned at 100Hz. LFP responses to whisker pad stimulation were extracted from 0 to 16 seconds after stimulus using in-house MATLAB scripts. To quantify neuronal response magnitudes, amplitude (mV) was calculated from the average response to the first pulse in the stimulation train (over a 50 ms period following stimulation onset), as described previously (22).

All animals were transcardially perfusion fixed under terminal anesthesia with 0.9% heparinized saline followed by periodate lysine paraformaldehyde (PLP) containing only 0.025% glutaraldehyde (PLP_light) at the end of LSCI and LFP measurements. The brains were postfixed, cryoprotected, embedded in tissue-tek (Sakura Finetek Europe) and frozen in isopentane at −40°C. Frozen, 20 μm thick, serial sections spanning the somatosensory cortex were cut from fixed tissue and mounted on gelatin-coated glass slides. IHC was performed with antibodies against markers of activated astrocytes (1:300, GFAP, Dako), STAT3 (1:300, STAT3α, Cell Signaling Technology), neurons (1:500, NeuN, Millipore), blood vessels (1:100, CD31, Abcam), microglia [1:200, OX42 (cd11b/c), Abcam], cyclooxygenase-1 and 2 (1:100, COX-1/2; Abcam), inducible isoform of nitric oxide synthetase (1:50, iNOS, Abcam), nitrotyrosine (1:400, Millipore), alpha-smooth muscle actin (1:100, αSMA; Abcam), β-dystroglycan (1:200, Abcam), Claudin-5 (1:200, Abcam) and cytochrome p450-4A (1:100, Abcam), the enzyme responsible for the biosynthesis of 20-hydroxyeicosatetraenoic acid (20-HETE).

Immunofluorescent microscopy was performed to identify reactive astrocytes (GFAP), microglia (OX42), neurons (NeuN), blood vessels (CD31), cyclooxygenase-1 and 2 (COX-1/2), inducible isoform of nitric oxide synthase (iNOS), cytochrome p450-4A, αSMA, β-dystroglycan, and Claudin-5. More details about the performed protocols can be found in Supplementary Methods.

Photomicrographs of each brain section were obtained by scanning each section with Aperio Image Analysis (Leica), or using a Nikon Microscope (Nikon E800) coupled to a RoHS camera (RoHS). Subsequently, images were analyzed using ImageScope (Leica) and Qcapture Pro software (Qimaging), respectively.

Immunofluorescent images were acquired using an inverted confocal microscope (LSM-710, Carl Zeiss Microimaging) or Leica DM IRBE (Leica). Detection ranges were set to eliminate cross-talk between fluorophores: 409–485 nm for AMCA, 494–553 nm for Alexa Fluo 488 and 561–595 nm for Cy3/Texas Red. Images were analyzed using in ImageJ software package (NIH; ref. 23). For full information regarding the colocalization studies, please see Supplementary Data.

All data are given as mean ± SEM. Statistical analysis was performed using GraphPad Prism (GraphPad Software, Inc.). Unpaired or paired t tests were used to compare data between ipsilateral and contralateral cortices. Unpaired t tests were used to identify differences between DMSO and WP1066 groups. The relationships between ASL-CBF maps and areas of reactive astrocytes were tested by a bivariate correlation analysis, including calculation of the Pearson correlation coefficient (r).

One week after intracortical injection of ENU1564 cells, 94% of all brain metastases were associated with blood vessels (Fig. 1A–C; P < 0.001 compared with nonassociated), indicating that focally induced metastases were predominantly perivascular. Of the blood vessels associated with ENU1564 cells, the majority (74%) were capillaries or small veins (<8.5 μm diameter (Fig. 1D; ref. 25), rather than arterioles, arteries, and large veins (>8.5 μm; ref. 25; P < 0.001 small vs. large).

LSCI measurement of CBF responses to electrical stimulation of the whisker pad in vivo, revealed a focal region of hyperaemia in the contralateral (normal) cortex during stimulation, but a considerably reduced response in the ipsilateral (metastasis bearing) cortex of ENU1564-injected animals (Fig. 1E). Quantitatively, the CBF response was significantly reduced in the ipsilateral versus contralateral cortex (∼50%; Fig. 1F; P < 0.05). Simultaneous LFP recordings of neuronal responses showed no significant differences between ipsilateral and contralateral cortices (Fig. 1G). No significant differences between ipsilateral and contralateral cortices were seen in PBS-injected animals for either CBF or LFP responses (Supplementary Fig. S1A–S1C). Immunohistochemically, no differences in the number of NeuN-positive neurons between cortices in either group (excluding metastatic foci) were evident (Supplementary Fig. S1D). Together, these findings suggest that neurovascular coupling is disrupted in the presence of brain metastases.

Within the metastatic foci, astrocyte end-feet appeared to be displaced from blood vessels owing to the perivascular growth pattern of the metastatic cells [mean displacement = 31.3 ± 2.4 μm; Fig. 1H (i)]. At the same time, a widespread area of reactive astrocytes was evident around the metastatic foci in the ipsilateral cortex [Fig. 1H (ii), dotted line, 0.54 ± 0.11 mm²]. Moreover, a significant decrease in colocalization between β-dystroglycan, GFAP and CD31 was evident throughout the area of reactive astrocytes in ENU1564 animals (Fig. 1I; Table 1; P < 0.05), suggesting detachment of astrocyte end-feet from the vascular endothelium unrelated to the direct mechanical effects of the metastatic cells. No decrease in colocalization was observed in PBS-injected animals (Fig. 1I; Table 1).

Despite the apparent dissociation of astrocyte end-feet from the vasculature, BBB integrity remained intact, as determined by post-Gd-DTPA T₁-weighted MRI (Supplementary Fig. S1E) and IHC detection of Claudin-5, a key tight junctional protein in BBB paracellular permeability (Supplementary Fig. S1F and S1G). Moreover, although blood vessels associated with metastatic cells were abnormal in appearance (Supplementary Fig. S1H), no change in blood vessel number was evident in either metastatic foci or areas of reactive astrocytes (Supplementary Fig. S1I).

To assess the effect of reactive astrocytes per se on neurovascular function, we used a model of CNTF-induced reactive astrocytes (Fig. 2A; ref. 17). Six weeks after intracortical injection, an area of reactive astrocytes was observed in CNTF-Lv animals (Fig. 2B; 0.46 ± 0.09 mm²), while negligible astrocyte reactivity was observed in control LacZ-Lv animals (Fig. 2B). Reactive astrocytes induced by CNTF-Lv did not alter BBB integrity, as determined by post-Gd-DTPA T₁-weighted MRI (Fig. 2C) and Claudin-5/CD31 (vessel marker) colocalization (Supplementary Fig. S2A), neuronal number (Supplementary Fig. S2B), or blood vessel numbers (Supplementary Fig. S2C).

LSCI measurements in CNTF-Lv–injected animals indicated a significant reduction in the amplitude of the CBF response to whisker barrel stimulation in the ipsilateral cortex (∼70%) compared with the contralateral cortex (Fig. 2D and E; P < 0.01). No significant difference was observed in control LacZ-Lv–injected animals (Supplementary Fig. S2D and S2E). As for the metastasis model, LFP responses to electrical stimulation were unaltered in both groups of animals (Fig. 2F; Supplementary Fig. S2F), supporting the concept that neurovascular coupling is disrupted when astrocytes become reactive.

Again, a significant decrease in colocalization of β-dystroglycan, GFAP and CD31 was observed (Fig. 2G, Table 1; P < 0.05), which was not seen in LacZ-Lv animals (Fig. 2G; Table 1). When assessing colocalization of β-dystroglycan and GFAP alone, no change was found in either ENU1546 or CNTF-Lv models (Table 1), indicating that reactive astrocytes exhibit reduced adhesion to the vascular endothelium rather than reduced expression of β-dystroglycan per se.

Animals injected with ENU1564 cells and treated with WP1066 (Fig. 3A), a selective inhibitor of the JAK/STAT3 pathway mediating astrocyte reactivity (15), showed a significant decrease in the area of astrocyte reactivity compared with vehicle (DMSO)-treated animals for both GFAP (∼55%; Fig. 3B and C; P < 0.05) and STAT3 staining (∼77%; Fig. 3B and C; P < 0.05). However, no significant reduction was evident in the area of metastatic foci (0.63 ± 0.12 vs. 0.55 ± 0.27 mm²; Fig. 3C). In WP1066-treated animals, nuclear STAT3 expression in reactive astrocytes was decreased compared with DMSO-treated animals (Fig. 3B).

As expected, (cf. Fig. 1F), DMSO-treated ENU1564 animals, showed a significantly reduced (∼48%) CBF response to electrical stimulation in the metastasis-bearing cortex compared with the contralateral cortex (Fig. 3D; P < 0.05). In contrast, WP1066-treated ENU1564 animals showed a much smaller and nonsignificant reduction (∼26%) in CBF response compared with the contralateral hemisphere (Fig. 3E). As above, LFP responses to whisker pad stimulation were unaltered in both groups (Fig. 3F and G).

Subsequently, we demonstrated that cerebrovascular responses to hypercapnia (26) were significantly reduced in both ENU1564 (∼44%) and CNTF-Lv (∼40%) models, but not in their respective control groups (PBS and LacZ-Lv; Supplementary Fig. S3; P < 0.001 and P < 0.05, respectively). Again, STAT3 inhibition improved the hypercapnia-evoked vascular responses in ENU1564-injected animals, with no significant difference evident between cortices (Supplementary Fig. S3A–S3L).

Together these findings indicate that inhibition of STAT3-mediated astrocyte reactivity can, at least partially, reverse cerebrovascular dysfunction associated with brain metastases.

Measurements of basal CBF using ASL-MRI showed a significant reduction compared with the contralateral cortex of animals injected with either ENU1564 cells or CNTF-Lv (Fig. 4A–D; P < 0.01). Notably, the area of reactive astrocytes as a percentage of the somatosensory cortex was much larger than the metastatic foci (40% vs. 3%, respectively), and correlated spatially with the region of reduced CBF for both ENU1564 and CNTF-Lv models (Fig. 4E; P < 0.05; Pearson coefficient correlation r² = 0.33). In contrast, no changes in basal CBF were observed in PBS or LacZ-Lv control groups (Fig. 4F and G).

STAT3 inhibition with WP1066 largely reversed the reduction in basal CBF observed in ENU1564 animals (Fig. 4H and I), with a significant difference compared with DMSO-treated animals (P < 0.05). Given that the area of reactive astrocytes showed a much greater reduction after WP1066 treatment (∼55%) than the metastatic foci (∼12%), we suggest that reactive astrocytes, rather than the presence of metastases per se, underlie the reduced basal CBF in metastasis-bearing animals.

Large vessels (>8.5 μm) showed significant vasoconstriction in the area of reactive astrocytes within the ipsilateral versus contralateral cortex, in both ENU1564 and CNTF-Lv models (Fig. 5A and B; P < 0.05). Moreover, colocalization between cytochrome p450–4A, an endothelium-dependent vasoconstrictor (27, 28), and αSMA, a marker of smooth muscle cells on large blood vessels (29), was significantly greater in ipsilateral versus contralateral cortex for both models (Fig. 5C and D; Supplementary Table SII; P < 0.05). No differences were evident in control animals (Fig. 5C and D; Supplementary Table SII). No changes in cytochrome p450–4A levels were observed in either astrocytes or microglia (Fig. 5E and F).

In contrast, a significant increase in the diameters of smaller vessels (<8.5 μm diameter) in the region of astrocyte reactivity was evident in both ENU1564 and CNTF-Lv animals (Fig. 5G and H), compared with controls (PBS and LacZ-Lv; Fig. 5G and H; P < 0.001). Colocalization of the vasodilatory molecule iNOS with astrocytes was significantly greater in ENU1564 and CNTF-Lv animals compared with controls (Fig. 5I; P < 0.05 and P < 0.01; respectively), together with increased cortical levels of nitrotyrosine, a marker of nitric oxide (Fig. 5J; ref. 30). Similarly, astrocytic levels of COX-1, a key enzyme in the arachidonic acid vasodilatory pathway (1), were significantly greater in ipsilateral versus contralateral cortex of ENU1564 and CNTF-Lv groups (Fig. 5K; P < 0.05 and P < 0.01, respectively), but not in controls (PBS and LacZ-Lv; Fig. 5K). COX-1 showed greater colocalization with astrocytes than any other cells (Supplementary Table SIII), suggesting they are the dominant cell population driving COX-1–mediated vasodilation, while both astrocytes and endothelial cells showed the highest levels of iNOS colocalization. A significantly higher number of OX42-positive pixels was found in the ENU1564 group compared with PBS, CNTF-Lv or LacZ-Lv–injected groups (Supplementary Table SIV), confirming an inflammatory microglia/macrophages response in presence of brain metastases (31, 32), but no contribution of microglia/macrophages in the CNTF-Lv model (9, 17).

No changes were observed in the levels of COX-2 in astrocytes, iNOS in endothelial cells, or iNOS/COX-1 in microglia in any of the groups (Supplementary Fig. S4A–S4D), suggesting that vasodilation of small vessels reflects activation of the nitric oxide and arachidonic acid pathways primarily in astrocytes. A significant decrease in both iNOS/GFAP and COX1/GFAP colocalization was evident in WP1066- versus DMSO-treated ENU1564 animals (Fig. 5L). However, neither iNOS nor COX-1 was altered in microglia, endothelial cells or smooth muscle cells following WP1066 treatment (Supplementary Fig. S5A–S5D). In addition, expression of iNOS and COX-1 was not altered in microglia (OX42) in the ipsilateral cortices of ENU1564 animals treated with WP1066 (Supplementary Fig. S6A–S6C). These results indicate that STAT3 inhibition is primarily limited to reactive astrocytes within the metastatic microenvironment.

Reactive astrocytes were evident in human brain metastasis biopsies (Oxford Brain Bank: OBB-SH-1088-2013) primarily in the areas surrounding tumor colonies (Fig. 6A). As in the animal models, colocalization between cytochrome p450–4A and αSMA staining was seen on blood vessels associated with brain metastases (Fig. 6B), and both iNOS and COX-1 staining were present in astrocytes (Fig. 6C and D). The percentage of αSMA- and cytochrome p450–4A–positive blood vessels was greater around metastatic foci compared with control brain (Supplementary Table SII). Significantly greater levels of both iNOS and COX-1 per GFAP-positive pixel, and cytochrome p450–4A per αSMA-positive pixel, were found in the area of reactive astrocytes around metastatic foci compared with nontumor tissues (Fig. 6E; P < 0.001 and P < 0.05, respectively).

In a rat model of brain metastasis, we have demonstrated impairment of cerebrovascular function, disruption of astrocyte-vascular connections and widespread reactive astrocytes. Studies using a model of CNTF-induced astrocyte reactivity, indicated that reactive astrocytes per se can compromise cerebrovascular function, while inhibition of STAT3-mediated astrocyte reactivity partially reversed the cerebrovascular dysfunction associated with brain metastases. These findings suggest that STAT3-mediated astrocyte reactivity in brain metastasis may contribute to cerebrovascular, and hence neurological, dysfunction in patients.

In glioma, tumor cells displace astrocyte end-feet from blood vessels enabling tumor growth in the perivascular space (12, 33). Similarly, we and others have shown that brain metastases grow in this co-optive fashion, particularly during the micrometastatic stages (10, 11, 34). Here, we show that not only are astrocytes displaced from the vasculature mechanically, but that reactive astrocytes distant from the metastatic foci dissociate from the vasculature. Watkins and colleagues (12) hypothesized that displacement of astrocytes from the vasculature by invading glioma cells prevents vasoactive molecules from reaching endothelial cells, thus disrupting neurovascular coupling (12). While displacement-mediated vascular disruption may also occur in brain metastasis, our findings suggest that widespread reactive astrocytes, outside the spatially restricted metastatic foci, contribute significantly to disruption of vascular function. Critically, a similar compromise of cerebrovascular function was observed in the CNTF-Lv model, in which astrocytes were activated to a reactive phenotype for a prolonged period, while neuronal activity was preserved (17). Similarly, CBF responses to hypercapnia were compromised in both models. To date, only two studies have suggested that vasoactive pathways in astrocytes are involved in CBF responses to hypercapnia (35, 36). Our data support this concept and suggest that astrocyte reactivity may compromise autoregulation.

One possible explanation for the reduced CBF responses to both neuronal activation and hypercapnia is that vasodilation across the vascular bed leads to reduced vascular reserve (37, 38). However, although we found vasodilation in the microvascular network, the arteriolar vessels appeared to be constricted and basal blood flow was reduced, in keeping with restricted arteriolar flow. Moreover, we found more cytochrome p450–4A, the enzyme responsible for the biosynthesis of vasoconstrictor 20-HETE (27, 28), on smooth muscle cells at the arterioles. Consequently, we speculate that suppression of CBF responses to stimuli in the presence of reactive astrocytes may reflect either basal upregulation of vasoconstrictory pathways in the arteriolar bed and/or upregulation of enzymes that mediate vasodilation (iNOS and COX-1) downstream of the arterioles and reduced vascular reserve within the microvascular bed. A recent study has suggested that astrocytes regulate CBF at the capillary and arteriole levels through two distinct mechanisms (39), reinforcing the concept that reactive astrocytes exert differential effects on arterioles and capillaries.

Importantly, upregulation of the above vasoactive mediators was mirrored in all human brain metastasis samples studied, typically at the leading edge of the tumor. Recent studies have identified the JAK/STAT3 pathway in reactive astrocytes as a novel target for neurological diseases (40), brain metastasis (15) and glioblastoma (41). Together with significant antitumor effects (15), data suggest that inhibition of STAT3 may also reduce the immunosuppressive environment of brain tumors induced by cross-talk between astrocytes and microglia (15, 42), implicating this inflammatory microenvironment as a potential target for treatment in brain metastasis (43). Given our data showing that STAT3 inhibition restores impaired cerebrovascular function in brain metastasis, we propose that therapy suppressing astrocyte reactivity may also be effective in reducing metastasis-associated neurocognitive effects and in improving the response of current therapies that are limited by tumor-associated hypoxia and abnormal vasculature (44, 45). Notably, reduced blood flow has been linked to the neurological symptoms associated with brain metastases, such as seizures (46–48) and stroke (49), and this could be investigated in ongoing brain tumor clinical trials with STAT3 inhibitors (e.g., NCT01904123).

With regards to limitations of this study, although we have shown that reactive astrocytes are the dominant cell type producing prostaglandin mediated vasodilation via COX-1 activity, NO-mediated vasodilation is not solely driven by reactive astrocytes. Thus, separating the contributions of different stromal cell populations to the production of vasoactive mediators requires further investigation, potentially using genetically encoded calcium indicators, transgenic techniques, or viral constructs, in combination with in vivo recording of hemodynamic response. In addition, only one brain metastasis model was used in this study. Given that ENU model is well established (19, 50–52), rather than using a second brain metastasis model we considered it to be more relevant to use a model of CNTF-induced reactive astrocytes to determine the effect of reactive astrocytes specifically, in the absence of metastases, on neurovascular function. At the same time, the IHC data obtained from human brain metastasis samples lend further support to our experimental findings. Finally, we did not fully suppress astrocyte reactivity in our models via STAT3 inhibition, suggesting that astrocyte activation may be complex and heterogeneous (53). For example, the NFκB signaling pathway has been shown to contribute to the neurotoxicity of a subpopulation of reactive astrocytes induced by activated neuroinflammatory microglia (54) and, thus, may be an important additional pathway of astrocyte activation in brain metastasis that warrants further investigation.

In summary, this study is the first empirical demonstration that STAT3-mediated astrocyte reactivity associated with brain metastasis contributes to acute cerebrovascular dysfunction. STAT3 inhibition may, therefore, limit detrimental neurologic symptoms associated with disease progression and improve outcome for patients with brain metastases.

J.R. Larkin reports a patent for Algorithm for accurate quantification of perfusion using multi-phase arterial spin labeling MRI pending to University of Oxford. N.R. Sibson reports grants from Cancer Research UK during the conduct of the study. No disclosures were reported by the other authors.

M. Sarmiento Soto: Conceptualization, resources, formal analysis, investigation, methodology, writing-original draft, writing-review and editing. J.R. Larkin: Conceptualization, software, formal analysis, investigation. C. Martin: Data curation, software, investigation. A.A. Khrapitchev: Data curation, software, investigation. M. Maczka: Software, formal analysis, methodology. V. Economopoulos: Methodology. H. Scott: Data curation. C. Escartin: Conceptualization, methodology. G. Bonvento: Conceptualization, methodology. S. Serres: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, investigation, methodology, writing-original draft, writing-review and editing. N.R. Sibson: Conceptualization, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

The authors thank James Meakin for assistance with implementation of the pCASL sequence and G. Stoica (Texas A&M University) for the ENU1564 cells.

This work was supported by Cancer Research UK (grant number C5255/A15935 to N.R. Sibson), a FRB's 4-star research booster scheme from the University of Nottingham and Brain Tumour Charity (GN-000537 to S. Serres) and a Marie-Sklodowska Curie Action-IF (H2020-795695 to M. Sarmiento Soto).

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.