BGPT: Paper Review: Padlock Probe–Based Targeted In Situ Sequencing: Overview of Methods and Applications

Explore by Goal

Quick Explanation Copied

Concise verdict: This 2023 Annual Review is a thorough, up-to-date, method-forward synthesis of padlock probe–based in situ sequencing (ISS, HybISS, direct-RNA approaches, SCRINSHOT), its computational pipelines, and key applications (neurobiology, cancer, infectious disease). The review clearly summarizes workflows, trade-offs (multiplexity vs optical crowding, RT vs direct-RNA efficiency), and computational bottlenecks, and rightly stresses the need for standardization and automated pipelines. Key primary-method references (Ke et al. 2013; Gyllborg et al. 2020) are cited and consistent with the review's claims about specificity, rolling-circle amplification (ϕ29), and HybISS improvements in SNR and panel scale

Long Explanation

Visual summary & critical analysis — Padlock probe–based targeted in situ sequencing (Magoulopoulou et al., 2023)

What this review does well

Clear, method-centric synthesis of padlock-probe ISS variants (original ISS, HybISS, direct-RNA HybISS, SCRINSHOT) and their biochemical steps (RT, padlock hybridization + ligation, ϕ29 RCA, decoding)
Balanced discussion of trade-offs: high specificity via Thermus ligase / ligation discrimination vs low RT/ligation efficiency; HybISS trade-off between unlimited theoretic multiplexity and optical crowding
Useful consolidation of computational ecosystem: ASHLAR (stitch/registration), ISTDECO/BarDensr/PoSTcode (decoding), Cellpose/StarDist/Baysor/SSAM for segmentation or segmentation-free cell typing, and downstream tools (Squidpy/Giotto/pciSeq) for domain and cell-type inference

Quick visual: paper meta-scores (from supplied raw metadata)

Concise technical critique (visual-first)

Core workflow fidelity

The biochemical description (padlock arms ~20nt, ~70nt padlock length, anchor + transcript barcode, Tth/Tth-like ligase discrimination, ϕ29 RCA producing RCPs) precisely matches primary method accounts (Ke 2013) and HybISS elaborations (Gyllborg 2020)

Strengths

Comprehensive cross-study examples (brain atlases, cancer OncoMaps, tuberculosis granulomas) with appropriate citations and use-cases
Balanced appraisal of bottlenecks (optical crowding, autofluorescence, segmentation errors) and computational remedies (deconvolution decoders, segmentation-free methods).

Limitations & blindspots

Limited quantitative benchmarking across tissue types: HybISS claims (higher SNR, larger panels) are supported by 2020 NAR data on brain, but cross-tissue, multi-lab benchmarks are sparse and not deeply analyzed in the review
Practical 'optical crowding' mitigation strategies are described qualitatively, but the review could better quantify expected RCP densities vs decoding error across modalities and imaging systems (objective NA, pixel size, z-sectioning) — important for labs planning deployment.
Standardization & reproducibility: the review recommends starfish/pipelines but does not prescribe minimal reporting standards (probe sequences, probe QC metrics, RCP density, raw image access), which hampers cross-lab replication (authors note this gap)

Detailed evidence-backed observations

Detection efficiency and direct-RNA approaches. The review correctly summarizes that direct-RNA padlock designs (chimeric padlocks, SCRINSHOT, BOLORAMIS) skip RT and can substantially increase per-molecule detection efficiency (~5x reported in applied studies), consistent with later direct-RNA reports (Lee et al., 2022) — but authors should have included a comparative table that normalizes detection efficiency versus sample type, fixation, and imaging optics to guide users selecting protocols
Multiplexing ceiling: concept vs practice. The review explains HybISS theoretically permits unlimited barcodes by sequential hybridization of L-probes/readout probes; in practice, optical crowding, probe cross-hybridization, and imaging throughput limit usable panel size. Primary HybISS reports mapped ~119–120 genes across whole sections but also required heavy data volumes and quenching steps for human tissue (lipofuscin) — so "unlimited" = coding capacity, not a practical immediate feature without advanced imaging and deconvolution.
Specificity: ligation discrimination vs hybridization trade-offs. ISS ligase-based discrimination (Thermus ligase) affords single-nucleotide specificity valuable for SNV/allele detection; HybISS sacrifices ligation-dependence for hybridization readout, increasing signal but relying on probe design and decoding accuracy to preserve specificity. The review correctly frames this trade-off and cites the biochemical fidelity of ligases (Tong/Barany lineage)
Computational decoding and segmentation: current best practices. The review lists recent decoders (ISTDECO, BarDensr, PoSTcode) and segmentation/segmentation-free approaches (Cellpose, StarDist, Baysor, SSAM, spage2vec). This is accurate and practical — yet an explicit performance comparison under matched image SNR / RCP density conditions (precision/recall vs optical crowding) is missing and would be the natural next step for the community to benchmark decoders on standardized image sets (the review recommends starfish pipelines but lacks a benchmarking dataset)

Practical guidance distilled from the review (actionable)

If you need SNV-level calls or allele-specific detection: prefer ligation-based padlock ISS (retain ligation and ligase discrimination), design arms carefully (3' end mismatch penalization), and include orthogonal validation (amplicon seq or targeted NGS) .
If you need larger panels (100+ targets) and high SNR in autofluorescent tissues: HybISS is recommended, but plan for autofluorescence quenching (TrueBlack) and substantial imaging/data storage; also design decoding schemes to minimize optical crowding (sparse codes, combinatorial schemes) .
For high detection efficiency of rare transcripts: consider direct-RNA padlock (SCRINSHOT/BOLORAMIS) approaches that skip RT, but validate across tissue fixation regimes; probe chemistry and accessibility matter greatly.

What would change my evaluation (falsification tests)

If independent multi-lab benchmarks showed that HybISS SNR/panel claims do not generalize beyond neural tissue (e.g., dense fibrotic or lipid-rich tissues) the practical value would be lower than stated.
If large-scale comparisons revealed systematic cell-type misassignments due to segmentation biases (not corrected by pciSeq or segmentationfree methods), then the review's optimistic tone on cell atlas readiness would require downgrading.

Confidence & recommendations

I assign the review high methodological fidelity and utility for practitioners (quality score ~9 in our metadata). It is best used as a field map and a methods-reader: labs should follow its recommendations but expect to perform pilot benchmarking for their tissue and imaging platform before scaling panels. The review correctly centers biochemical logic, computational needs, and practical constraints, and it appropriately calls for standardization and shared datasets to advance reproducibility .

Actions you can take next (buttons)

Author review quick-links (per your request):

Primary review citation:

Feedback:

Updated: March 13, 2026