We retrospectively analyzed 18 NSCLC and 18 melanoma patients at Yale Cancer Center enrolled in a phase 2 trial of pembrolizumab (10 mg/kg, IV every 2 weeks) for untreated BMs, NCT02085070 [4, 12]. Patients were ineligible if they had symptomatic edema requiring corticosteroids or intracranial metastasis > 2 cm not previously irradiated. BM response was evaluated by MRI using modified RECIST 1.1 every 8 weeks.

Enrollment MRIs were analyzed using 3D Slicer (https://www.slicer.org) [13–15]. Images were segmented using 3D T1-weighted MP-RAGE or FLAIR sequences. We applied the Fast GrowCut Extension with Laplacian 0 settings to generate 3D models with corresponding tumor (V_T) and edema (V_E) volumes. To normalize edema to tumor volumes, (edema:tumor) ratios were calculated as (V_E)/(V_T). Up to five of the largest tumors per patient were evaluated. Tumors were excluded if nearby lesions caused overlapping edema or if images could not be processed by 3D Slicer.

A previously constructed and characterized tissue microarray containing formalin-fixed, paraffin-embedded tumor tissue from 30 melanoma BM patients was stained with anti-CD34 (Dako) to determine micro-vessel density, which was quantitated as detailed previously (Additional file 1: Supplemental Methods) [16, 17].

Short-term melanoma cultures were obtained from resected MBMs, used within 20 passages, and denoted by the prefix “YU” for Yale University (Additional file 2: Table S1). A375P and its cerebrotropic daughter cell line A375Br were a gift from Dr. Huang [18]. Co-cultures of primary human umbilical vein endothelial cells (HUVEC, ScienCell) and primary human astrocytes (ScienCell) recapitulating the BBB were established, as previously published and detailed in Additional file 1: Supplemental Methods [19]. Establishment of the BBB was confirmed via expression of brain endothelial markers glucose transporter-1 (GLUT1) and γ-glutamyl transpeptidase (GGT1) (Additional file 3: Figure S1) [20]. BBB leakiness was assessed by changes in transendothelial electrical resistance (TEER) and confirmed via lack of Evans Blue-labeled albumin (0.45% in phenol red-free medium) passage after 30 min at 37 °C using a previously established protocol (Additional file 4: Figure S2) [19, 21].

Correlations were analyzed by the coefficient of determination. Clinical variables were compared by unpaired t-tests. Progression free survival (PFS) and overall survival (OS) analyses were depicted by the Kaplan–Meier method and stratified by log-rank test. Comparison of in vitro TEER to in vivo edema was done by Fisher’s exact test.

Based on clinical observations that BMs of similar size can induce variable edema volumes (examples shown in Fig. 1a), we used 3D modeling to quantitate tumor and edema volumes. The correlation between tumor and edema volumes prior to anti-PD-1 was weak (R ² = 0.30, p < 0.0001, Fig. 1b). While melanoma BMs tended to be more edematous, this observation did not reach statistical significance (P = 0.059, Additional file 5: Figure S3A). Given the weak relationship between size and edema, we assessed edema:tumor volume ratios, and these were similar in NSCLC and melanoma both when comparing all tracked BMs (P = 0.26, Fig. 1c) or the largest lesion per patient (P = 0.67, Additional file 5: Figure S3B). Edema:tumor volumes by NSCLC subtype (adenocarcinoma, squamous cell, or poorly differentiated carcinoma) (P > 0.05, data not shown) or by lactate dehydrogenase level in melanoma patients (P = 0.83, data not shown) were also similar. Patient characteristics were previously described [4].Fig1Fig. 1

Tumor volume weakly correlates with edema among NSCLC and melanoma patients. a Representative baseline MRIs taken from patients treated with pembrolizumab showing similar tumor volumes on T1 post gadolinium 3D MP-RAGE (white arrows) but discordant amounts of edema volume on FLAIR sequences. b The correlation between tumor and edema volume of NSCLC and melanoma BMs is weak. c Pre-treatment edema:tumor volumes are not significantly different in melanoma and NSCLC patients

As there was no difference in edema volumes between melanoma and NSCLC patients when correcting for tumor size, we combined the cohorts for subsequent analyses. Historically, there has been hesitancy to treat BMs with CPIs due to concern for edema exacerbation, but we found that pre-treatment edema:tumor volume did not impact the likelihood of response to pembrolizumab in the brain (P = 0.82, N = 14 evaluable NSCLC and 13 melanoma patients, Fig. 2a).Fig2Fig. 2

Pembrolizumab response and survival were independent of initial edema:tumor volumes and concordant with edema change. a Pre-treatment BM edema:tumor volume ratio does not impact response to pembrolizumab (PD-progression of disease, SD-stable disease, PR-partial response, and CR-complete response, by modified RECIST version 1.1). b Tumor and edema volume changes from baseline after 4 cycles of pembrolizumab were strongly correlated. As pembrolizumab-sensitive BMs shrank, the tumor-associated edema also shrank and vice-versa for pembrolizumab-resistant tumors. Due to the wide range in values, the log percent change was used for graphical presentation. c Kaplan-Meier survival curves show no significant difference between BM PFS in patients whose edema:tumor volume was above the median compared to those whose ratio was below the median; d overall survival of NSCLC and melanoma patients was similarly not affected by baseline edema:tumor volume ratios

Changes in edema and tumor volume after 4 cycles of pembrolizumab compared with baseline were strongly correlated (R ² = 0.81, N = 11 NSCLC and 8 melanoma patients with 27 and 17 target lesions, respectively, as 3 NSCLC and 5 melanoma patients were excluded due to disease progression or immune-related side effects prior to receiving 4 cycles of treatment, Fig. 2b). Patients who responded to therapy had corresponding decreases in tumor and edema volume and vice versa.

To determine whether more edematous lesions were associated with worse PFS or OS, we binarized edema:tumor volume ratios by the median. There was no statistically significant difference between low and high edema:tumor ratios and PFS (P = 0.23, Fig. 2c). There was a trend towards an association between improved OS and lower edema:tumor ratios, but this was not significant (P = 0.17, Fig. 2d).

One potential mechanism of edema is increased tumor neo-angiogenesis resulting in formation of functionally aberrant, leaky neo-vessels. Tumor vessel density was determined in 23 melanoma craniotomy samples with available pre-operative MRIs; similar samples were not available for NSCLC patients. There was no correlation between vessel density and either tumor volume (R ² = 0.087), edema volume (R ² = 0.12), or edema:tumor ratio on pre-resection MRI (R ² = 0.037, Fig. 3a-d).Fig3Fig. 3

Edema is not correlated with micro-vessel density. a Representative images of melanoma BM core sections with either high or low anti-CD34⁺ staining. There were no correlations between density of CD34⁺ staining cells in melanoma BMs and pre-resection b tumor volume, c edema volume, or d edema:tumor ratio

To determine the effects of tumor cells themselves on edema, we first established TEER as a surrogate for BBB leakiness by demonstrating that passage of albumin strongly correlated with decreased TEER (R = -0.69, P = 0.0045, Additional file 4: Figure S2). As vasogenic edema is uniquely associated with cerebral metastases, we determined whether cells cultured from extra-cerebral metastases resulted in decreased TEER when co-cultured with HUVEC cells and astrocytes in our BBB model. Co-cultures with A375P cells (derived from a lymph node) resulted in increased TEER (P < 0.0001 compared to CTRL containing only HUVECs and astrocytes), whereas its cerebrotropic daughter cell line A375Br resulted in decreased TEER (P = 0.0005) [18]. Further testing with other extracerebral cultures YUCOT, YUSIT, and YUVON also did not result in decreased TEER (P > 0.05, Fig. 4a). To assess how intracerebral metastases affect the BBB, 13 short-term cultures of human melanoma BMs were assayed using the in vitro BBB model. Similar NSCLC cultures were not available. TEER decreases were observed after the addition of 9 of the 13 cultures (Fig. 4b), compared to CTRL or to 293 T human embryonic kidney and 3T3/J2 murine fibroblast cell lines. These results are consistent with the clinical picture in which not all cerebral metastases are associated with edema, and perilesional edema is not typically observed in other organs.Fig4Fig. 4

Specific human melanoma BM cells induce BBB compromise without contributions of immune cells in vitro. a TEER was measured as a surrogate for inter-endothelial tight junction integrity in an in vitro BBB model. A significant decrease in TEER was found in a cerebrotropic human melanoma daughter cell line (A375Br), derived by repeated intra-carotid injection of A375P parental cells and expansion of cells isolated from the brain. A375P cells derived from a metastatic lymph node increased TEER. Short-term melanoma cultures from other extracerebral sites (YUCOT, YUSIT, and YUVON) similarly did not result in decreased resistance compared to CTRL (HUVEC and astrocytes co-cultured without the addition of melanoma cells). b A panel of short-term cultures derived from human melanoma BMs was similarly studied and TEER changes determined. Co-culture with some but not all melanoma BMs resulted in decreased resistance compared to CTRL. c TEER changes were compared to MRI-based tumor and edema volume measurements of resected lesions from which the cultures were derived. Black circles represent cell lines with decreased TEER, whereas white circles correspond to cell lines with increased TEER changes. d Four of the nine cell cultures that decreased TEER (black circles) are from more edematous tumors, whereas all the cell cultures that did not decrease TEER (white circles) were from less edematous tumors