Transcriptomic subtyping of malignant peripheral nerve sheath tumours highlights immune signatures, genomic profiles, patient survival and therapeutic targets

Abstract

Background: Malignant peripheral nerve sheath tumour (MPNST) is an aggressive orphan disease commonly affecting adolescents or young adults. Current knowledge of molecular tumour biology has been insufficient for development of rational treatment strategies. We aimed to discover molecular subtypes of potential clinical relevance.

Methods: Fresh frozen samples of MPNSTs (n = 94) and benign neurofibromas (n = 28) from 115 patients in a European multicentre study were analysed by DNA copy number and/or transcriptomic profiling. Unsupervised transcriptomic subtyping was performed and the subtypes characterized for genomic aberrations, clinicopathological associations and patient survival.

Findings: MPNSTs were classified into two transcriptomic subtypes defined primarily by immune signatures and proliferative processes. "Immune active" MPNSTs (44%) had sustained immune signals relative to neurofibromas, were more frequently low-grade (P = 0.01) and had favourable prognostic associations in a multivariable model of disease-specific survival with clinicopathological factors (hazard ratio 0.25, P = 0.003). "Immune deficient" MPNSTs were more aggressive and characterized by proliferative signatures, high genomic complexity, aberrant TP53 and PRC2 loss, as well as high relative expression of several potential actionable targets (EGFR, ERBB2, EZH2, KIF11, PLK1, RRM2). Integrated gene-wise analyses suggested a DNA copy number-basis for proliferative transcriptomic signatures in particular, and the tumour copy number burden further stratified the transcriptomic subtypes according to patient prognosis (P < 0.01).

Interpretation: Approximately half of MPNSTs belong to an "immune deficient" transcriptomic subtype associated with an aggressive disease course, PRC2 loss and expression of several potential therapeutic targets, providing a rationale for molecularly-guided intervention trials.

Funding: Research grants from non-profit organizations, as stated in the Acknowledgements.

Keywords: DNA copy number aberrations; Data integration; MPNST; Prognosis; Transcriptomic subtypes.

Fig. 1

Genome-wide copy number aberrations across MPNSTs. (a) Percentage of the genome affected by DNA copy number gain, loss and both (total CNA; n = 93), as well as LOH (n = 91). Number at dashed line is the median value. Scatter plots show the association between (b) loss and gain, and (c) loss and LOH. Filled circles (black) indicate tumours (n = 14) with a disproportionate percentage of LOH and loss. Statistics are from Spearman's rank correlation and the asterisk indicates the correlation coefficient after exclusion of tumours with filled circles. (d) Frequency of genome-wide gains, losses, and LOH among tumours. (e–g) Kaplan–Meier survival curves for patients with MPNSTs with high and low levels of copy number gain, loss and LOH (see Supplementary Fig. S8a for dichotomization of the tumours). Hazard ratios and 95% CIs (in brackets) are from Cox regression analyses and P-values from Wald tests. Results from analysis of the linear variables (percentage aberration) are given in parentheses. (h) Kaplan–Meier survival curves for the combined variable of loss and LOH, grouped according to panels (f) and (g). Hazard ratio was calculated by comparing patients with low percentage of genome-wide loss and LOH (orange) versus the combined group of patients with high percentage of loss and/or LOH (grey, purple, black).

Fig. 2

Gene expression characteristics of the transcriptomic MPNST subtypes. (a) Gene set enrichment analyses of the Hallmark (n = 50), TP53 and PRC2 gene sets between MPNST transcriptomic subtypes and benign neurofibromas as indicated. Gene sets with FDR-corrected P < 0.01 from camera gene set test were included for plotting, and the Hallmark gene sets were ordered according to significance levels from comparison of the two MPNST subtypes. (b) Proportion of genes per gene set with concurrent gain and upregulated expression (red), or loss and downregulated expression (blue) among MPNSTs (n = 63). Vertical dashed lines indicate the median number of concurrent genes among the 50 Hallmark gene sets, and asterisks indicate gene sets with a significantly larger number of concurrent genes than the median (evaluated by one-sided Fisher's tests). (c) Expression levels of selected immune genes in MPNSTs and neurofibromas (mean-centred and scaled to standard deviation 1). Tumours are sorted according to transcriptomic subtype and the expression level of CD274. Single-sample enrichment (ssGSVA) scores of selected gene sets plotted (d) against PC1 from principal component analysis of tumours (dashed line shows the linear regression); (e) according to tumour type and coloured as in panel (c); and (f) according to DNA copy number status of PRC2 core components. Statistics are from Spearman's rank correlation in panel (d) (calculated among the 64 MPNSTs only, neurofibromas were included for illustration), and Wilcoxon's test in panels (e) and (f).

Fig. 3

DNA copy number aberrations according to transcriptomic MPNST subtype. (a) Principal component analysis for neurofibromas and MPNSTs, coloured according to transcriptomic subtype (asterisks mark tumours from the same patient and dagger marks the plexiform neurofibroma), percentage of CNAs (gains and losses), and loss of heterozygosity (LOH). (b) Percentage of CNAs among immune active (yellow) and immune deficient (blue) MPNSTs. Number at dotted line is the median value. P-values are from Wilcoxon's test. (c) Venn diagram of genes with upregulated expression in immune deficient (n = 35) versus immune active MPNSTs (n = 28), with concurrent copy number gain and upregulated gene expression (investigated among 63 MPNSTs with gain versus neutral copy number), and gain in >20% of MPNSTs (investigated among 93 tumours). The scatter plot illustrates genes at the intersection of the Venn diagram (n = 221), and cancer-critical genes and top-ranked differentially expressed/aberrant genes are highlighted.

Fig. 4

Patient survival according to transcriptomic subtypes and DNA copy number burden. Kaplan–Meier survival curves for patients with MPNST stratified by (a) transcriptomic subtypes and (b–e) transcriptomic subtypes plus copy number gain or loss/LOH. Patients were grouped according to high and low levels of CNAs using the same thresholds as for the complete MPNST series (Supplementary Fig. S8a). Hazard ratios and 95% CIs (in brackets) are from Cox regression analyses and P-values from Wald tests.

Fig. 5

Potential target genes in immune deficient MPNSTs with a high burden of DNA copy number loss and/or LOH. (a) Venn diagram of affected genes in the subset of immune deficient MPNSTs with a particularly high frequency of loss/LOH (genes with difference in aberration frequency >40% relative to immune deficient MPNSTs with a low burden of loss and/or LOH). Thresholds to group MPNSTs according to high and low aberration levels are shown in Supplementary Fig. S8a. Cancer-critical genes with concurrent loss and downregulated expression (n = 44) were listed in the table below. The asterisk indicate genes with small-scale mutations that have previously been reported in MPNSTs.^,,,, Table columns indicate from left to right the number of immune deficient MPNSTs with loss and/or LOH of each gene; associations with disease-specific survival (hazard ratios from Cox regression analysis and P-values from Wald tests in comparisons of immune deficient MPNSTs with neutral copy number versus loss and/or LOH of each gene); and results from differential gene expression analyses between immune deficient MPNSTs with high burden of loss and/or LOH versus low burden of both (fold change on log2-scale [logFC] and P-values from limma analyses). Red bars indicate genes with significant P-values from both survival and differential gene expression analyses and (b) selected genes from each chromosomal region were illustrated with Kaplan–Meier curves of disease-specific survival and box plots of gene expression in different tumour groups as indicated (log2-scale).