Long-Read DNA Methylation Analysis for Single-Molecule Resolution

Native Methylation Detection: How It Works Without Bisulfite

Both ONT and PacBio long-read platforms detect DNA methylation directly from native double-stranded DNA, without the chemical conversion step that has defined methylation analysis since the 1990s. This matters because bisulfite treatment converts unmethylated cytosine to uracil under harsh acidic conditions that degrade DNA, introduce GC bias, and erase the distinction between 5mC and 5hmC — both read as "C" after bisulfite unless additional chemical steps (oxBS, TAB-seq) are applied.

Native detection avoids all of these artifacts. Both platforms share a common principle — modified bases are identified by their physical signature on individual DNA molecules — but differ in how that signature is read: ONT measures ionic current disruption as DNA passes through a protein pore, while PacBio measures altered polymerase kinetics during synthesis in a zero-mode waveguide. The illustration below compares both detection principles at the molecular level.

Oxford Nanopore: Raw Current Signal Disruption

In Nanopore sequencing, a DNA molecule passes through a protein nanopore embedded in a membrane. An ionic current flows through the pore; as each base (or modified base) transits the narrowest constriction, it disrupts the current in a characteristic way. 5-Methylcytosine and 5-hydroxymethylcytosine alter the current signal differently from unmodified cytosine — and from each other.

A neural network basecaller (Dorado with the 5mC/5hmC model, or Remora/Megalodon for earlier workflows) analyzes the raw current trace in ~5-mer to ~9-mer context windows, assigning each base a modification probability. The output is a standard BAM file with MM/ML tags specifying per-base modification calls and their confidence scores. Tools like Modkit convert these to methylation frequency bedGraph tracks and phased methylation haplotypes.

Key advantages of ONT methylation detection:

• Ultra-long reads — routinely 50–200 kb, enabling phasing across entire imprinted domains, segmental duplications, and centromere-proximal regions
• Simultaneous 5mC + 5hmC discrimination — the current signal differs measurably between the two modifications
• 6mA and 4mC detection — relevant for prokaryotic methylation, mtDNA, and specific eukaryotic contexts (e.g., 6mA in glioblastoma, C. elegans)
• Adaptive sampling — software-defined enrichment (ReadUntil API) to selectively sequence regions of interest without additional wet-lab capture

PacBio HiFi: Kinetic Signature Detection

PacBio SMRT sequencing detects modified bases through polymerase kinetics. During sequencing, a DNA polymerase incorporates fluorescently labeled nucleotides into a complementary strand in a zero-mode waveguide (ZMW). The time interval between successive incorporations — the inter-pulse duration (IPD) — slows measurably when the polymerase encounters a modified base. 5mC and 5hmC each leave a characteristic IPD and pulse-width (PW) signature.

The PacBio Revio system combines circular consensus sequencing (CCS) with kinetic analysis: the same molecule is read multiple times (typically 10–15 passes for HiFi), producing both a high-accuracy consensus sequence (Q30+) and aggregated kinetic information that reveals modification status. The pb-cpg-tools pipeline calls 5mC at CpG sites genome-wide using a 400-bp context window. In 2025, a CUHK team introduced the HK2 model enabling 5hmC discrimination from PacBio HiFi kinetics.

Key advantages of PacBio HiFi methylation detection:

• High base accuracy — HiFi consensus reads achieve Q30+ (>99.9% accuracy), so SNV/indel calling and methylation calling are performed on the same high-quality reads
• Long-range phasing — HiPhase can phase variants and methylation across the full HiFi read length (15–25 kb), and with linked reads, across tens of megabases
• No reference bias — kinetic detection does not rely on a reference genome model, making it suitable for non-model organisms and highly divergent regions
• Prokaryotic methylation — SMRT kinetic data is the gold standard for detecting bacterial methyltransferases (m6A, m4C motifs); KinMethyl pipeline available

Haplotype-Resolved and Allele-Specific Methylation Analysis

The defining advantage of long-read methylation sequencing is the ability to assign every methylation call to a specific DNA molecule — and therefore to a specific haplotype or parental allele. Short-read bisulfite sequencing produces a population-average methylation level at each CpG (beta value = M/(M+U)), but cannot tell you whether 50% methylation means both alleles are half-methylated or one allele is fully methylated and the other is fully unmethylated. Biologically, these are entirely different regulatory states.

How Methylation Phasing Works

Long reads that span heterozygous variants (SNVs or indels) carry both the variant allele and the methylation status of every CpG on that same molecule. Phasing tools — MethPhaser (ONT), HiPhase (PacBio), or Whatshap — use these variant-phased reads to sort methylation calls into two haplotypes, producing a genome-wide haplotype-resolved methylome.

The biological insights enabled by haplotype-resolved methylation include:

• Imprinting validation — at known imprinted loci (e.g., IGF2/H19, SNRPN, KCNQ1), one parental allele is fully methylated while the other is fully unmethylated. Long-read phased methylation confirms this allele-specific pattern directly from sequence reads — no family-based genotyping required.
• Allele-specific methylation (ASM) at disease-associated loci — a regulatory variant on one allele may create or destroy a CpG site, alter transcription factor binding, or induce local hypermethylation on that allele only. Short-read methylation arrays average this signal; long reads resolve it per allele.
• X-inactivation skewing — X-chromosome inactivation silences one X chromosome in female cells via promoter CpG island methylation. Long-read phasing across the X chromosome quantifies skewing ratios and identifies escape genes (genes that resist silencing) with allele-level resolution.
• Cancer LOH + methylation — tumors frequently undergo loss of heterozygosity (LOH). If the retained allele carries a somatically hypermethylated tumor suppressor promoter, short-read methods may detect the methylation but misattribute it to both alleles. Long reads reveal the true allele-specific pattern.

Co-Methylation Haplotypes (mHap)

Beyond allele-level phasing, long reads reveal co-methylation patterns — whether adjacent CpGs on the same molecule tend to be methylated together or independently. Metrics such as single-molecule methylation entropy, Proportion of Discordant Reads (PDR), and Methylation Haplotype Load (MHL) quantify these patterns, which reflect chromatin state heterogeneity within a cell population. A 2023 Communications Biology study by Magi et al. demonstrated that stochastic co-methylation disruption at CpG-poor regions — not genetic mutation — drives chemotherapy resistance in relapsed AML, an insight inaccessible to short-read methylation analysis.

Service Options: Whole-Genome, Reduced Representation, and Targeted Methylation

Long-read methylation sequencing can be applied at three scales depending on your research question and sequencing budget.

Bioinformatics Analysis and Deliverables

Long-read methylation data analysis involves raw signal processing, base modification calling, alignment, methylation frequency quantification, DMR detection, and phasing. Our pipeline uses community-standard tools with fully documented parameters and versions.

Standard Deliverables:

Optional Advanced Analysis:

• Methylation haplotype (mHap) analysis — single-molecule co-methylation patterns, PDR, MHL, methylation entropy
• Multi-omics integration — joint analysis of methylation + SVs + CNVs from the same sequencing data
• Episignature classifier construction — diagnostic methylation signature models for specific disorders
• Imprinting / X-inactivation analysis — allele-specific methylation quantification at imprinted loci and X-chromosome-wide skewing ratios
• Longitudinal methylation tracking — analysis of methylation changes across time points or disease progression stages
• Epigenomic data analysis — see our dedicated Epigenomic Data Analysis service

Applications Across Research Areas

Frequently Asked Questions