BGPT: Paper Review: Joint single-cell multiomic analysis in Wnt3a induced asymmetric stem cell division

Fuel Your Discoveries

Quick Answer Copied

What this paper adds: It introduces SET-seq/scSET-seq to jointly profile histone marks (H3K27me3, H3K4me3) and the transcriptome in the same single cell, then uses it to show that during Wnt3a-induced asymmetric mESC division, H3K27me3 (not H3K4me3) tracks gene-expression changes and aligns with epigenomic clusters; it also implicates Aebp2 in PRC2-linked control of daughter-cell fate balance.

Long Answer

Paper under review

“Joint single-cell multiomic analysis in Wnt3a induced asymmetric stem cell division” (doi:10.1038/s41467-021-26203-0)

Core claim: During Wnt3a-driven asymmetric mESC division, H3K27me3 correlates with transcriptomic changes and epigenomic cluster structure, while H3K4me3 shows weaker/“disordered” correspondence with expression dynamics; Aebp2 (PRC2 regulator) modulation shifts the asymmetric Nanog-expression outcome ratio.

Figure-data provenance (numbers used above) are taken directly from the manuscript text: passing-QC single-cell counts (335 vs 210), cluster proportions (24.4/22.2/53.4), selected gene detection examples (avg 4,665 per cell; 17,097/9,371 genes for 300 ng/30 pg; 4,206 Smart-seq2 reference), and the reported module correlation fraction (21.5% for H3K27me3 vs 10.7% for H3K4me3).

1) Method development (SET-seq / scSET-seq): what is new, and what is constrained?

Design intent: The method attempts to “co-profile” cytoplasmic RNA expression and nuclear chromatin marks from the same single cell by combining Tn5-based directed tagmentation of mRNA/cDNA hybrids with CUT&Tag-style antibody-guided tagmentation on nuclei. Optimization evidence: The authors report empirically tuning tagmentation temperature/time, Tn5 concentration, and starting RNA amount (down to 30 pg total RNA), with correlations to conventional ligation-based RNA-seq reported as strong across tested conditions and cell-count scaling. Concrete throughput: scSET-seq is positioned as suited to hundreds of cells; their asymmetric-division experiment produced 1,504 genome-wide profiles before QC, with 335 and 210 passing for H3K27me3 and H3K4me3 subsets.

2) Biological question: Wnt3a-directed asymmetric division—what do they actually show?

Spatiotemporal setup: Proximal vs distal daughters are assigned by manual inspection relative to Wnt3a-coated beads because conventional FACS sorting was not feasible by marker fluorescence; proximal/distance daughters are then indexed for scSET-seq. Three epigenomic/transcriptional clusters: They find three main clusters (Mix/Proxi/Dista) and show that when clustering is performed using H3K27me3 signatures, it largely recapitulates expression-defined clusters, whereas clustering using H3K4me3 behaves differently (more strongly projecting onto the Mix transcriptomic cluster). Pseudotime dynamics: The authors reconstruct trajectories via pseudotime ordering and describe Proxi→Dista progression, with a branch corresponding to the Mix cluster occurring as a distinct lineage stage rather than a midpoint between Proxi and Dista in gene-marker behavior. Main mechanistic correlation claim: Using co-embedding and WGCNA module correlations, they report stronger epigenome↔transcriptome module coupling for H3K27me3 than H3K4me3, including a higher fraction of significantly correlated modules (21.5% vs 10.7%). Functional link via Aebp2 perturbation: They identify Aebp2 expression as a Mix cluster marker gene and perform CRISPR knockouts (Ezh2 and Aebp2). They report Aebp2 knockout shifts daughter-cell outcome ratios for Nanog asymmetric expression (and they discuss possible compensatory PRC2 recruitment/state changes due to altered H3K27me3 patterns).

3) Skeptical critique: where results could be sensitive or misleading

Manual proximal/distal assignment: Because daughter-cell assignment relies on manual picking relative to bead contact rather than an automated molecular sorter, labeling errors or subtle selection bias could distort cluster composition (especially the “Mix” state). The authors explicitly note technical difficulty for sorting, but this still leaves a potential bias pathway.
Noise + peak calling thresholding: Single-cell histone-mark profiling yields sparse signal; peak calling overlaps with ENCODE at ~60% in their discussion, implying that stringent thresholds may omit true biology or systematically under-detect one mark more than the other—potentially affecting the apparent “H3K27me3 vs H3K4me3” asymmetry.
Correlation ≠ causation: The central mechanistic statement is correlation of H3K27me3 with gene-expression changes. WGCNA and module correlations support co-variation but do not, by themselves, prove that H3K27me3 changes are the cause rather than a consequence or a parallel readout of cell-cycle timing, chromatin accessibility changes, or Wnt-mediated signaling state. Their Aebp2 perturbation is a stronger link, but still depends on assumptions about how Aebp2 knockout alters PRC2 local recruitment/turnover in this asymmetric-division context.
Trajectory/branch interpretability: Pseudotime reconstruction in branching differentiation can be sensitive to marker selection, dimensional reduction choices, batch integration, and the definition of “differential genes used for ordering.” If cell-state transitions are not monotonic or if “Mix” corresponds to a combination of states, the branch interpretation may be partially underdetermined.

4) Reproducibility & data access

The authors state raw and processed sequencing data are deposited in GEO under GSE168637, and provide external reference datasets from ENCODE (e.g., ENCSR059MBO, ENCSR326ULS, ENCSR000CGO, ENCSR095IPH) plus Aebp2 ChIP-seq dataset GSE83082. They also provide code availability via a GitHub repository for scSET-seq.

5) What would most disprove or change the paper’s conclusions?

Show that H3K27me3↔expression coupling (and module-correlation dominance) disappears under alternative peak-calling thresholds / denoising settings, or after using an orthogonal mapping strategy for histone mark assignment to genes. The paper itself notes stringent peak-calling may reduce overlap with ENCODE.
Demonstrate that Aebp2 KO shifts Nanog asymmetry via a PRC2/H3K27me3 mechanism that can be independently validated (e.g., temporally acute perturbation or rescue), rather than via compensatory changes that could indirectly alter division outcomes. The paper discusses possible compensatory PRC2 recruitment patterns when Aebp2 is depleted.

Bespoke author reviews (jump to each author)

Feedback:

Updated: April 26, 2026