Methylation Pattern

Methylation in the human genome occurs almost exclusively at cytosine bases followed by a guanine (the CpG dinucleotide). These sites are asymmetrically distributed across the genome, concentrating in dense regions known as CpG islands near gene promoters. The methylation status of these promoter-associated islands directly controls transcriptional silencing. In liquid biopsy analytics, the binary methylation state (methylated vs. unmethylated) across thousands of targeted CpG sites creates a high-dimensional feature vector that machine learning classifiers use to distinguish between dozens of potential tissue sources.

Every differentiated cell type in the body possesses a unique methylation blueprint established during development by DNA methyltransferases (DNMTs). These patterns are remarkably consistent across individuals, making them ideal biomarkers. Key characteristics include:

• Promoter Hypermethylation: Silences genes not required for that tissue's function
• Enhancer Demethylation: Marks active, tissue-specific regulatory elements
• Gene Body Methylation: Positively correlates with transcriptional elongation

Liquid biopsy algorithms leverage reference methylome databases from sorted cell types to deconvolve the mixture of cfDNA fragments in plasma back to their organ of origin.

The gold-standard method for interrogating methylation patterns relies on sodium bisulfite treatment of DNA. This chemical reaction selectively deaminates unmethylated cytosines to uracil while leaving 5-methylcytosine intact. During subsequent PCR amplification, uracils are read as thymines. The resulting C-to-T conversion creates a digital binary code in sequencing data: a readout of 'C' indicates a methylated base, while 'T' indicates an unmethylated base. Computational pipelines must account for incomplete conversion rates and the destruction of up to 90% of input DNA during the harsh chemical treatment.

Unlike somatic mutation analysis which looks for a single variant, methylation-based tissue-of-origin determination requires probabilistic modeling across millions of CpG sites. Modern algorithms treat each cfDNA fragment as an independent observation drawn from a mixture of latent tissue types. Key computational approaches include:

• Non-negative matrix factorization (NMF): Decomposes the bulk methylation signal into constituent tissue components
• Hidden Markov Models (HMMs): Exploit the spatial correlation of methylation states along a single read
• Deep neural networks: Learn complex, non-linear patterns from bisulfite sequencing data without explicit reference methylomes

Methylation patterns are physically coupled to chromatin architecture. The positioning of nucleosomes—histone octamers around which DNA is wrapped—protects DNA from enzymatic cleavage, creating characteristic fragment size oscillations in cfDNA. These nucleosome footprints are themselves cell-type-specific and correlate with underlying methylation status. Advanced liquid biopsy classifiers integrate fragmentomics features (fragment length, end motif frequencies, and nucleosome spacing) with methylation data to improve the signal-to-noise ratio for detecting low-abundance tumor-derived cfDNA fractions, often below 0.1%.

An emerging alternative to harsh bisulfite chemistry is enzymatic methyl-seq (EM-seq) , which uses TET2 and APOBEC enzymes to oxidize and deaminate methylated cytosines without damaging DNA. This preserves fragment length integrity—critical for downstream fragmentomics analysis—and requires significantly less input material. For liquid biopsy applications, EM-seq enables the simultaneous extraction of methylation, copy number, and fragmentomic information from a single low-input plasma sample, simplifying multi-analyte machine learning pipelines and reducing pre-analytical noise.

Short fragments of DNA released into the bloodstream through apoptosis, necrosis, and active secretion. cfDNA carries the epigenetic signatures of its cell of origin, including tissue-specific methylation patterns. The nucleosome-protected fragmentation of cfDNA preserves the methylation status of CpG dinucleotides, making it the primary substrate for non-invasive methylation analysis.

• Typical fragment length: ~167 bp (nucleosome unit)
• Half-life in circulation: 16 minutes to 2.5 hours
• Concentration in healthy plasma: 1-10 ng/mL

The study of cfDNA fragmentation patterns including fragment length distributions, end motif frequencies, and nucleosome positioning. Fragmentomic features correlate strongly with methylation status because nucleosome occupancy and DNA methylation are mechanistically linked. Tumor-derived cfDNA typically exhibits shorter fragment lengths and distinct end motif profiles compared to non-tumor cfDNA.

• End motifs: CCCA, CCTG, CCAG enriched in cancer
• Fragment size ratio: tumor-derived fragments are 20-50 bp shorter
• Nucleosome spacing informs tissue-of-origin inference

Genomic regions with a high frequency of CpG dinucleotides, typically 200-2000 bp in length, located in or near ~60% of human gene promoters. CpG islands are predominantly unmethylated in normal tissues but become hypermethylated in cancer, leading to transcriptional silencing of tumor suppressor genes. These regions are the primary targets for methylation-based liquid biopsy assays.

• Definition: GC content > 50%, observed/expected CpG ratio > 0.6
• ~28,000 CpG islands in the human genome
• Hypermethylation of SFRP, MGMT, CDKN2A promoters in colorectal cancer

A chemical treatment that deaminates unmethylated cytosines to uracil while leaving 5-methylcytosine intact, enabling methylation detection via sequencing. After PCR amplification, unmethylated cytosines are read as thymine, while methylated cytosines remain as cytosine. This is the gold-standard method for single-base resolution methylation profiling, though it causes DNA degradation and GC bias.

• Conversion efficiency: typically > 99%
• DNA loss during treatment: 50-90%
• Alternative: enzymatic conversion (e.g., EM-seq) preserves fragment length

A computational method that uses tissue-specific methylation signatures to determine the cell type or organ contributing cfDNA to the bloodstream. By comparing the methylation pattern of circulating DNA against reference methylation atlases from multiple tissues, algorithms can estimate the fraction of cfDNA originating from each source, enabling detection of tissue-specific damage or tumor localization.

• Reference atlases: Roadmap Epigenomics, ENCODE, TCGA
• Deconvolution algorithms: CiberSort-like methylation deconvolution, MethylPurify
• Applications: transplant rejection monitoring, cancer of unknown primary

A genomic interval exhibiting a statistically significant difference in methylation levels between two biological conditions, such as tumor versus normal tissue. DMRs serve as cancer-specific biomarkers in liquid biopsy, with hypermethylated DMRs in promoter regions and hypomethylated DMRs in repetitive elements being hallmarks of malignancy.

• Detection methods: BSmooth, DSS, methylKit
• Typical DMR length: 100-1000 bp
• Pan-cancer DMRs: ZNF154, SEPT9, VIM promoters