Disclaimer: This blog provides an educational summary of a peer-reviewed research article published in Science Advances. It does not represent endorsement by the study authors or the journal. Several authors are affiliated with SeekGene, and the study was performed on SeekGene’s droplet platform; performance figures should be read in that context. Observations described are specific to this study and shall not be generalized as universal product claims.
For Research Use Only. Not for use in diagnostic procedures.
Figure 1 | scEM-seq labels each nucleus inside a single droplet, converts methylation enzymatically after the emulsion is broken, and recovers CpGs efficiently at droplet scale
DNA methylation is the quiet layer of cell identity. It does not alter the genome, yet it fixes which genes a cell can use, keeps lineages apart, and records the regulatory history of disease. Bulk whole-genome bisulfite sequencing reads that layer beautifully — but only on average. In a tumor or an immune infiltrate, the average blends cell types and states that may never coexist in one cell, and erases the rare population that matters most.
Single-cell methylome methods have struggled to resolve that blend at scale. Plate-based approaches such as scBS-seq and snmC-seq2 reach deep coverage but profile hundreds of cells; combinatorial indexing scales further through repeated rounds of pooling; droplet methods promise throughput, yet Drop-BS-seq’s bisulfite chemistry and repeated emulsification cost DNA, and scTAM-seq reads only about 650 CpGs per cell.
That is the gap a peer-reviewed study in Science Advances sets out to close. Published on 23 October 2026 and led by investigators at the Chinese PLA General Hospital, the Beijing Institute of Genomics of the Chinese Academy of Sciences, the China National Center for Bioinformation, Peking University First Hospital, and Nankai University, with SeekGene, the paper introduces scEM-seq — a droplet-based enzymatic methylome workflow built on a single emulsification. Applied to 3,812 cell-line methylomes, 9,551 PBMCs, and 9,364 prostate cancer cells, it shows that methylation alone can resolve immune lineages, expose patient-specific tumor architecture, and flag states linked to biochemical recurrence.
Methylation alone can resolve what the bulk average hides.
The scientific challenge
Reading methylation at scale without losing the cell
Two constraints shape every single-cell methylation method. Conversion is destructive: bisulfite fragments DNA, and each extra purification or emulsion step removes molecules a single cell cannot spare. And the barcode must survive conversion. scEM-seq addresses both by design — the cell barcode and UMI are synthesized without cytosine, the read 1 primer is methylated, and barcoding is completed by ligation before the droplets are broken. Conversion then happens once, in bulk, after every molecule already carries its cell’s identity.
The second challenge is analytical. Single-cell methylomes are sparse and high-dimensional, and solid tumors have no methylation reference atlas to annotate against. A method that produces data without a framework for doublet removal, clustering, annotation, and copy-number inference leaves much of its value unused. This study pairs the chemistry with an end-to-end pipeline.
Why it matters
Why SeekOne™ DD Single-Cell DNA Methylation matters
01
One droplet, one barcode, one cell
Nuclei are fixed, nucleosome-depleted, and tagmented, then co-partitioned with barcoded gel beads. A bridge oligonucleotide links each bead’s barcode to the Tn5 insertion site, and ligation proceeds inside the droplet — so all downstream steps can run in one pooled tube.
02
Enzymatic conversion instead of bisulfite
Post-emulsion C-to-U conversion uses TET2 and APOBEC, avoiding bisulfite fragmentation. The workflow runs in about 16 hours, supports up to eight samples in parallel, and profiles more than 10,000 cells per experiment. At matched output it recovers 11.3% more CpGs than Drop-BS-seq, for an estimated reagent cost of about $0.024 per cell.
03
Sparse per cell, complete across cells
Cell-line methylomes averaged 85,348 CpGs per cell and PBMC methylomes 69,569. Aggregating 9,364 prostate cancer cells yielded 28.6 million CpG observations covering more than 95% of CpGs in hg38 — enough to define methylation domains, differentially methylated regions (DMRs), and copy-number profiles from the same data.
04
A framework, not just a library
The pipeline includes MethSCAn-based imputation, a doublet classifier combining read-depth decay (Di) with the nonbinary methylation rate (NBMR), scMethID annotation for tissues without a reference, fluctuating-CpG (fCpG) phylogenetics, and ichorCNA copy-number inference. Data and scripts are deposited in GSA, Zenodo, and GitHub.
This study profiles the methylation arm alone. SeekOne™ DD Single-Cell DNA Methylation can be paired with single-cell transcriptomics on the same droplet platform, so methylation and expression can be interpreted side by side.
Validation
From cultured cells to clinical tissue: validation at three levels
Validation began in cell lines. 22Rv1 and PC3 prostate cancer cells grown in 2D and 3D, alone and in mixtures, yielded 3,812 methylomes with 64% unique alignment. Clustering separated cell identity rather than culture condition; within 22Rv1, one cluster sat epigenetically between the 2D and 3D states, and CTCF-binding sites — not gene bodies or CpG islands — carried the clearest condition-specific signal. A human–mouse mixing experiment estimated an 8.7% multiplet rate.
In PBMCs, 9,551 cells resolved into six clusters that matched BLUEPRINT hypomethylated regions and correlated with sciMET-cap profiles at B cell DMRs. In two prostate tumors (Gleason scores 7 and 8), 9,364 cells passed quality control at about 69% mapping, and the aggregated methylome reproduced bulk WGBS profiles from the Chinese Prostate Cancer Genome and Epigenome Atlas.
Key findings
Four findings from the study
FINDING 1 immune lineages
Immune lineages resolve from methylation alone
The PBMC result is a useful stress test because the answer is known. Each lineage lost methylation at its marker promoters and BLUEPRINT-defined hypomethylated regions, and cell type-specific hypo-DMRs were enriched for B cell activation, myeloid leukocyte differentiation, and T cell differentiation. B cells and NKT cells carried a higher proportion of hypo-DMRs and showed globally reduced methylation across gene bodies, CpG islands, and CTCF sites — lineage identity is written across the methylome, not at a single locus.
Figure 2 | Methylation alone separates monocytes, B cells, T cells, NK cells, and NKT cells, and reproduces BLUEPRINT lineage signatures
Initial clustering of the prostate tumors produced 13 clusters, which separated into six tumor clusters and seven nontumor clusters; subclustering yielded TC0–TC6 and NC0–NC14. Tumor clusters showed globally reduced methylation and extensive hypomethylated blocks. With no prostate cancer methylation atlas available, the authors built scMethID, which combines single-cell RNA-seq marker genes with matched bulk WGBS and RNA-seq to assign lineage identity probabilistically; assignments were corroborated against a pan-tissue methylation atlas. The contrast was stark: malignant epithelial cells clustered by patient, while nonepithelial cells intermingled across patients.
Figure 3 | Tumor and nontumor clusters separate by methylation, and malignant epithelial cells retain patient-specific architecture
FINDING 3 copy number
Methylation and copy number couple — patient by patient
Because methylation reads map back to the genome, the same libraries yield copy-number profiles. Copy-number variants (CNVs) were highly similar within a patient and divergent between patients, and TC0 and TC1 carried relatively higher burdens of structural alterations. The methylation–CNV relationship, however, was not universal: positive across TC0, TC2, and TC6 in PCa1 and absent in PCa2. The overlap between CNVs and methylation domains also varied by cluster. Epigenome and genome are not decoupled; their coupling is itself a patient-specific variable that bulk data cannot resolve.
FINDING 4 recurrence
Two tumor states mark aggressive disease
The clinical signal came from cluster-specific hypomethylated regions. Scoring TCGA-PRAD patients by methylation at TC0 and TC1 hypo-DMRs, the authors found that the lowest-methylation group had markedly higher biochemical recurrence (21.7% versus 4.7% for TC0 and 20.9% versus 3.3% for TC1), was enriched for Gleason scores 8–10, and showed worse progression-free survival (log-rank P = 0.0023 and P = 0.00012). The regions were enriched for carboxylic acid, amide, and fatty acid metabolic processes. An fCpG-based tree placed TC0 in PCa1 and TC1 in PCa2 on terminal, distantly branched positions, consistent with late-emerging malignant clones.
Figure 4 | TC0- and TC1-specific hypomethylation tracks with biochemical recurrence, higher Gleason grade, and worse progression-free survival in TCGA-PRAD
Outlook
From a single methylome to a scalable framework for epigenetic heterogeneity
scEM-seq does not try to beat deep plate-based methods at per-cell coverage; the authors note that scBS-seq and snmC-seq2 still recover more CpGs per cell. What it changes is scale. One emulsification instead of several, and enzymatic conversion instead of bisulfite, remove two major sources of loss; C-free barcodes keep cell identity intact through conversion. The accompanying framework then turns sparse per-cell data into cell types, domains, phylogenies, and copy-number states.
The limits are equally clear. Coverage remains sparse, low-frequency events may need deeper sequencing, and extensive CNVs can distort the doublet metrics that rely on read-depth decay and methylation bimodality. The current version profiles methylation alone. And two tumors cannot establish that TC0/TC1 signatures generalize; that requires larger cohorts and prospective validation.
What the study does establish is a template: a droplet-scale methylome readout that resolves what bulk sequencing averages away, and a computational path from barcode to biology. For labs asking which cells differ, how they evolved, and whether it matters clinically, this is a workflow worth watching.
Interested in SeekOne™ DD Single-Cell DNA Methylation?
Li G, Yang Y, Li Z, et al. High-throughput droplet-based enzymatic DNA methylome profiling enables single-cell resolution of epigenetic heterogeneity. Sci Adv. 2026;12(43):eaef2844. doi:10.1126/sciadv.aef2844