Integrative multi-omics reveals two biologically distinct groups of pilocytic astrocytoma

Abstract

Pilocytic astrocytoma (PA), the most common pediatric brain tumor, is driven by aberrant mitogen-activated protein kinase signaling most commonly caused by BRAF gene fusions or activating mutations. While 5-year overall survival rates exceed 95%, tumor recurrence or progression constitutes a major clinical challenge in incompletely resected tumors. Here, we used similarity network fusion (SNF) analysis in an integrative multi-omics approach employing RNA transcriptomic and mass spectrometry-based proteomic profiling to molecularly characterize PA tissue samples from 62 patients. Thereby, we uncovered that PAs segregated into two molecularly distinct groups, namely, Group 1 and Group 2, which were validated in three non-overlapping cohorts. Patients with Group 1 tumors were significantly younger and showed worse progression-free survival compared to patients with group 2 tumors. Ingenuity pathways analysis (IPA) and gene set enrichment analysis (GSEA) revealed that Group 1 tumors were enriched for immune response pathways, such as interferon signaling, while Group 2 tumors showed enrichment for action potential and neurotransmitter signaling pathways. Analysis of immune cell-related gene signatures showed an enrichment of infiltrating T Cells in Group 1 versus Group 2 tumors. Taken together, integrative multi-omics of PA identified biologically distinct and prognostically relevant tumor groups that may improve risk stratification of this single pathway driven tumor type.

Keywords: Integrative multi-omics; Intertumoral heterogeneity; Pilocytic astrocytoma.

Fig. 1

Similarity network fusion identifies two pilocytic astrocytoma groups using integrative multi-omics data. Similarity network fusion (SNF) representation clearly segregates two PA groups. Shorter edge length and greater thickness between samples (nodes) indicate more similarity

Fig. 2

SNF groups are recapitulated in an extended PA and PA validation cohorts. a, b Hierarchical clustering of 100 gene/protein signatures based on p value allows for the expansion of groups to non-overlapping RNA sequencing (a) and mass spectrometry (b) samples. c, d 100-gene/protein signature applied to non-overlapping transcriptomic (ICGC) [17], c and proteomic (CPTAC) [31], d validation cohorts segregate samples into two groups. Lower panel, submap analyses show close relatedness between discovery and both validation cohorts. e–g Combined data set analysis of clinical features shows that the majority of younger patients belong to Group 1 and adults belong to Group 2. (p < 0.0001, Mann Whitney test, e), location shows an enrichment of infratentorial regions for Group 2 (p = 0.0024, Fisher’s Exact test, f), and Kaplan–Meier plot shows patients in Group 1 are more likely to develop recurrent tumors (p = 0.0142, log rank method, g)

Fig. 3

Ingenuity pathway analysis identifies differential canonical pathway activation between PA groups. a Distribution of protein/mRNA ratios in PA groups. Both groups display a non-centric ratio distribution (Mann Whitney test; p value = 0.0144), with an imbalance in favor of mRNA for Group 1 and protein for Group 2. Dotplots show the median protein/mRNA ratios for individual samples. b, c Group 1 PA were compared to Group 2 PA for both transcriptome (b) and proteome (c). Significant genes (fold change ± 2 and False Discovery rate q value < 0.05) were processed using IPA and significantly activated canonical pathways are shown. (Activation z-score ± 2, – log₁₀ (p value) > 1.30.)

Fig. 4

Integrative multi-omics identifies highly divergent pathways. a GSEA-based enrichment map representations based on groups (p value = 0; FDR < 0.03). Nodes (circles) represent enriched pathways identified on the basis of proteome or transcriptome expression, which are light or medium colored, respectively, and dark nodes represent overlapping terms between transcriptome and proteome expression. Edges connect pathways/nodes that share at least half of the terms defining them. Nodes grouped according to functional families are indicated on each network. b, c Enrichment plots of top transcriptomic gene set for Group 1 showing Response to Type I Interferon (b) and top proteomic gene set for Group 2 showing Hallmark Oxidative Phosphorylation (c)