Background
Patients with melanoma brain metastasis (intracranial melanoma metastasis) still have a very unfavorable prognosis. The mean overall survival of patients with untreated intracranial melanoma metastases is as little as 4 months [
1], improving only to a median survival of 22.7 months with surgery followed by immunotherapy [
2]. There are several emerging treatment options like targeting MAPK signaling using BRAF or MEK inhibitors or immunotherapy using anti-CTLA4/anti-PD-1 antibodies that improve patient survival. These treatment options are effective for patients with extracranial metastases [
3]. For patients with intracranial metastases, the treatment response duration of BRAF/MEK inhibitors is only limited to a few months and the efficacy of immune checkpoint inhibitors is substantially reduced in symptomatic patients [
1,
4‐
8]. Results of important clinical immunotherapy trials for melanoma patients with intracranial metastases are summarized in [
9]. Clinical outcomes of melanoma patients with intracranial metastases treated with stereotactic radiosurgery and various systemic therapies have been analyzed in [
10]. The overall 12 month survival rates for the combined anti-PD-1-CTLA4 therapy was 68%, 62% for BRAF/MEK inhibitor treatment, 59% for anti-PD-1 therapy, and 45% for anti-CTLA4 therapy compared to only 21% for BRAF inhibitor treatment and 15% for conventional chemotherapy. The combined treatment of patients with anti-PD-1-CTLA4 therapy showed the best median overall survival of about 21 months. The updated five year data from patients with intracranial melanoma metastases enrolled on the ABC trial investigating nivolumab plus ipilimumab or nivolumab alone confirmed the high anti-tumor activity of nivolumab plus ipilimumab in patients with asymptomatic intracranial metastases (5-year intracranial progression-free survival 52%, 5-year overall survival 55%) [
11]. The NIBIT-M2 trial showed persistent therapeutic efficacy of ipilimumab plus nivolumab with a seven year overall survival rate of 42.8% in asymptomatic patients [
12]. Further, in a recently published retrospective study with 376 patients with intracranial melanoma metastases treated with nivolumab plus ipilimumab, long-term survival was seen in treatment-naive, asymptomatic, steroid-free patients as well as in those patients that received stereotactic radiosurgery in combination with nivolumab plus ipilimumab [
13].
Still, the therapy resistance of intracranial metastases is the leading cause of death of melanoma patients [
14,
15]. Unfortunately, intracranial metastases are also very common affecting about 50% of all metastatic melanoma patients [
1]. Therefore, a detailed molecular characterization of differences between intra- and extracranial metastases is needed to better characterize molecular alterations and mechanisms that distinguish both types of metastases to provide a basis for future developments of novel therapeutic strategies.
Several studies were done over the last years to identify differences between intra- and extracranial melanoma metastases. Different genetic alterations in genes like
BRAF,
NRAS and
CDKN2A were suggested to contribute to a unique molecular profile of intracranial metastases [
16]. Up-regulation of PI3K/Akt signaling in intracranial metastases has been reported as a key mechanism for uncontrolled proliferation of intracranial metastases [
17‐
20]. Epigenetically regulated genes (e.g.
CSSP1,
GRB10,
NMB,
PDXK,
PRKCZ,
RASL11B,
STK10 and
WDR24) with altered promoter methylation and corresponding differential expression in intra- compared to extracranial metastases were recently identified [
21]. Single-cell sequencing of metastases revealed a neuronal-like cell meta-program of intracranial melanoma metastases [
22]. Another single-cell sequencing study delineated brain metastases programs into a proliferative and an inflammatory archetype [
23]. All these studies compared intracranial metastases from several patients to extracranial metastases from other patients. However, melanoma metastases in different organs from a patient are more similar to each other than to metastases from other patients in the same organs. This patient-specific heterogeneity of molecular data from melanoma metastases has been observed in other studies before [
18,
24,
25]. Therefore, such differences between individual patients should be included in the data analysis to further improve the identification of molecular differences between intra- and extracranial melanoma metastases.
To account for the inter-patient heterogeneity, some studies already started to investigate patient-matched metastasis pairs at different molecular layers. Chen et al. [
18] analyzed gene mutations, DNA copy number alterations and gene expression profiles identifying the PI3K/Akt signaling pathway as a potential therapeutic target. Fischer et al. [
26] performed RNA sequencing and multiple immune-relevant sequencing analyses and identified a significant immunosuppression and enrichment of oxidative phosphorylation in intracranial melanoma metastases. We have recently performed a personalized analysis of genome-wide DNA-methylation profiles and revealed a global decrease of DNA-methylation intracranially [
25]. These DNA-methylation changes between patient-matched intra- and extracranial melanoma metastases affected many genes involved in cellular signaling, growth, adhesion and apoptosis and further supported the presence of a neuronal phenotype. Further, we were also able to predict potential downstream targets of genes with altered promoter methylation, which allowed to group heterogeneous patient-matched melanoma metastasis pairs into three homogeneous subgroups utilizing a network-based approach [
27]. In addition, mutations in driver genes (most frequently
ARID1A,
ARID2 and
BRAF), which distinguished intra- from extracranial melanoma metastases, were identified by targeted next-generation sequencing [
28].
Here, we perform a personalized analysis of patient-matched gene expression profiles of intra- and extracranial melanoma metastasis pairs to account for the common developmental origin of patient-matched metastases. To realize this, each patient-matched metastasis pair was analyzed by a Hidden-Markov Model (HMM) approach to identify differentially expressed genes for each pair. This was done by transferring the recently used HMM-approach for the personalized analysis of DNA-methylation profiles by [
25] to the analysis of gene expression profiles. The predicted differentially expressed genes were further used to identify frequently affected cellular pathways and to derive a set of genes that was altered in the same manner in the majority of patients. In-depth literature analysis in combination with comparisons to independent related studies were performed to provide further hints which genes potentially play an important role to establish molecular differences between intra- and extracranial melanoma metastases. Additionally, significant expression associations of several of these genes with decreased survival of melanoma patients from a large public patient cohort indicate the relevance of our findings.
Discussion
Melanoma brain metastases still present a major clinical challenge compared to extracranial metastases [
59]. Therefore, the identification of driver genes and pathways that distinguish intra- from extracranial metastases is an important next step to provide a basis for the development of new therapeutic strategies. To contribute to this, we analyzed gene expression profiles of melanoma metastases from 16 patients by utilizing a computational strategy for the personalized analysis of individual patient-matched melanoma metastasis pairs.
In a first analysis step, we could show that melanoma metastases are clearly more similar to other melanoma metastases of the same patient than to melanoma metastases from different patients in the same tissue. The corresponding patient-specific clustering of gene expression profiles of melanoma metastases can be accounted to the common evolutionary development of patient-matched metastases from the same primary tumor. A very similar behavior of melanoma metastases has previously been observed for gene expression data [
26] and DNA-methylation profiles of melanoma metastases [
25]. Such a patient-specific clustering of melanoma metastases samples suggested the strong need for a personalized analysis of patient-specific melanoma metastasis pairs. This was realized by using a three-state Hidden Markov Model (HMM) approach to predict the most likely underlying gene expression state of each gene in each metastasis pair. HMMs are valuable tools to classify molecular alterations that distinguish intra- from extracranial melanoma metastases in individual patients. This has recently been demonstrated for the prediction of DNA-methylation changes of patient-matched melanoma metastasis pairs [
25] and was also previously shown to work well for the identification of differentially expressed genes in individual tumor expression profiles [
32,
33].
The individual prediction of differentially expressed genes for each patient-matched melanoma metastasis pair provided by the HMM was used to analyze individual enrichments of differential expression in cancer-related pathways. This revealed six cancer-associated signaling pathways that were frequently enriched for differential expression (cytokine-receptor interaction, calcium signaling, ECM-receptor interaction, cAMP signaling, Jak-STAT, and PI3K/Akt signaling).
PI3K/Akt signaling has already been frequently reported to play an important role in melanoma brain metastasis [
17‐
19,
25,
60]. Further, cytokine-receptor interaction pathway genes and ECM pathway genes were reported to be differentially methylated in melanoma brain metastasis [
25] and cytokine receptors have been reported to be significantly differentially expressed in melanoma cell lines [
61]. Further, only little is known about the role of the cAMP signaling pathway in regard to melanoma. However, there is some evidence that cAMP signaling plays an important role in melanoma through a link to the MAPK pathway [
62]. Moreover, the calcium signaling pathway was also frequently significantly enriched for differentially expressed genes and has been reported to influence tumor cell proliferation, invasion and cell death [
63]. Recently, calcium signaling was reported to trigger communication between glioblastoma tumor cell networks with impacts on tumor cell viability and tumor growth [
64]. In our cohort, a significant enrichment of the calcium signaling pathway genes with increased expression in intra- compared to extracranial metastases was found for every patient-matched metastasis pair. In addition, our candidate genes with increased expression in patient-matched intra- compared to extracranial metastasis pairs, which were shared across 50% of all patients, were enriched for functional terms associated with a brain-like phenotype. These results are also supported by the identification of a brain-like phenotype in other related studies [
22,
25,
43]. Our additional direct comparisons of intracranial metastases to normal brain tissues further suggest that the observed brain-like phenotype is potentially jointly driven by normal cells in the metastases microenvironment and tumor cells of the intracranial metastases. Especially the genes predicted to be significantly up-regulated in the intracranial metastases compared to the normal brain tissues could be interesting candidates for further experimental validations of the brain-like phenotype (
CDH15,
GJC3,
MSX1,
OR7A5,
ZIC1).
Immune-relevant pathways were almost exclusively (except patient P42) enriched for decreased expression in intracranial metastases in the personalized analysis of the patient-matched metastasis pairs by the HMM. Importantly, this downregulation of immune pathway genes in intracranial metastases was found in relation to all types of extracranial metastasis independent of the fact in which tissue they occurred. This observation was also supported by the gene ontology analysis of the top-ranked decreased genes shared across multiple patients, which included many immune-related terms. Such an underexpression of immune-relevant genes in bulk samples of intracranial metastases could be associated with the blood brain barrier, but it has previously been reported that this barrier is compromised in melanoma patients with intracranial metastases [
65,
66]. Further, in accordance with our study, a significant immunosuppression in intracranial melanoma metastases was previously reported in a closely related study [
26]. Thus, a downregulation of immune pathways may contribute to the poor prognosis of intracranial metastasis.
Next, we derived a candidate set of genes that consistently showed increased or decreased expression in intracranial metastases of multiple patients. This candidate gene set was compared to the candidate gene sets of three related melanoma metastases studies [
18,
22,
26] to determine how frequently individual altered genes were observed in other transcriptome analyses. Overall, a total of eight of our candidate genes (down:
CILP,
DPT,
FGF7,
LAMP3,
MEOX2,
TMEM119; up:
GLDN,
PMP2) were also differentially expressed in the same direction in two or three of the related studies. All of these genes were reported to play a role in cancer (Table
2). These genes could therefore be of great interest for future experimental studies. This gene set contained
TMEM119, which was the only gene that was altered in the same way in all three related studies and our study.
TMEM119 can be used as a marker for microglia [
44]. However, microglia are predominantly found in the brain environment and not in the extracranial environment. Therefore, it is surprising that
TMEM119 expression was observed to be decreased in intra- compared to extracranial metastases. Further experimental studies are required to investigate this observation.
In a next step, we analyzed the expression behavior of our candidate genes in relation to publicly available data of primary and metastatic melanoma from TCGA [
42] to identify if the expression behavior of these genes was associated with patient survival. In total, 38 of our 103 candidate genes were measured in the public TCGA data set. We observed a significant association between gene expression and survival for 11 of those 38 genes (from most to least significant:
CXCL11,
CCL8,
ST6GALNAC5,
PLA2G2D,
KRT5,
FGF7,
CD38,
CD48,
CD3D,
HSD11B1,
CD8B). All of these genes showed decreased expression in our intracranial melanoma metastases compared to their corresponding extracranial metastases. In the context of the TCGA cohort, reduced expression of ten of these genes was associated with significantly shorter survival compared to patients who showed increased expression. Thus, these genes may also have the potential to contribute to the poor prognosis of intracranial metastases. Most of these survival-associated genes are involved in the regulation of immune responses. This finding supports the generally known association between immune infiltration and survival for metastatic melanoma patients [
67,
68]. However, there were also three other survival-associated genes that are involved in cell metabolism (
HSD11B1,
ST6GALNAC5) and cell growth (
FGF7). We also found genes that are associated with melanoma metastases formation (
CCL8,
CD38) [
55,
57] and melanoma cell migration and invasion (
KRT5) [
58]. Thus, the significant association between the expression of candidate genes and patient survival suggests that the expression behavior of these genes may also contribute to the poor prognosis of intracranial melanoma metastases. Still, a limitation of our survival analysis is that only 30 of the included TCGA patient samples were from distant metastases, whereas the majority of samples were from regional metastases. All metastases in our cohort were distant metastases, but a robust and potentially best matching survival analysis that would only consider the 30 distant metastases from TCGA was not possible, because only seven patients would have been in each group of the gene-specific survival comparison of the high and low expression group.
Further, several of the genes with increased expression in intra- compared to extracranial metastases predicted in at least 11 of 16 melanoma patients could potentially be of therapeutic relevance for the development of targeted treatment strategies for intracranial metastases (e.g.
PPBP,
HEPACAM,
SLC24A2,
SLC38A11,
FMN2,
PMP2).
PPBP, also known as chemokine
CXCL7, is involved in the stimulation of PI3K/Akt signaling [
69] and known to promote breast cancer progression [
70].
CXCL7 can be targeted
in vitro,
in vivo and clinically by PI3K/Akt inhibitors [
19,
71,
72]. Further, CSF1R inhibitors (GW2580) reduce myeloid cells in the tumor microenvironment of gliomas and significantly decrease the expression of chemokine
CXCL7, thus inhibiting tumor growth [
69,
73].
HEPACAM is involved in cell motility and cell-matrix interactions, known to control astrocyte self-organization and coupling [
74], and able to suppress cancer cell growth and to induce migration [
75]. Further analyses are necessary to analyze the role of the overexpression of
HEPACAM in intracranial metastases to characterize its therapeutic potential. The two transporters
SLC24A2 and
SLC38A11 might be involved in metabolic reprogramming and adaptation to the brain-specific microenvironment [
76]. Amino acid transporters are investigated in clinical trials and can be targets for cancer therapy [
77,
78].
FMN2 has essential roles in the organization of the actin cytoskeleton and cell polarity and has been reported to promote cell cycle arrest by inhibiting the degradation of the cyclin-dependent kinase inhibitor p21 [
79]. Circular RNA of
FMN2 has been reported to play a role in colorectal cancer [
80]. The role of the increased expression of
FMN2 in intracranial metastases should be analyzed by additional experiments. The myelin protein
PMP2 is involved in the regulation of melanoma cell invasion and may present a novel therapeutic target [
51].
Finally, it is important to note that our study only includes a limited number of patients for which patient-matched metastases were available. This reduces the ability to generalize all findings, but the cohort size was still large enough to clearly demonstrate the need for a personalized analysis of the transcriptomes of the patient-matched metastasis pairs. Our predicted top candidate genes were in good accordance with three closely related melanoma metastasis studies and expression associations of candidate genes with patient survival indicate their clinical importance. Both things clearly support the relevance of our work and novel findings.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.