Brain metastasis DNA methylomes, a novel resource for the identification of biological and clinical features

Abstract

Brain metastases (BM) are one the most lethal and poorly managed clinical complications in cancer patients. These secondary tumors represent the most common intracranial neoplasm in adults, most frequently originating from lung cancer, breast cancer, and cutaneous melanoma. In primary brain tumors, such as gliomas, recent advances in DNA methylation profiling have allowed for a comprehensive molecular classification. Such data provide prognostic information, in addition to helping predict patient response to specific systemic therapies. However, epigenetic alterations of metastatic brain tumors with specific biological and translational relevance still require much further exploration. Using the widely employed Illumina Infinium HumanMethylation 450K platform, we have generated a cohort of genome-wide DNA methylomes from ninety-six needle-dissected BM specimens from patients with lung cancer, breast cancer, and cutaneous melanoma with clinical, pathological, and demographic annotations. This resource offers an unprecedented and unique opportunity to identify novel DNA methylation features influencing the behavior of brain metastasis, and thus accelerate the discovery of BM-specific theranostic epigenetic alterations.

Figure 1. Study design for the construction of genome-wide DNA methylation profiles of metastatic brain tumors.

(a) Patients with metastatic brain tumors from breast cancer, lung cancer or cutaneous melanoma origin were selected for the study. (b) A representative magnetic resonance imaging (MRI) scan of a single metastatic brain tumor lesion used as part of the clinical diagnosis is shown in the scheme. (c) After surgery, resected tumors were routinely stored as formalin-fixed and paraffin-embedded (FFPE) tissue blocks and stained with hematoxylin and eosin (H&E) for anatomic pathology diagnosis. (d) FFPE tissue sections underwent routine immunohistochemistry (IHC) evaluation to confirm the tumor of origin and molecular subtypes of each case. (e) After tumor cell-rich areas were identified, tissue microdissection followed by DNA purification was performed in each case. (f) DNA specimens passing the quality control metrics were converted with sodium bisulfite, enzymatically fragmented, and hybridized in the HM450K BeadChips. Raw intensity data were normalized and corrected β values for each specimen were generated. A representative heat map of the DNA methylation data is shown in the study scheme.

Figure 2. Density distribution of β values across samples.

(a) The distribution of raw β values for all three BM subtypes before normalization. (b) The distribution of β values after normalization for all three BM subtypes. BM subtype is indicated by the color of the line, with green lines representing breast cancer brain metastases (BCBM), orange lines representing lung cancer brain metastases (LCBM), and purple lines representing melanoma brain metastases (MBM).

Figure 3. Distribution of β values among genomic features and BM type specific clustering.

(a) The distribution of DNAm levels in low-CpG density regions (i.e. open sea, CGI shelf, and CGI shore) with respect to high-CpG density CGIs. (b) The distribution of DNAm levels for CpG sites with respect to gene compartments in the promoter regions, 5’UTRs, the 1^st exon, gene body, and intergenic regions (IGRs). Box colors indicate BM subtype, with green boxes representing breast cancer brain metastases (BCBM), orange boxes representing lung cancer brain metastases (LCBM), and purple boxes representing melanoma brain metastases (MBM). (c) t-SNE analysis reveals BM subtype specific clustering of samples. The top 2,500 most variable CpG sites were used for the analysis with the first two dimensions shown. Each of the three BM subtypes is confined to a cluster, with MBMs depicted in purple, BCBMs depicted in green, and LCBMs depicted in orange. No samples were found to fall outside of a BM subtype specific cluster

Figure 4. Quality metrics used for sample filtering. A three-part quality control pipeline was used to identify samples with low quality.

(a) The distribution of mean detection p-values for all probes in each sample. (b) The distribution of the number of probes with missing data per sample. (c) The distribution of the top 25 probes with missing data. For panels a and b, colors represented BM subtype with green bars representing breast cancer brain metastases (BCBM), orange bars representing lung cancer brain metastases (LCBM), and purple bars representing melanoma brain metastases (MBM).