The patient is a 52-year-old female who initially underwent resection of a sporadic intracranial meningioma, WHO grade I, at age 27. At age 38, she underwent salvage resection for multifocal intracranial recurrence, with surgical pathology revealing transformation to atypical meningioma, WHO grade II. At age 48, she again presented with intracranial progression and was treated with external beam radiotherapy, followed by stereotactic radiosurgery to satellite intracranial lesions at age 50. At age 52, a surveillance MRI of the brain revealed further intracranial recurrence (Fig. 1a), and she underwent whole body imaging that identified a liver metastasis (Fig. 1b). Subsequent salvage resection (Fig. 1c) with concurrent Cs-131 brachytherapy of the growing intracranial tumor, and ultrasound-guided liver biopsy, again demonstrated intracranial atypical meningioma, WHO grade II (Fig. 1d), as well as metastatic meningioma in the liver (Fig. 1e), which was verified using immunohistochemistry for somatostatin receptor type 2a (Fig. 1f) [12].
To elucidate the molecular features associated with the meningioma metastasis, we performed DNA methylation profiling and RNA-seq on 6 spatially distinct sites from the intracranial meningioma and the liver metastasis. Intracranial samples, as well as the liver core biopsy, were flash frozen in liquid nitrogen immediately after collection, and DNA and RNA were simultaneously extracted from the same sample from each site. Unsupervised hierarchical clustering of methylation data revealed that the liver metastasis demonstrated a distinct epigenetic profile from the 6 intracranial lesions (Fig. 2a), likely resulting from hepatocellular contamination in the metastatic sample. In support of this hypothesis, deconvolution of cell types from methylation data [13], with a focus on hepatocytes, showed a small hepatocyte fraction exclusively in the liver metastasis (0% versus 9.8%, Fig. 2b). However, tumor purity analysis from methylation data [15] demonstrated similar percentages in the 6 intracranial samples compared to the liver metastasis (83–88% versus 81%, Fig. 2c). Consistently, all 7 samples demonstrated high concordance with meningioma methylation profiles based on tumor classification via a random forest model (99% versus 99%, Fig. 2d) [14], suggesting that the biopsied liver lesion was indeed primarily composed of metastatic meningioma cells, rather than contaminating stromal cells or infiltrated hepatocytes. Moreover, when we calculated copy number variants (CNVs) based on DNA methylation profiles [16], we found no private CNVs in the liver metastasis compared to the intracranial samples (Fig. 2e), and that all samples demonstrated loss of chromosome 22q, which harbors the meningioma tumor suppressor gene NF2. However, we did observe 4 CNVs that were present in the intracranial lesions but lost in the liver metastasis, which may have been driven, in part, by the underlying normal hepatocyte contamination in the metastatic sample. Notably, these changes did not appear to affect DNA methylation-based tumor classification, and could, alternatively, have been reflective of metastasis of a meningioma clone not captured in the 6 intracranial meningioma samples we profiled. In summary, DNA methylation profiling indicates that metastatic meningioma, while containing detectable contaminating cells, is primarily composed of meningioma cells with a similar CNV profile to matched intracranial samples.
We next used RNA-seq and differential expression analysis to compare the transcriptomes of the 6 intracranial samples with the liver metastasis. Unsupervised hierarchical clustering of transcriptomic data segregated the liver metastasis from the intracranial samples (Fig. 3a). Notably, a large number of genes were detected exclusively in the intracranial or metastatic samples, consistent with contaminating non-meningioma cells. In order to minimize contaminating hepatocyte signatures, we filtered RNA-seq data to identify only those genes expressed at a transcripts per million (TPM) level greater than 1 in all 7 samples, resulting in a total of 16,513 genes (45% of the initial gene list). We then selected genes with a log2 fold change greater than 2, which resulted in 628 enriched genes in the intracranial meningioma samples, and 726 enriched genes in the metastasis (Supplementary Table 1). Gene ontology analysis revealed enrichment of SUZ12 and FOXM1 transcription factor networks (Fig. 3b) and mitotic spindle function (Fig. 3c) in the intracranial meningioma samples, consistent with the established roles of these pathways in regulating meningioma cell proliferation [9]. In contrast, genes enriched in the liver metastasis showed overrepresentation of metabolic pathways, and SUZ12 and hepatocyte nuclear factor 4a (HNF4A) transcription factor networks, suggestive of liver-enriched gene expression programs, rather than metastatic meningioma (Fig. 3d). In support of this hypothesis, analysis of the tissue specific expression patterns of the metastasis gene set revealed enrichment of liver restricted transcripts (Fig. 3e). These data are consistent with the notion that bulk RNA-seq has limited utility for identifying molecular signatures in meningioma metastases, even with stringent filters from a relatively pure biopsy, as evidenced by pathology (Fig. 1e), histology (Fig. 1f), cell type deconvolution (Fig. 2b), tumor purity analysis (Fig. 2c), random forest tumor classification (Fig. 2d), and copy number variants (Fig. 2e).
