BGPT: Paper Review: Optimizing genome editing strategy by primer-extension-mediated sequencing

Explore by Goal

Quick Explanation Copied

PEM-seq (Yin et al., 2019) provides a single, barcode-aware sequencing workflow intended to measure on-target editing outcomes (indels + germline states) together with genome-wide DSB outcomes (translocations, and associated large deletions/inversions), then uses that unified readout to compare Cas9 variants (eCas9/FeCas9), nickases, and the anti-CRISPR AcrIIA4 timing/inhibition trade-offs.

Main strengths: (i) conceptual unification of efficiency + “structural safety” readouts, (ii) molecular barcoding to mitigate PCR duplicate inflation, (iii) availability of pipeline code and deposited PEM-seq data.

Long Explanation

Paper Review (Scientific, Skeptical, Evidence-based)

“Optimizing genome editing strategy by primer-extension-mediated sequencing” — Yin et al., Cell Discovery (2019)

DOI: 10.1038/s41421-019-0088-8 • Open access; GEO dataset and GitHub pipeline are provided.

Figure A — PEM-seq decomposes on-target outcomes (RAG1A example)

PEM-seq reports that (in HEK293T treated with Cas9:RAG1A) indels = 35.7%, translocations = 2.7%, summing to an editing efficiency ≈ 38.4%.

Figure B — Indel micro-spectrum near the RAG1A break site

The paper reports (for RAG1A) 11.3% small insertions and 24.0% tiny deletions among total editing events, within ±20 bp around the target.

Figure C — Off-target hotspots: PEM-seq vs LAM-HTGTS recovery

In the RAG1A setting, PEM-seq identified 53 total off-target sites (24 new vs LAM-HTGTS and 4 weak sites that LAM-HTGTS had but PEM-seq did not). It also reports that 8 selected off-target sites were checked by in vitro Cas9 digestion (including 4 new sites and 4 overlapped ones).

Figure D — Efficiency–specificity trade-off across Cas9 forms (RAG1A)

The paper reports eCas9 yielded 7 detected off-target hotspots without obvious editing-efficiency loss, while FeCas9 yielded only 3 detected hotspots and retained comparable on-target editing efficiency. For Cas9 nickases (RAG1A/G), they report editing efficiency at 19.3% with 5 off-target hotspots detected.

Figure E — AcrIIA4 inhibition: on-target drops more than off-target (fold change)

The paper reports: at a 3:1 Cas9:AcrIIA4 mass ratio, Cas9 editing efficiency decreases ~11-fold, while off-target translocation junctions decrease only ~1.7–4.6-fold.

Mechanistic reading of PEM-seq (what it claims it measures)

Core design: PEM-seq uses primer extension to create a single-copy “surrogate fragment” from a genome region near a bait-induced DSB (within ~200 bp), then isolates these surrogate fragments, adds adapters carrying a 14-bp random molecular barcode (RMB), and sequences (Illumina 2×150). Outcome categories: The analysis pipeline decomposes sequencing reads into germline (uncut or perfectly repaired), indels (errored rejoining at break-site/±5bp neighborhood), and translocations (genome-wide DSB outcomes captured relative to bait). Why barcodes matter: The paper argues PEM-seq reduces PCR amplification bias (relative to methods that include linear amplification steps). It attributes this to primer extension producing one copy of the template and to RMB clustering removing PCR duplicates at an editing distance threshold.

Scientific context: PEM-seq builds on translocation-capture DSB mapping approaches such as LAM-HTGTS, where “bait” DSB sites enable genome-wide translocation detection. LAM-HTGTS is itself described as using linear amplification that can complicate PCR duplicate discrimination.

What the paper demonstrates (and what it does not)

1) Off-target hotspot discovery sensitivity

In HEK293T cells targeting RAG1A, PEM-seq identifies 53 total off-target sites (24 new relative to LAM-HTGTS; 4 weak LAM-HTGTS sites not recovered).
Eight selected off-target sites were validated via in vitro Cas9 digestion, with cleavage efficiencies varying from 18–60%.
Critical caveat: in vitro digestion validates potential cleavage of candidate loci, but it does not by itself prove the in vivo cutting frequency at each locus matches PEM-seq’s translocation/junction-based inference.

2) Editing efficiency quantification

PEM-seq quantifies germline vs indels vs translocations and estimates editing efficiency as a function of indel + translocation proportions normalized to controls (SuperQ).
The paper cross-checks indel fractions at RAG1A using RFLP, T7EI, and TIDE, reporting broadly comparable indel percentages (27–40%).

3) Structural consequences: deletions/inversions and resection orientation biases

The paper reports that deletions/inversions concentrate near the target/primer region and that junction distributions show an end resection-like pattern, with resection frequency in the 5–50 kb region around ~0.05% of total editing events.

4) Strategy optimization results (variants, nickases, inhibitors)

Nickases: RAG1A/G nickases show lower editing efficiency (19.3%) and reduced off-target hotspot count (5) compared to Cas9:RAG1A (38.4% and more hotspots).
High-fidelity variants: eCas9 (7 hotspots) and FeCas9 (3 hotspots) reduce off-targets with minimal/no obvious editing-efficiency loss at RAG1A; FeCas9 is reported to outperform eCas9 at additional loci (e.g., EMX1 and a region near c-MYC).
AcrIIA4 inhibitor: AcrIIA4 suppresses Cas9 activity more on-target than off-target (e.g., ~11-fold vs ~1.7–4.6-fold at the reported ratio), implying the timing/availability of active nuclease can reshape the efficiency–specificity trade-off.

Limitations, blind spots, and skeptical interpretation

Bait-dependence: PEM-seq “relies heavily on bait DSBs” to capture translocations; thus, it may not faithfully represent loci whose translocation capture probability differs due to distance, chromatin architecture, or DSB kinetics.
Off-target indel quantification constraints: the manuscript explicitly notes that PEM-seq cannot quantify frequencies of identified off-target sites’ indels when the bait is an on-target site, because indel detection is tied to bait sites.
Rare event sampling: translocation is rare; they state PEM-seq needs at least tens of thousands of cells for off-target site identification.
Interpretation of “off-target intensity” vs cutting frequency: PEM-seq uses junction enrichment and hotspot calling; that is a proxy for DSB outcomes, but proxy ≠ absolute cleavage probability. This matters when comparing inhibitors/variants that alter kinetics rather than final cleavage spectrum.

Assays vs in vivo truth: several validations are performed in vitro (for selected sites) and cross-comparisons are made against assays like TIDE/RFLP/T7EI, but each of those has its own measurement biases (e.g., PCR bias, indel-size sensitivity limits).

Conflict of interest & reporting transparency (what we can verify)

The paper states the authors are applying for a patent for PEM-seq and FeCas9, and one author (J.H.) is a Bayer investigator.
Code availability is provided (SuperQ on GitHub) and PEM-seq data are deposited in GEO (GSE116231).

Skeptical note: patents and industry ties don’t automatically invalidate results, but they increase the importance of independent replication and careful methodological scrutiny (especially around the mapping/peak-calling and normalization choices).

Jump to author-specific BGPT reviews

Feedback:

Updated: April 01, 2026