According to KEGG pathways enriched by DEGs relevant to brain metastasis using data of breast cancer, lung carcinoma and melanoma, we observed important roles of PI3K/AKT signaling and cytokine-cytokine receptor interactions in driving such a process (Supplementary Fig. 3). PI3K/AKT is the upstream regulator of many signaling pathways with known functionalities in carcinogenesis and progression (Fig. 2E). Among them, the critical roles of the PI3K/AKT/mTOR axis23 and the PI3K/AKT/MYC axis24 in mediating brain metastasis of melanoma and lung cancer have been recently proposed, respectively. The PI3K/ATK signaling can be activated by many factors such as cytokines, integrins and growth factors. The enrichment of ‘cytokine-cytokine receptor interactions’ in our brain metastatic related DEGs suggested the general role of cytokine/PI3K/AKT signaling in mediating brain metastasis. In fact, cytokine-mediated signaling has already been heavily implicated in cancer metastasis25. There is no wonder that cytokines act as important players particularly in brain metastasis, as most cytokines are secreted proteins from glial cells in the nervous system that function as local regulators to activate lymphocytes and lymphocytes have been proposed as critical predictors of brain metastasis26–29.
According to the GO analysis, top identified terms are closely related to immune response such as ‘T cell activation’, ‘leukocyte proliferation’, and ‘mononuclear cell differentiation’ (Supplementary Fig. 4). In consistent with this, the central role of the immune system in brain metastases has already been recognized, with the molecular strategies of tumors to evade immunosuppression and how the molecular mechanism dictating interactions between the immune system and the central nervous system towards brain metastasis being covered by Leibold et al. 30. During cancer metastasis, malignant cells must break the constraints of the environment of the primary tumor loci, migrate into the circulation system and survive the migration process, extravasate at their destination, and establish a new niche within a foreign environment, the process of which all require reprogrammed immune system to help them evade detection by the immune system.
Among the three types of cancers under study, melanoma diverged most significantly from the other two types of cancers regarding both KEGG pathways and GO terms. Specifically, all top enriched KEGG pathways and approximately 76% enriched GO terms in melanoma were not shown in the pan-cancer analysis (Supplementary Fig. 3, Supplementary Fig. 4). This reflects molecular differences on the mechanisms of tumor initiation and progression regarding tissue-of-origin but not differences brought by the size of gene list as DEGs from melanoma (i.e., 2324) are over double folds of that from lung cancer (i.e., 1028). Also, the degree of divergence from the pan-cancer results may be a reverse reflection on the likelihood of gaining brain metastasis, as breast and lung cancers constitute as the most important sources of increased incidence of brain metastasis observed in Sweden31.
We identified and proposed the pivotal role of CD44 in mediating brain metastasis in this study. Once a tumor cell has left the primary tumor site, invaded the surrounding stroma, crossed the endothelial barrier into the systemic circulation, and made its way to the vessels connecting to the brain, it may be arrested in the microvasculature at the vascular branch points that facilitates its attachment with the endothelium, the process of which involves interactions between CD44, selectins, integrins, cadherins, and ICAMs/VCAMs32,33. CD44 is a known cell surface adhesion receptor and a cancer stem cell marker actively involved in the epithelial-mesenchymal transition (EMT) process34, with the CD44+/CD24− phenotype being considered contributing to malignant relapse35. Intriguingly, CD44 conveyed favorable prognostic values on cancer OS, RFS and DMFS (Supplementary Fig. 5) and was lowly expressed in metastatic samples in this study. Consistent with this, CD44 is highly expressed in patient samples without circulating tumor cells (CTCs) but lowly presented in those having CTCs 36. Thus, brain metastasis is a dynamic process involving multiple steps, where CD44 expression may be elevated during EMT and drive tumor stemness but decreased once entering the metastatic site.
Proteins highly connected with CD44 and with similar profiles regarding brain metastasis, i.e., MMP9 and TP53 (Fig. 2B), are both canonical markers associated with cancers. In particular, MMP9 (matrix metalloproteinase 9) high expression, which is necessary for tumor cells to break the constraint of the extracellular matrix during invasion, has been reported in breast cancer patients with brain metastasis 37. TP53 (a well-known tumor suppressor with critical roles in genome stability maintenance), once mutated, has been associated with increased frequency of brain metastasis in many tumors38,39. Other proteins with close associations with CD44 include EGFR40 and KRAS41, which are all well-studied players during carcinogenesis and, in particular, brain metastasis.
KEGG pathways enriched by the ‘brain metastasis transcriptional profile’ suggested the critical role of cancer-associated proteoglycans and miRNAs in brain metastasis. Indeed, the role of proteoglycan in metastasis cascade including cancer cell proliferation, angiogenesis, EMT, migration, intravasation, survival during circulation, extravasation, and colonization at the metastatic site have been critically reviewed by Ahrens, T.D. et al. 42. There have already been several publications reporting the importance of epigenetic regulations on brain metastasis involving miRNAs. For instance, loss of miRNA-let-7d43 and exosomal miRNA44 were shown to promote brain metastasis of breast cancer cells. Also, miRNA-768-3p was shown to suppress brain metastasis of lung cancer cells via suppressing KRAS expression 45, miRNA-197 and miRNA-184 were found to be associated with brain metastasis in EGFR-mutant lung cancers 46, a panel composed of miRNA-210, miRNA-214 and miRNA-15a was proposed for the prognosis of IIIA-N2 lung adenocarcinoma brain metastasis47, and the miRNA-1207-5p/EPB41L5 axis was shown to promote non-small cell lung cancer brain metastasis 48. Lastly, a miRNA-based signature composed of miRNA-150-5p, miRNA-15b-5p, miRNA-16-5p, and miRNA-374b-3p was been proposed to detect brain metastasis of melanoma cells 49. GO terms enriched by the ‘brain metastasis transcriptional profile’ implicated the importance of ‘intrinsic apoptotic pathway’ in this process, consolidating the tight connection observed between CD44 and TP53 (Fig. 2B).
Identified from data on brain cancer metastasis for breast cancer, lung carcinoma and melanoma, we assessed the expression level of CD44 and its prognostic value on patient OS among primary brain primary cancers taking glioma as the example. The purpose here is to establish the connection between primary and metastatic brain cancers and construct the index feasible for brain cancer prognosis independent of its origin.
CD44 was over-represented in both LGG and GBM types of gliomas, but the same pattern was not observed in all tumor types available in GEPIA (Fig. 3A), suggesting the expression profile and functionality of CD44 is tumor-type specific. In addition, CD44 gene expression was positively associated with tumor grade (Fig. 3B), suggestive of the promotive role of CD44 on brain cancer carcinogenesis. Interestingly, CD44 gene expression was low in patients carrying IDH mutation or 1p/19q co-deletion (Fig. 3C), implicative of the non-redundant role of CD44 in brain cancer prognosis that is independent from IDH mutation and 1p/19q co-deletion.
The poor prognostic value of CD44 gene expression on glioma (including both LGG and GBM) OS has been validated using multiple transcriptional data stored in CGGA, GEPIA and TCGA (Fig. 4A-4C), and by stratifying data into 1-year, 3-year, 5-year groups, CD44 gene expression was found to perform best in predicting 1-year glioma OS (Fig. 4D) that is of the most practical relevance in clinics given the fast progression of brain cancers and their short course of disease.
Through univariate and multivariate Cox regression analysis, we identified four factors with independent prognostic value on gliomas, i.e., CD44 gene expression, WHO grade, IDH mutation and 1p19q co-deletion (Fig. 5). By unifying them into one single index, i.e., ‘Total Points’, the prognostic power of the model on patient OS substantially improved, with the best performance being obtained on predicting glioma 3-year OS.
Lastly, we experimentally explored the effect of CAP on CD44 and its associated onco-therapeutic potential against brain cancers. The results showed that CAP significantly reduced CD44 expression (Fig. 6A). In consistent with this, CAP effectively arrested the growth of LGG and GBM cells (Fig. 6C), halted their migration abilities (Fig. 6D) and reduced their cancer stemness (Fig. 6E).