Bioinformatic Analysis of Gene Expression for Melanoma Treatment

Abstract

Bioinformatic analysis of genome-wide gene expression allows us to characterize cells, including melanomas. Gene expression profiles have been generated in various stages of melanomas and analyzed by researchers in unique ways. Lauss et al. compared their melanoma subtypes with those of The Cancer Genome Atlas Network and found consistency between the two studies.

Molecularly dissecting the biology of cancers may help to subclassify them in a manner that assists in their optimal treatment. Gene expression profiling is a powerful method that can aid in identifying important mechanistic pathways. High-throughput RNA sequencing approaches provide strong tools to comprehensively quantify gene expression. Bioinformatic analyses such as gene ontology strategies enable investigators to assess cell states by categorizing discrete and recurrently upregulated or downregulated genes. Researchers have generated useful gene expression datasets in melanomas and have analyzed them in multiple ways to classify melanomas. In these molecular classification approaches, it is important to determine whether the findings are consistent among those analyses. Lauss et al. (2016) attempted to answer this question by comparing the underlying biological processes in their classification schemes with those of The Cancer Genome Atlas Network (TCGA) (Cancer Genome Atlas Network, 2015).

Subtypes of melanoma

In 2010, Jönsson et al. categorized melanomas into four groups based on gene expression: “High-Immune,” “Normal-like,” “Pigmentation,” and “Proliferative” (Jönsson et al., 2010). Simultaneously, the multi-institutional TCGA initiative led to a large database from which investigators categorized melanomas into three groups, with related but nonidentical designations: “Immune,” “Keratin,” and “MITF-Low” (Cancer Genome Atlas Network, 2015). Lauss et al. (2016) compared these datasets and identified common genes between several of these groups. Specifically, their study revealed overlap between: (i) the High-Immune/Pigmentation (Jönsson dataset) and Immune (TCGA) groups, (ii) the Normal-like/Pigmentation (Jönsson) and Keratin (TCGA) groups, and (iii) the Proliferative (Jönsson) and MITF-Low (TCGA) groups. Collectively, their analyses suggested that the two research teams categorized genes into similar gene ontologies, even though each group used a different gene set for subtype discovery based on unique algorithms (486 genes [Jönsson] and 1,500 genes [TCGA]), among which only 34 genes were common to both gene sets.

It remains incompletely understood what each subtype means functionally and biologically and how genes of each subtype are regulated. For example, Lauss et al. (2016) found that “High-Immune” and “Immune” subtypes predicted better prognosis for patients. However, expression of both archetypal proimmunogenic and immunosuppressive genes can be highly enriched in the immune subtypes. How the balance in the expression of different immune genes determines function—for example, activation or suppression of antitumor immunity—remains to be determined. It will be important to gain further understanding of these mechanisms to predict melanoma status and to treat patients optimally.

MITF as a marker for subtypes

Microphthalmia-associated transcription factor (MITF) plays a dominant role in transcriptional regulation in melanocytes and melanoma, and it may function as a melanoma oncogene (Levy et al., 2006). MITF regulates not only melanocyte-specific genes (such as pigmentation genes), but also ubiquitously expressed fundamental growth or survival genes such as BCL2 and CDK2 (Levy et al., 2006). Thus, it is reasonable that expression levels of MITF and MITF-target genes, especially pigmentation genes, define some of the subtypes in the two studies. It remains incompletely understood how MITF is regulated in each melanoma. MITF appears likely to be regulated by distinct pathways or factors in different melanomas, and these factors could affect responses to therapies and thereby predict prognosis. By combining gene signatures that may be related to MITF activity, such as immune genes, it may become possible to categorize melanomas into new subtypes, and the recognition of these subtypes might improve predictions of responses to therapies and prognostication in the future.

Advanced gene expression profiling

Single-cell gene expression analytic technologies allow us to characterize gene expression diversity among individual cells within a tumor, thereby providing higher resolution analysis of variability within cell populations. Using these technologies, Ennen et al. (2015) and Tirosh et al. (2016) found that gene expression levels in individual cells of a melanoma tumor are notably heterogeneous. High- and low-MITF cells coexist within the same tumor, thus demonstrating that this prior categorization of melanomas was based more on average gene expression among tumor cells than on uniform levels among cells within a tumor (as previously suggested by immune-staining observations). These technologies are also useful to characterize minor—but key—populations of cancer cells, such as stromal cells, immune infiltrating populations, circulating tumor cells, and cancer initiating/stem-like cells that may influence tumor progression and responses to treatments. Moreover, combining gene expression with DNA sequencing and epigenetic analyses may further increase the depth and accuracy of analyses. For example, Cancer Genome Atlas Network (2015) characterized melanomas by analyzing combinations of genome-wide gene expression, whole-genome sequencing, and certain epigenetic hallmarks such as DNA methylation and histone modifications.

Gene expression analysis for melanoma therapy

In treating patients with melanoma, a gene expression profile might be informative for predicting responses to specific therapies and uncovering mechanisms of sensitivity or resistance to the therapies. Hugo et al. (2015) analyzed gene expression, whole-exome sequencing, and DNA methylation in pretreated and mitogen-activated protein kinase inhibitor-resistant clinical melanomas. They found that upregulation of c-MET and downregulation of LEF1 by DNA methylation may play key roles in resistance. They also found, based on gene expression, that T cells are exhausted in resistant melanomas.

Tirosh et al. (2016) characterized gene expression of single cells in various stages of melanoma, with or without treatment, including tumor cells, immune cells, and stromal cells. Their findings included high- and low-MITF melanoma cells within individual tumors, potential interactions between cancer-associated fibroblasts and immune cells, and immune cell exhaustion states.

Van Allen et al. (2015) undertook to find markers to predict responses to anti-CTLA-4 treatment against melanoma. They performed whole-exome sequencing and RNA sequencing in the pretreated clinical samples. They found that overall mutation and neoantigen loads, as well as mRNA expression levels of GZMA and PRF1, correlated with better clinical outcomes, although the magnitude of differences in mutational/neoantigen load was not sufficient to predict easily how individual patients would respond to immune checkpoint blockade.

Hugo et al. (2016) performed whole-genome sequencing and RNA sequencing to study anti-PD-1 treatment resistance mechanisms in melanomas. They found that overall mutation loads and BRCA2 mutations correlated with better clinical responses. They also found that genes categorized as mesenchymal, immunosuppressive, monocytic, or inflammatory, as well as wound healing- and angiogenesis-associated genes, were enriched in the anti-PD-1 treatment-resistant melanomas.

Thus, combinatorial analyses of gene expression and other datasets such as whole-exome sequencing may predict responses to therapies and prognosis more accurately than analysis of gene expression alone. Additional data such as characterization of histone modifications and other epigenetic features, coupled with high-throughput sequencing approaches, may refine such combinatorial analyses even further.

Future of big data analysis

Although many molecular fundamentals of cell biology have been uncovered through studies of highly defined cell populations in vitro, the analysis of cells within in vivo contexts affords opportunities to uncover important functional aspects that are especially relevant to disease settings. Use of high-throughput “Big Data” analytic datasets and tools is vital to deconvolute the inherent variability that occurs between individual humans, between discrete subtypes of a related tumor, and among cells within a single tumor population. Therefore, evidence of concurrence across datasets is important for validation, while even leading to new questions. Attempts to understand the functional biology underlying expression “classifiers” may provide important insights into key biological pathways that regulate the myriad features of tumor biology, including responses to therapy.

Some extensions of big data could enrich bioinformatics analyses. Not only coding RNAs but also noncoding RNAs clearly play important roles in cells, and RNA sequencing has revealed an increasing number of unique noncoding RNAs. Many of these, including microRNAs and long noncoding RNAs, remain to be understood functionally, although they often modulate expression or translation of coding RNAs. These bioinformatic analyses may thus be refined by the inclusion of both coding and noncoding RNAs in the datasets—both to enhance clinical predictions and to reveal functionally important relationships. Currently, elevated serum lactate dehydrogenase remains one of the best clinical prognostic markers for patients with advanced melanoma, although this is limited by a high false positive rate (Kluger et al., 2011). Molecular and physiological data not only from tumors but also from other cell types and tissues could aid in predicting tumor responses to therapies as well as clinical prognosis. Although the study of gene expression is no longer a “new” technology in cancer, the explosion of high-throughput and high-resolution technologies, including analytic tools, and access to multiple patient populations collectively suggest that opportunities for biological discovery and clinical translation still exist, and these should provide valuable insights in the coming years.

Clinical Implications.

• Genetic comparisons of melanoma subtypes may now be made from distinct population datasets.

• From two independent groups, gene classes were consistently enriched in the corresponding melanoma subtypes.

• Based on genetic analyses, immune subtypes are better at predicting prognosis than the classification scheme.