Development of latent and highly metastatic human melanoma variants
To study factors regulating the switch from latent to progressively growing melanoma metastases we developed animal models that recapitulate relevant steps of metastasis, including tumor cell transit through blood vessels, extravasation, and establishment of micrometastases. We reasoned that cells endowed with different metastatic potential can be isolated from the heterogeneous tumor population of melanoma metastases and can be enriched for by rounds of stress and adaptation. Given the particular propensity of melanoma to metastasize to the brain, we started with brain metastasis (BrM) models. YUGEN8, a BRAF-mutated (V600E) and PTEN-null, short-term culture, established from a patient with BrM was subjected to in vivo and in vitro selection, and yielded several clonally related (isogenic) cell line variants of latent or highly brain-metastatic human melanomas (Fig. 1A). When delivered systemically to athymic nude mice, via left ventricle (LV) injection, parental cells have a relatively low metastatic potential and induce BrM in <50% of mice within three months. BrMs isolated from mice were dissociated into single cell suspensions, expanded in culture and subjected to two additional cycles of LV injections. We independently selected several single-cell derived clones from cultures of successive BrM, and performed an in vivo screen for their potential to generate either latent or fast-growing BrM when re-delivered systemically. For further studies, we chose three stable clones (Cl.1A, Cl.2A, and Cl.2B), which maintained their initial phenotypes at reinjection. All established cell lines express luciferase allowing for bioluminescence imaging (BLI) and tracking of metastases. Cl.1A is a metastatic latent-competent cell line variant; 70% of mice injected with Cl.1A remain BrM-free at least six months post-injection but disseminated melanoma cells can be recovered in culture from dissociated brain tissues. At the other extreme, Cl.2A and Cl.2B are aggressive cell line derivatives from different BrM and are endowed with significantly enhanced and preferential brain metastatic competency; these clones induce BrM in 100% of mice as early as three weeks, with mice demonstrating overt brain disease by eight weeks post-injection. Presence of macrometastases was confirmed by gross and microscopic examination (Fig. 1B). As expected, the survival of mice inoculated with indolent Cl.1A cells was significantly longer than that of mice inoculated with the aggressive Cl.2A and Cl.2B cells (Fig. 1C). Interestingly, latent and aggressive brain metastatic clones had similar subcutaneous growth potential, with no differences in mean tumor volume or survival (Fig. 1D and data not shown). To identify key molecular determinants of brain metastatic outgrowth in our model, we conducted transcriptomic comparison of melanoma variants with indolent or high metastatic potential. Gene expression profiles generated by isogenic clones highly correlated, indicating that modest differences emerged in their transcriptome (Fig. 1E-G). We found 50 and 61 genes (FC±2.5) differentially expressed between Cl.2A and Cl.1A or parental YUGEN8 respectively; 88 and 91 genes (FC ±2.5) were differentially expressed between Cl.2B and Cl.1A or YUGEN8 respectively (Supplementary Table 1). Of these, 32 and 41 genes were common between Cl.2A and Cl.2B, when compared to Cl.1A and YUGEN8, respectively. Several genes that encode extracellular matrix (ECM) proteins and molecules that regulate their assembly were significantly enriched in the highly metastatic variants compared to latent cells (Fig. 1H-I). The topmost differentially expressed gene in Cl.2A and Cl.2B cells compared to parental cells or Cl.1A cells was fibromodulin (FMOD).
FMOD is a mediator of the transition to progressive melanoma metastases
Early and overt brain disease was only achieved by Cl.2A and Cl.2B cells, which retain production and secrete high levels of FMOD in culture, suggesting its dysregulation might be directly linked to achieving brain metastatic competence (Supplementary Fig. 1A). Utilizing the CRISPR/Cas9 technique, we successfully silenced FMOD in Cl.2A cells and thereafter we investigated the biological consequences of FMOD loss (Supplementary Fig. 1B-C). FMOD depletion decreased the BrM activity of Cl.2A cells, which was evident by BLI, and we observed a statistically significant difference in survival between these mice and the control group (Fig. 2A). In addition, the incidence of brain macrometastasis was significantly lower (60%) with FMOD silencing compared to control (Supplementary Table 2). FMOD depletion slightly attenuated extra-cranial metastasis (Supplementary Table 2), though the difference was not statistically significant. Brain tumors from both control and FMOD depleted variant cells showed similar staining for the proliferation marker Ki67 and TUNEL reaction for apoptosis, as well as for the endothelial cell specific marker CD34 (Fig. 2B and Supplementary Fig. 1D-E).
To understand the mechanism by which FMOD regulates brain metastatic outgrowth in our model, we investigated its role in several rate-limiting steps of the metastatic cascade. Once in arterial circulation, cancer cells can extravasate, survive, and then arrest or develop as perivascular lesions. We first evaluated the in vitro tumorigenic potential of Cl.2A cells. FMOD silencing did not affect cell proliferation either under standard culture or starvation conditions (Supplementary Fig. 1F). Assessment of annexin V/propidium iodide staining showed that suppressing FMOD does not induce spontaneous cell death in our model, hence no effects on cell viability were noted (Supplementary Fig. 1G). Likewise, FMOD depletion did not impair colony formation whether cultured under standard conditions or serum starvation (Supplementary Fig. 1H). Next, we investigated the role of FMOD in cell motility in vitro, using the gap closure assay. FMOD knockout in Cl.2A cells significantly attenuated cell migration (Fig. 2C) but did not affect invasion (Supplementary Fig. 1I). To form BrMs, cells need to transmigrate through the BBB which requires them to first adhere to the brain endothelium, rich in laminin and collagen IV . FMOD silencing significantly decreased the adhesion of Cl.2A cells to tenascin, laminin, and vitronectin (Fig. 2D) but not fibronectin, collagen I, II or IV. Permeability of tumor vasculature is altered in advanced disease. To test the effect of FMOD on the BBB, we assessed tight junction integrity by measurement of trans-endothelial electrical resistance (TEER) and albumin leakiness in an in vitro co-culture model comprised of human astrocytes and primary human endothelial cells (HUVECs), which we previously developed . We showed that HUVECs, though are extracranial endothelial cells, have the ability to acquire brain endothelial cell characteristics and tight junctions . We observed a significant increase in TEER in the presence of FMOD depleted cells compared to controls, a phenomenon that was reversed in the presence of recombinant FMOD (Fig. 2E). The increase in BBB permeability in the presence of FMOD was confirmed by analysis of Texas-Red albumin extravasation through trans-wells (Fig. 2F). This suggests that FMOD might promote melanoma brain metastatic relapse by altering endothelial barrier functions to facilitate extravasation of cancer cells into the brain parenchyma.
Previous studies demonstrated that metastatic cells entice local endothelium to form angiogenic vessels . FMOD was previously linked to angiogenesis in several malignancies [22, 23]. To explore the role of FMOD in angiogenesis at the cellular level, we studied the effects of Cl.2A cells, in a co-culture system, or recombinant FMOD on endothelial cells. We found no differences in the proliferation of HUVECs when sub-cultured in conditioned medium from Cl.2A cell variants or in the presence of recombinant FMOD. (Figure 3A-B). Endothelial cell transmigration is an important step during angiogenesis regulated by chemotactic stimuli and involves degradation of ECM to allow passage of the migrating cells . HUVECs transmigration rates towards Cl.2A FMOD-depleted cells in an indirect co-culture transwell system, was significantly slower than towards control cells (Fig. 3C). Lastly, we evaluated HUVEC’s ability to form tube-like structure (TLS) in the presence of recombinant human FMOD. Consistent with previous reports [22, 23], FMOD treatment induced HUVEC tube formation, and its effects were comparable to those induced by vascular endothelial growth factor (VEGF), a potent angiogenic molecule (Fig. 3D). We noted an increased number of branches, junctions, meshes and total tube length in treated compared to non-treated cells (Fig. 3E-H). These findings suggest that the metastatic aggressiveness of Cl.2A cells may result from FMOD’s promotion of angiogenesis.
SOX2 is a rate-limiting step of metastatic outgrowth upon FMOD silencing
Though FMOD depletion significantly inhibited BrM, we hypothesized that FMOD is not functioning independently to drive the expansion of tumor foci into macrometastases. We elected to study the SOX2 transcription factor, a critical regulator of self-renewal, because like FMOD, its expression was elevated in the aggressive Cl.2A and Cl.2B cells and lost in the metastatic latent Cl.1A variant (Supplementary Fig. 1A). SOX2 expression significantly and selectively promotes metastasis to the brain in lung and breast cancers [8, 20]. We therefore examined if SOX2 was required for or contributed to metastatic formation at distant sites in our model. We successfully ablated the expression of SOX2 or SOX2 and FMOD in Cl.2A cells using CRISPR/Cas9 (Supplementary Fig. 1B-C). For reproducibility, we carried out two independent transductions from which stable clones were selected and studied further. Silencing SOX2 did not affect metastasis as we observed similar rapid weight loss and overt brain metastatic colonization in 100% of mice following systemic injection with either SOX2-depleted or control Cl.2A cells.By contrast, 92% of mice injected with SOX2 and FMOD-depleted Cl.2A cells failed to develop BrM even after 6 months of follow-up (Supplementary Fig. 2A). Moreover, macroscopic metastases at sites other than the brain remained undetectable for over 6 months in 76% of mice, as indicated by the absence of bioluminescent signal (Supplementary Fig. 2A and Supplementary Table 2). The marked inhibition of metastatic outgrowth was confirmed by a statistically significant difference in the survival rates between the experimental and control groups (Fig. 4A). The impaired metastatic capacity of dual SOX2 and FMOD-depleted cells was confirmed by microscopic examination of H&E staining of histological brain sections at the experimental endpoint. Pathology revealed rare, small tumor foci that did not develop into macroscopic metastasis (Supplementary Fig. 2B). By contrast, the brains of mice inoculated with control or SOX2-depleted cells showed multiple large lesions (Fig. 4B). Brain tumors derived from either SOX2 or control cells showed similar Ki67, TUNEL and CD34 staining (Fig. 4B and Supplementary Fig. 2C). SOX2 accumulation is a dynamic process; brain metastases that escape from dormancy accrue cells with varying levels of SOX2 as they expand . To verify if SOX2 was present only in a distinct subpopulation of cells, we conducted immunofluorescence staining and found SOX2 positive staining in the majority of DAPI-positive Cl.2A cells (Supplementary Fig. 2D). These results suggested that FMOD-regulated tumor-host interactions might be coupled with SOX2-driven signaling mechanisms in BrM in our model.
We next examined concomitantly the functions of SOX2 and FMOD. Similar to FMOD, depleting SOX2 did not alter proliferation rates of Cl.2A cells. In contrast, dual silencing of SOX2 and FMOD inhibited cell proliferation under both standard culture conditions and serum starvation, suggesting that FMOD expressing Cl.2A cells are more capable of overcoming stress conditions (Fig. 4C). The reduction in proliferation was likely attributable to cell-cycle modulation. Flow cytometry analysis showed that dual silencing of SOX2 and FMOD inhibited cell cycle progression via delayed G1 to S phase transition (Fig. 4D). Neither SOX2 depletion nor SOX2 and FMOD dual silencing had an effect on spontaneous apoptosis (Supplementary Fig. 2E). However, in the presence of etoposide, a topoisomerase II inhibitor, which induces apoptosis, SOX2 and FMOD dual silencing caused elevated levels of cytotoxicity under both, standard conditions and serum starvation (Fig. 4E). The ability of cells to form colonies reflects their capacity to initiate tumor growth. Dual SOX2 and FMOD depletion impeded colony formation in Cl.2A cells (Fig. 4F). To explore the relevance of these findings in vivo, we compared subcutaneous tumor growth rates of Cl.2A variants and found no significant differences (Supplementary Fig. 2F). Lowering the number of transplanted cells from 300,000 to 3,000 per flank did not affect tumor formation (data not shown). This suggests that the latent metastatic phenotype generated by SOX2 and FMOD dual knockout in the aggressive clone is independent of the intrinsic capacity of Cl.2A cells to proliferate and possibly controlled by specific microenvironmental factors in metastatic sites.
We next assessed the migratory and invasive properties of SOX2 depleted cells. SOX2 silencing did not affect cell motility in a wound healing assay. SOX2 and FMOD dual gene editing had no additive inhibitory effects on cell migration when compared to FMOD depletion (Supplementary Fig. 2G). There were no significant differences between the matrigel invasion indices of Cl.2A variant cells when compared to control cells (Supplementary Fig. 2H). As we previously found that FMOD depletion significantly decreased the adhesion of Cl.2A cells to tenascin, laminin, and vitronectin, we continued this study with regard to SOX2 depletion. SOX2 silencing did not alter the adhesion of Cl.2A cells to any of the ECM components tested. However, dual SOX2 and FMOD inhibition further decreased Cl.2A adhesion to tenascin and vitronectin compared to FMOD depletion only, with no additive effect on laminin (Figure 4G). Interestingly, SOX2 and FMOD dual gene editing significantly impaired cell adhesion to collagen I and II, indicating presence of a fine-tuned cooperative mechanism between SOX2 and FMOD cellular programs (Fig. 4H). Notably, SOX2 silencing had no effect on endothelial cells proliferation and did not significantly alter the effects of FMOD depletion on HUVECs transmigration in vitro (Supplementary Fig. 2I-J). Our data suggest that SOX2 and FMOD might have overlapping and distinct functions.
Upon examination of the H&E stained tumor sections formed by Cl.2A, we noted the presence of vessel-like structures seemingly formed by tumor cells. Like other type of cancer cells, melanoma cells form vascular-like structures, a process recognized as a form of neovascularization, independent of angiogenesis, and referred to as vasculogenic mimicry (VM) [25, 26]. To determine whether VM was present in BrM collected from our models, we conducted thorough microscopic examination of sections for CD34 immunohistochemical and Periodic acid-Schiff (PAS) histochemical staining. A specific feature of VM channels is a wall structure negative for CD34 staining but positive for PAS staining [26, 27]. Quantification of PAS(+)/CD34(-) vessels in tumor xenografts showed a reduction in VM content in tumors with FMOD depletion compared to tumors formed by Cl.2A cells or SOX2 knockout cells (Fig. 5A-B). However, the number of PAS(+)/CD34(+) vessels was comparable among the three tumor groups. (Fig. 5C). Since dual SOX2 and FMOD impeded formation of macrometastasis, it was impossible to assess VM in this group. Next, we sought to verify formation of tubal structures derived from our cells in vitro. Melanoma cells cultured on matrigel readily form tubular-like structures in vitro . As expected, Cl.2A cells exhibited VM in a tube formation assay conducted. Consistent with our findings in tumor xenografts, FMOD depletion partially reduced VM in vitro (Fig. 5D), as noted by the decreased number of branches, junctions, meshes, and total tube length compared to control cells (Fig. 5E-H). SOX2 depletion did not affect VM. Interestingly, cells inhibited for both FMOD and SOX2 expression completely lost the ability to form tube-like structures, suggesting once again a possible cooperation between SOX2 and FMOD-driven cellular programs (Fig. 5D-H).
Gene signatures of FMOD- and/or SOX2-silenced cells
To gain insights into how distant metastasis is regulated in our model, changes in the gene expression profiles of Cl.2A cells after suppression of FMOD and/or SOX2 were determined by microarray analysis. We focused primarily on changes following single FMOD or double FMOD/SOX2 silencing as these genetic approaches induced significant differences in vitro and in vivo. By hierarchical clustering and principal component analysis, we found that FMOD/SOX2 double knockout samples cluster separately from other samples, indicating these cell populations are biologically distinct, and supporting their dramatic phenotypic differences (Fig. 6A). By comparative analyses and using the FC±2.0 cut-off, we identified 80 differentially expressed genes (51 upregulated and 29 downregulated) in the FMOD-silenced cells and 531 differentially regulated genes (271 upregulated and 260 downregulated) in the SOX2/FMOD-silenced cells (Fig. 6B and Supplementary Table 3). To classify the differentially expressed genes into biological categories, gene sets were further subjected to enrichment analysis using the NCBI human genome as a reference list. Enriched functional categories included transcription misregulation in cancer, cell cycle, ECM-receptor interaction, focal adhesion, PI3K-AKT/TNF/IL-17/JAK-STAT signaling pathways, and cytokine-cytokine receptor interaction (Supplementary Table 4). Loss of FMOD expression was concomitantly associated with the dysregulated expression (FC ±2.0) of twelve ECM-remodeling genes including POSTN, MMP1, COL11A2, CTSK, LIF, and SPP1 (Supplementary Table 5). Dual SOX2/FMOD silencing led to increased number of dysregulated ECM-remodeling genes (FC±2.0) including POSTN, COL11A1, COL11A2, ADAMTS4, HAPLN1, TNC, TINAGL1, ITGB4, SPP1,MMP1, and MMP9, among others (Fig. 6C-E and Supplementary Table 5). We concluded that modulation of these ECM remodeling proteins likely interferes with tumor-matrix crosstalk and is at least in part responsible for the dramatic difference in the metastatic phenotype of our models.
Effects of FMOD and/or SOX2 on Hippo pathway signaling
We next sought to uncover the signaling mechanisms tied to the aforementioned phenotypic alterations. Since our models are derived from a BRAFV600E PTEN-null melanoma, we first evaluated changes in PI3K/Akt and MAPK signaling associated FMOD and/or SOX2 loss. Though FMOD depleted Cl.2A cells showed a slight decrease in p-Akt (S473) levels compared to control, this was not associated with changes in p-mTOR (S2448) or downstream effectors p-4EBP1 (S65) and p-RPS6 (S235/236) (Fig. 7A and data not shown). Furthermore, loss of FMOD resulted in negligible changes in p-ERK1/2 (T202/Y204) levels and no changes in p-MEK1/2 (S221) (Fig. 7A). FMOD is a ubiquitous component of the ECM involved in structural matrix assembly and organization . Since the Hippo pathway has been linked to ECM organization and remodeling , we explored possible effects of FMOD on Hippo signaling. Loss of FMOD was associated with decreased p-FAK (S397) and activation of the Hippo pathway via phosphorylation of its upstream mediators, MST1 (T183) and LATS1 (T1079) (Fig. 7B-C). Downstream of LATS1 we observed increased p-YAP (S127 and S397) and p-TAZ (S89) in both FMOD- and SOX2/FMOD-depleted cells (Fig. 7B-C). To confirm modulation of YAP/TAZ activity, we quantified mRNA expression levels of some of the known YAP/TAZ target genes, including BMP4, BIRC5, CDC20, and ZEB2 . We found decreased expression of these genes in both FMOD- and SOX2/FMOD-depleted cells, (Fig. 7D). Interestingly, loss of FMOD and SOX2/FMOD reduced expression of several known YAP/TAZ targets implicated in vascular mimicry, including ZEB2, SNAIL and SLUG, which are all key transcription factors that regulate the epithelial- mesenchymal transition (EMT) process, the mesenchymal markers vimentin, and survivin, also implicated in EMT (Fig. 7E) . Collectively these data indicate that FMOD and SOX2 might antagonize with Hippo signaling to regulate YAP/TAZ transcriptional activity.
Studies of FMOD and SOX2 expression and their prognostic value in melanoma patient samples
We next studied FMOD expression in human melanoma samples employing quantitative immunofluorescence and tissue microarrays. We started with a cohort of 169 metastatic melanoma patients with variable times to development of BrM . The fluorescent signal was quantified within the S100-positive area (tumor expression) or the tumor microenvironment (stromal expression) (Fig. 8A). There was a strong correlation between stroma and tumor FMOD expression (R= 0.78, p < 0.0001) (Supplementary Fig. 3A). FMOD in the stroma and tumor was higher in patients who developed BrMs at some point during their illness (t-test, p = 0.008 and p = 0.06, respectively) (Fig. 8B-C). However, there was no correlation with incidence of metastasis at other sites. This suggests that FMOD dysregulation might be preferentially associated with cerebral metastasis. We also evaluated matched cerebral and extra-cerebral metastases from a cohort of 37 melanoma cases previously described . FMOD levels in cerebral metastases were not significantly different when compared to their extra-cerebral counterparts (data not shown), indicating FMOD dysregulation is likely an early event in BrM. Examination of FMOD distribution across various metastatic sites showed a prevalence of higher stromal FMOD in visceral metastases and low expression in bone lesions (ANOVA, p = 0.03) (Supplementary Fig. 3C). We next assessed the association between FMOD levels and organ specific metastasis-free survival. There was no association between FMOD expression and time to development of brain, liver, lung or bone metastasis (Fig. 8D-E and data not shown). Increased stromal FMOD correlated with decreased overall survival (OS), though data only trended toward significance (log-rank p = 0.09) (Fig. 8F). When evaluating FMOD specifically in intracranial tumor stroma, high expression was significantly associated with decreased OS and this was confirmed in our smaller cohort of brain lesions (RR 1.99; Lower CL 1.00; Upper CL 5.10, p = 0.04; Fig. 8G and data not shown). No correlation was found between FMOD expression and gender, M stage, LDH levels or BRAF/NRAS mutational status, but levels were higher in patients over 50 years old (t-test, p = 0.02). Using anti-CD34 to identify endothelial cells we found no association between FMOD and vascular density, a finding which was validated in our second cohort of cases (t-test, p = 0.56; Supplementary Fig. 3D and data not shown). Interestingly, elevated FMOD levels were found in cranial lesions with a higher edema-to-tumor volume ratio by 3D modeling (t-test, p=0.02; Supplementary Fig. 3E) .
We next sought to investigate the possible combined effect of FMOD and SOX2 upregulation, in the context of clinical outcome. SOX2 staining was heterogeneous with a minority of positive tumor cells (Fig. 8A), a pattern consistent with its roles in a distinct subpopulation of cells such as ‘stem-like’ cells. SOX2 expression did not correlate with organ specific metastasis or survival (Supplementary Fig. 3F-G), suggesting that SOX2 dysregulation by itself is not associated with either a brain homing phenotype or with aggressive disease in general. High FMOD in either stroma or tumor was significantly associated with high SOX2 expression and this relationship held when cerebral lesions were examined separately (t-test, p = 0.02 and p = 0.0002, respectively; Fig. 9A-B). We next generated a composite FMOD/SOX2 score. We defined two groups of patients: high FMOD/high SOX2 and low FMOD/low SOX2. We found a higher prevalence of cases with elevated levels of both FMOD and SOX2 in BrM relative to other sites (χ2-analysis, 70% of cases vs. 42%, respectively, p = 0.004; Fig. 9C). In addition, we found an increased frequency of tumors with high levels of both proteins in patients who developed BrMs at some point during their illness (t-test, p = 0.05 and p = 0.08, in the stroma and tumor, respectively; Figure 9E-F). Moreover, patients with high FMOD in the stroma and high SOX2 had a higher risk for early BrM development (RR 1.32; Lower CL 0.99; Upper CL 1.82, p = 0.05; Fig. 9G). A similar trend was seen for cases with high tumor FMOD and SOX2 expression (p = 0.09; Fig. 9H). No association was found between this composite score and metastasis at other sites or OS. Our data suggests a possible role of SOX2 in tumors with FMOD dysregulation; concomitant SOX2 upregulation might contribute to its association with early development of melanoma BrMs, but not with disease aggressiveness.