Intracranial melanoma metastases display more vascular mimicry than extracranial metastases

We previously showed that aggressive brain metastatic competent cells cultured in matrigel can readily induce vasculogenic mimicry (VM) in vitro, and VM competency seems to correlate with the ability of these cells to induce metastatic outgrowth [32]. Moreover, our previous studies also show that human resected melanoma brain metastases have lower CD34 + vessel density compared to matched extracranial metastases, but non-angiogenic vasculature has yet to be studied in specimens from patients with melanoma brain metastases [48]. We therefore sought to evaluate VM in matched cerebral and extra-cerebral metastases from a cohort of 37 melanoma cases previously described [48]. A specific feature of VM channels is a wall structure negative for CD34 staining but positive for PAS staining [32, 56, 57]. CD34 + /PAS + staining identifies endothelium-lined blood vessels (BV) whereas CD34-/PAS + staining identifies VM (Fig. 1a). Using methods we previously described [32], we conducted thorough microscopic examination of sections for CD34 and PAS staining, followed by quantification of BV and VM structures. We found a significantly higher number of VM structures in intracranial metastases compared to their matched extracranial counterparts (Fig. 1b). As expected, BV levels were higher in extracranial cases compared to matched intracranial cases (Fig. 1c). The same trends were found when our comparative analysis included all matched and unmatched tumor specimens on the TMA (Intracranial: 75 samples, Extracranial: 117 samples; Supplementary Fig. 1a, b). When intra-tumoral BV and VM structures were compared, we found that intracranial cases have significantly more VM than BV whereas there was no difference between the number of VM and BV in extracranial cases, and this was true whether analysis included only matched lesions (37 patients, Fig. 1d and e, respectively) or the entire cohort (Supplementary Fig. 1c and d, respectively). Collectively, these data suggest that intracranial melanoma metastases, but not extracranial melanoma metastases, may utilize VM over angiogenesis as a form of neovascularization in humans.

Fig. 1

Intracranial melanoma metastases have higher levels of vascular mimicry than extracranial metastases: a Representative images of immunohistochemical staining for CD34 and PAS in extracranial and intracranial tumors isolated from patients at Yale University. Black arrow: CD34 + /PAS + , red arrow: CD34-/PAS + . Magnification of images are 20x with scale bars representing 50 µM. b Analysis of CD34-/PAS + vessels and CD43 + /PAS + vessels c in matched intracranial and extracranial metastasis patient samples. Black is intracranial, blue is extracranial, and the grey line connects the matched pairs. Analysis of CD34 + /PAS + and CD34-/PAS + vessels in intracranial metastases (d) and in extracranial metastases (e). The grey line connects values from the same tumor. Significance for all panels assessed by paired-Student’s t-test. ****p < 0.0001, ***p < 0.001, **p < 0.01

Intracranial vascular mimicry is linked with MBM tumor volume and edema

Obtaining nutrients from the tumor microenvironment is important for tumor growth and viability, which is predominantly acquired via angiogenesis [12]. However, our group showed that there is no correlation between BV density and tumor volume in intracranial melanoma metastases [58]. Therefore, we sought to study the association between VM and brain tumor volume. When tumor volume was dichotomized by the median value into high and low groups, high tumor volume independently associated with high levels of VM in intracranial cases (Fig. 2a). However, as expected, there was no significant association between tumor volume and BV number (Fig. 2b). These results were confirmed by Chi-square analysis of dichotomized VM (Supplementary Fig. 2a) or BV (Supplementary Fig. 2b) with continuous tumor volume scores. These results suggest that VM could facilitate brain metastasis tumor growth.

Fig. 2

Vascular mimicry in melanoma brain metastases is associated with melanoma brain metastasis tumor volume and edema. a Two-sample unpaired t-test showing that high VM staining (continuous tumor CD34-/PAS+ scores) is significantly associated with high brain metastasis tumor volume (tumor volume scores dichotomized by the median value) (p = 0.0124). b Two-sample unpaired t-test showing that high blood vessel staining (continuous tumor CD3 + /PAS + scores) is not significantly associated with high brain metastasis tumor volume (tumor volume scores dichotomized by the median value) (p = 0.2402). c Continuous scores for VM staining (CD34-/PAS+) were plotted against intracranial edema scores dichotomized by the median value and significance assessed via two-sample unpaired T-test (p = 0.0415). d Two sample t-test of continuous BV scores (CD34+ /PAS+) and dichotomized edema scores (P = 0.1996)

Using an alternative quantification method, we previously showed that brain tumor volume positively correlates with edema, derived by 3D modeling from patient MRIs [32] and this was once again confirmed in the current study by linear correlation (Supplementary Fig. 2c). Therefore, we further investigated the relationship between vasculature and edema, and found a statistically significant correlation between VM and increased edema (Fig. 2c). The same trend was seen for angiogenic vasculature, although the association between BV and high edema did not reach statistical significance (Fig. 2d). This suggests that VM could facilitate perilesional edema in patients with melanoma brain metastases.

Brain metastatic cell lines form vascular structures in vitro

To further study VM formation in advanced melanoma, we assessed the ability of several short-term human melanoma cell line cultures to form VM in vitro. Our assay includes cells derived from melanoma brain metastases (blue) and cells derived from extracranial metastases of patients who did not develop brain metastases (non-cerebrotropic, black) or from those who did (cerebrotropic, red). We found a tendency for cells established from melanoma brain metastases (YUVENA, YUGANK, YUKRIN, YUMETRO) to have a significantly better ability to induce VM in vitro compared to cells from other metastatic sites (YUKOLI, YUMUT, YUKSI, YUCOT, YUGASP) as determined by quantification of the number of meshes, junctions, and tube length (Fig. 3a–d). Similarly, YUSIK and YUSIV cells derived from extracranial specimens of patients who developed rapid and overt brain metastases, also exhibited high VM in vitro (Fig. 3a–d). Moreover, a pronounced ability to induce VM was seen in the brain metastatic mouse melanoma line derived from repeat left-ventricle injections and isolation of clones from the brain, YUMM1.1Br, compared to its parental counterpart, YUMM1.1, derived from a genetically engineered mouse model [50] (Fig. 3e–h). Similarly, the murine B16.F10 and human Cl.2A cell line derivatives which generate melanoma brain metastases [32, 59] exhibited VM as opposed to parental, non-brain metastatic B16 and Cl.1A cells, respectively (Fig. 3e–h). We also observed an increased amount of VM with mouse tumor endothelial cells as compared to normal endothelial cells (Supplementary Fig. 3). We did not see a significant difference in VM formation between YUMM1.7 and the mutagenized derivative, YUMMER1.7 (Supplementary Fig. 3). Lastly, in our previous study we identified paracrine stimulation of VM with Cl.2A cells [32]. Here, we found that conditioned medium from Cl.2A, YUSIK, and YUVENA cells enhance VM formation with the non-cerebrotropic cell lines, YUKOLI, YUKSI and YUCOT (Supplementary Fig. 4a–i). Additionally, we observed stimulation of VM formation by brain metastatic cells in the presence of endothelial cells in a co-culture model (Supplementary Fig. 5). Taken together, these studies indicate that VM is likely a process exploited to a greater extent in melanoma brain metastases.

Fig. 3

Melanoma cell lines form vascular networks in vitro. Quantification of the number of meshes (a), junctions (b) and tube length (c), in a melanoma cell line panel and the representative images (d). Blue represents short-term cultured patient derived cell lines from intracranial melanoma metastases, black represents short-term cultured patient derived cell lines from extracranial metastases, red represents short-term cultures derived from extracerebral metastasis in patients who developed overt brain metastases (cerebrotropic) (d; 10 × magnification). Quantification of the number of meshes (e), junctions (f) and tube length (g), in primary and brain metastatic cell lines. Black represents primary, blue represents cerebrotropic derivatives. B16 variants and YUMM1.1 variants are mouse cell lines. h Representative images of VM structures formed by the cell lines in panels e–g (10 × magnification). All data are expressed as mean \(\pm\) SD of three biological replicates and statistical significance is determined using an unpaired Student’s T-test. ****p < 0.0001

Targeting YAP/TAZ inhibits vascular mimicry in brain metastatic cell line derivatives

Recently, YAP/TAZ has emerged as a key mediator of VM [21, 29] and our previous work shows that altered YAP/TAZ signaling is associated with VM and tumor growth [32]. Here, we found YAP and TAZ to be expressed at variable levels in melanoma (Supplementary Fig. 3i, j) and that siRNA targeted inhibition of YAP, TAZ, and YAP/TAZ decrease VM formation in Cl.2A and YUSIK cells (Supplementary Fig. 6).

To further investigate YAP in human cerebral and non-cerebral melanoma tumors we employed quantitative immunofluorescence and a TMA of clinical samples from a historical cohort of metastatic melanoma cases with variable times to development of brain metastasis [49]. YAP activity was evaluated in a subset of these tumors (51 cases) for which transcript profiles were available from a previously published dataset [49]. The fluorescent signal was quantified within the S100-positive area or within the tumor DAPI positive nuclei. By two-way comparison, YAP levels in cerebral metastases were significantly higher when compared to extra-cerebral tumors suggesting that YAP expression might be associated with brain involvement of melanoma (Supplementary Fig. 7a, b). Consistent with these findings, examination of YAP distribution across various metastatic sites showed a prevalence of higher YAP in visceral metastases and low expression in skin lesions, further suggesting that YAP upregulation and activity could be associated with the location of the specific metastasis (Supplementary Fig. 7c, d). Furthermore, in this subset of cases, high YAP expression correlated with high VM, though data only trended toward significance (Supplementary Fig. 7e, f). Using a gene expression data set from our previously published tumor profiling studies, we found several known YAP/TAZ targets previously linked to angiogenesis to be differentially expressed in these metastatic tumors. For example, expression of DOCK5, RUNX1, ASAP1, JUN, FOS, positively correlated with YAP levels and for these comparisons p-values were for the most part significant or otherwise a trend was evident (Supplementary Fig. 7g). Interestingly, high YAP expression also correlated with downstream transcriptional targets previously implicated in VM such as SNAI2, an EMT regulator, and STAT3 and ANG2, two known potent proangiogenic factors (Supplementary Fig. 6h) [21, 60]. Additionally, CDC20 expression correlated with VM and a similar trend was seen for CCND1 and ANG2 genes (Supplementary Fig. 7h).

To expand on siRNA targeted inhibition of YAP and TAZ we tested the effects of pharmacologically targeting YAP/TAZ with verteporfin (VP) and CA3, two known YAP/TAZ inhibitors, on VM formation in vitro, using the aggressive Cl.2A human melanoma cells and the two highly cerebrotropic derivatives of mouse melanoma cell lines, YUMM1.1Br and B16.F10 [61,62,63,64,65]. In Cl.2A, YUMM1.1Br, and B16.F10 cells we observe a dose dependent decrease of VM when treating with CA3 and VP when quantifying number of meshes (Fig. 4a–c) alongside number of junctions and tube length (Supplementary Fig. 8a–f). CA3 and VP inhibit VM in Cl.2A and YUMM1.1Br at 1 µM and 4 µM, respectively (Fig. 4a, b), and VP inhibits B16.F10 at 0.5 µM whereas the active dose for CA3 is 1 µM (Fig. 4c). Evaluation of live cell density using Cell Titer Glo shows no significant difference in cell viability, which is observed in all cell lines tested at the active dose of CA3 or VP at 6 hrs, the time of VM formation, an indication that drug effects on VM are not an indirect result of cell death (Supplementary Fig. 9a–d). At this timepoint, we observed a decrease in both YAP and TAZ protein levels upon treatment with VP, but not CA3 in Cl.2A, YUMM1.1Br, B16.F10, (Supplementary Fig. 10a–c). We also observed a variable decrease in select YAP/TAZ target genes after VP and CA3 treatments (Supplementary Fig. 11a–c). Using the lowest dose for VM inhibition for each tested cell line, we next assessed the effects of VP and CA3 on cell proliferation. In Cl.2A cells, there is significant growth inhibition at 1 µM CA3 and VP at 72 hrs (Fig. 4d). In YUMM1.1Br cells, 4 µM CA3 had no effect on proliferation after 72 hrs, while VP significantly inhibited cell proliferation (Fig. 4e). In B16.F10 cells, there was no significant difference in cell proliferation throughout the time course for the active dose of CA3 (1 µM) or VP (0.5 µM) (Fig. 4f). Lastly, we confirmed the importance of YAP/TAZ signaling in mediating VM through testing the effects of a third compound, TED-347. TED-347 is a covalent, allosteric inhibitor of the YAP1-TEAD4 protein–protein interaction which suppresses TEAD4 transcriptional activity [66]. TED-347 robustly inhibited VM in a dose dependent manner in Cl.2A cells (Supplementary Fig. 12a–d).

Fig. 4

The effect of YAP/TAZ inhibitors on VM and growth in cerebrotropic derivatives of melanoma cell lines. CL.2A (a), YUMM1.1Br (b), and B16-F10 (c) cells were treated with increasing doses of CA3 and VP for 6 hrs at which point VM formation was assessed by mesh quantification. The number of meshes were counted for each treatment group. Black = DMSO, Blue = CA3, Green = VP. d CL.2A cells were treated with the active dose of CA3 (top, 1 µM) and VP (bottom, 1 µM), (e) YUMM1.1Br were treated with the active dose of CA3 (top, 4 µM) and VP (bottom, 4 µM), (f) and B16.F10 cells were treated with the active dose of CA3 (top, 1 µM) and VP (bottom, 0.5 µM), for 72 hrs and cell proliferation was assessed by Cell Titer Glo every 24 hrs. The 0 h timepoint represents cells assessed prior to drug treatment. Representative Relative Luminescent Units (RLU) are normalized by subtracting the 0 h values from subsequent measurements. All data are expressed as mean \(\pm\) SD of at least three biological replicates and statistical significance is determined using an unpaired Student’s T-test at each time point. ****p < 0.0001, **p < 0.01, *p < 0.05

To confirm our results in short-term patient-derived cultures, we used YUGANK, YUVENA, and YUKRIN, cells derived directly from patient MBM, and YUSIK, obtained from an extracranial tumor specimen of a patient with brain metastasis. We found that the optimal dose for VM inhibition in these cell lines for both CA3 and VP is 0.5 µM (Fig. 5a, Supplementary Fig. 8 g–j). Notably, with this dose, there was no significant difference in cell viability observed in any cell line tested at 6 hrs, the time of VM formation (Supplementary Fig. 9 a and c), and at this timepoint we observed that CA3 and VP decreased both YAP and TAZ total levels in YUGANK, YUKRIN, and YUSIK cell lines (Supplementary Fig. 10e–g) alongside a variable decrease in select YAP/TAZ target genes (Supplementary Fig. 11d–g). Overtime CA3 inhibited proliferation of YUVENA, YUGANK, YUKRIN and YUSIK cells, but this effect was only significant after at least 48 hrs (Fig. 5b–e). By contrast, VP did not impact proliferation of YUGANK and YUVENA cells and only had a negligible inhibitory effect on proliferation in YUKRIN and YUSIK cells after 72 hrs (Fig. 5b–e). Collectively, these data suggest YAP/TAZ signaling might drive VM formation in melanoma brain metastasis while also mediating oncogenic growth, but this appears to be context dependent as the cell lines tested were affected differently by drug treatment.

Fig. 5

The effect of YAP/TAZ inhibitors on VM and growth in patient derived melanoma brain metastasis cultures. a YUGANK, YUVENA, YUKRIN and YUSIK were treated with 0.5 µM CA3 (blue, left) or 0.5 µM VP (green, right) for 6 h at which point VM formation was assessed by mesh quantification. YUGANK (b), YUVENA (c), YUKRIN (d) and YUSIK (e), were treated with the active dose of CA3 (0.5 µM) and VP (0.5 µM) for 72 hrs and cell proliferation was assessed by Cell Titer Glo every 24 hrs. The 0 h timepoint represents cells assessed prior to drug treatment. Representative Relative Luminescent Units (RLU) are normalized by subtracting the 0 h values from subsequent measurements. All data are expressed as mean \(\pm\) SD of at least three biological replicates and statistical significance is determined using an unpaired Student’s T-test at each time point. ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05

The effect of the YAP/TAZ inhibitor, CA3, on melanoma metastasis

Since YAP/TAZ pharmacologic inhibition attenuated cell growth and VM formation, we next sought to study its effects on metastatic growth employing a left-ventricle injection model of Cl.2A cells, which form both intracranial and extracranial metastases. Nude mice received 1 mg/kg–10 mg/kg CA3 three days after injection, prior to formation of any overt distant metastases. CA3 doses as low as 1 mg/kg have been shown to inhibit growth in non-metastatic models and is well tolerated [62]. All mice developed brain metastasis before death and treatment of CA3 was non-toxic to mice as assessed by weight (Supplementary Fig. 12e). We found that CA3 treatment prolonged overall survival as measured by the time to death from treatment initiation (Fig. 6a). Additionally, the time to death after detection of the first extracranial metastasis was delayed by CA3 treatment, though the difference between groups only trended toward significance (Fig. 6b). Interestingly, CA3 significantly prolonged the survival rate from date of first MBM detection (Fig. 6c). As expected, CA3 treatment decreased the levels of YAP, TAZ, and the target gene, BIRC5 (Fig. 6d). Moreover, CA3 treatment did not affect blood vessel density but decreased VM density in melanoma brain tumor specimens (Fig. 6e–g). We then sought to validate these data using TED-347 (20 mg/kg). TED-347 significantly prolonged mouse survival, time to death from first metastasis, and time to death from MBM (Supplementary Fig. 12f–h) and no drug related toxicity was observed (data not shown). Like CA3 treatment, TED-347 treatment did not affect blood vessel density but decreased VM density in melanoma brain tumor specimens (Supplementary Fig. 12i–k) These data suggest that drugs targeting YAP/TAZ could have activity in melanoma brain and extracranial metastases and enhance survival, which might be linked to their inhibitory effects on VM.

Fig. 6

CA3 prolongs overall and brain metastasis survival in a murine model of metastasis. a–c Kaplan–Meier curves for mice receiving 1 mg/kg to 10 mg/kg treatment in a left ventricle injection murine model of brain metastasis. KM curves demonstrating the correlation between CA3 treatment and time overall survival (a, log-rank test, p = 0.022), the time to death from first metastasis diagnosis (b, log rank test, p = 0.126) and from brain metastasis diagnosis (c, log rank test, p = 0.004). d Representative images of immunohistochemical staining for YAP, TAZ, BIRC5, and IgG in Cl.2A brain tumors developed after left-ventricle injection in nude mice treated as controls, or with 1 mg/kg CA3. Scale bars represent 200 µM. e Representative images of immunohistochemical staining for CD34 and PAS in the same tumors. Black arrows point to BV, red arrows point to VM structures. Scale bars represent 50 µM. Quantification of VM (f; CD34-/PAS+) and BV (g; CD34+/PAS +) in nine areas from two tumors of each treatment group. Students t-test was used to assess significance between vascular density in control and CA3 treated mice. **p < 0.001

Anti-angiogenic drugs do not inhibit vascular mimicry

Use of anti-angiogenic monotherapy in treatment of melanoma has failed to produce survival benefit [38], but anti-angiogenic therapies may have a role in advanced melanoma when used in combination with other drugs including immunotherapy (NCT03820986, NCT03776136) [39]. We therefore sought to study the effects of anti-angiogenic drugs currently used in the clinic on VM. Both subcutaneous xenograft and left-ventricle syngeneic mouse models were used to assess the effect of anti-VEGF (αVEGF), and the multiple tyrosine kinase inhibitor, lenvatinib on the formation of VM in vivo. C57BL/6 WT mice were subcutaneously injected with B16.F10 cells and treated with anti-VEGF, lenvatinib, or an IgG isotype starting 7 days after injection [54]. As expected, lenvatinib reduced the number of blood vessels identified by CD34 + /PAS + staining of subcutaneous murine tumor sections (Fig. 7a, b). The number of blood vessels were also reduced in anti-VEGF compared to control, but the difference only trended toward significance (Fig. 7a, b). Notably, the number of VM vessels (CD34-/PAS +), was comparable among the three groups (Fig. 7c). These results were confirmed using the YUMMER1.7 model where treatment with either anti-VEGF or lenvatinib significantly reduced the amount of CD34 + /PAS + vessels compared to control (Fig. 7d, e). As seen with the B16.F10 model, the number of CD34-/PAS + vessels were comparable among the three treatment groups in the YUMMER1.7 model (Fig. 7f). Moreover, these findings were further validated using the YUMMER1.7 left-ventricle metastasis model, where we similarly found a decrease in BV but not VM density with both anti-VEGF and lenvatinib treatments when compared to vehicle treated mice (Fig. 7g–i). These effects were next verified through in vitro VM assessment (Supplementary Fig. 13). As expected, we observed no significant difference in the number of VM meshes between treatments of YUMMER1.7 cells (Supplementary Fig. 13a). This result was further confirmed with YUMM1.1Br, Cl.2A, and B16.F10 cells (Supplementary Fig. 13b–d). These data suggest that highly metastatic melanoma cells might subvert inhibition of VEGF and perhaps the other targets of lenvatinib including FGFR, PDGFRα, KIT and RET to generate and maintain VM structures.

Fig. 7

Assessment of VM in subcutaneous and brain metastatic mouse models of melanoma treated with anti-angiogenic drugs. a Representative images of immunohistochemical staining for CD34 and PAS in B16.F10 subcutaneous tumors harvested from C57BL/6 mice treated as controls, with 5 mg/kg αVEGF, or with 10 mg/kg lenvatinib. Black arrow: BV; CD34 + /PAS + , red arrow: VM; CD34-/PAS + . Magnification: 20×. b Quantification of the number of BV and c Quantification of VM vessels in three areas from two tumors of each treatment group. d Representative images of immunohistochemical staining for CD34 and PAS in YUMMER1.7 subcutaneous tumors harvested from C57BL/6 mice treated as controls, with 5 mg/kg αVEGF, or with 10 mg/kg lenvatinib. Black arrow: BV; CD34 + /PAS + , red arrow: VM; CD34-/PAS + . Magnification: 20×. e Quantification of the number of BV and VM (f) in three areas from two tumors of each treatment group. g Representative images of immunohistochemical staining for CD34 and PAS in YUMMER1.7 brain tumors developed after left-ventricle injection in C57BL/6 mice treated as controls, with 5 mg/kg αVEGF, or with 10 mg/kg lenvatinib h Quantification of the number of BV, VM (i) in three areas from two tumors of control and αVEGF, and one tumor from lenvatinib. All data are expressed as mean \(\pm\) SD showing all points. All scale bars represent 200 µM. Statistical significance is determined using a one-way ANOVA ****p < 0.0001, **p < 0.001, *p < 0.05

We next assessed the effects of a CA3, and lenvatinib, and the combination thereof in a subcutaneous model using YUSIK cells which were derived from an extracranial specimen of a patient who developed rapid and overt brain metastasis (Supplementary Fig. 14). We identified a trend in the reduction of tumor growth with CA3 treatment and lenvatinib treatment, but this did not reach statistical significance (Supplementary Fig. 14a). However, combined CA3 and lenvatinib therapy led to a significant decrease in tumor volume compared to control (Supplementary Fig. 14a). Quantification of blood vessel density showed a reduction with lenvatinib treatment when compared to both control and CA3 treatments (Supplementary Fig. 14b,d). VM density in CA3 and combination therapy compared to control was significantly decreased while no reduction in VM was observed in the lenvatinib treatment group compared with control (Supplementary Fig. 14c, d). Notably, no drug related toxicity was observed (Supplementary Fig. 14e). These data suggest that dual targeting of the YAP/TAZ and angiogenic pathways could be beneficial to limit primary tumor growth, as well as the intrinsic capacity to metastasize to the brain, though the influence of VM mediating the propensity of MBM at these early stages remain unknown.