Paper Review — verify claims with raw data

Brief critical summary

circVDJ-seq is presented as a simple, low‑cost method to recover full TCR VDJ sequences from 3'-barcoded single‑cell and spatial cDNA (Visium, Multiome, droplet 3' libraries), achieving high mapping and barcode validity and strong concordance with 5' immune profiling and long‑read sequencing (example metrics: 93% valid CBC, 83% VDJ mapping; correlations R=0.84–0.93) circVDJ-seq preprint

In-depth review and critique of circVDJ-seq for T cell clonotype detection in single-cell and spatial multi-omics

1) What the paper does (concise)

The authors present circVDJ-seq: a library manipulation and targeted PCR approach that circularizes 3'-barcoded cDNA (via Gibson assembly), amplifies across conserved constant regions to capture full VDJ sequences, and recovers TCR alpha and beta paired clonotypes from 3'-directed single-cell and spatial workflows (3'GEX, Multiome nuclei, Visium) with post-processing using CellRanger (modified barcodes) and Dandelion for contig annotation circVDJ-seq method description

2) Key empirical results (data-first)

• Mapping and barcode recovery: ~93% of circVDJ reads contained a valid cell barcode and ~83% mapped to a VDJ gene (from 3'GEX-derived PBMC libraries) Mapping stats
• Contig recovery: TCRa contigs in 73% of cells, TCRb contigs in 84%, yielding paired clonotypes in ~57% of cells in optimized 3' libraries (triplicates) Contig recovery metrics
• Cross-method agreement: circVDJ vs 5'IPv2: high correlation of clonotype frequencies (R≈0.93 for 3'circVDJ vs 5'IPv2; R≈0.78 for MO circVDJ vs 5'IPv2) and strong agreement with long-read MAS-ISO-seq (R=0.84, p<2.2e-16) Method concordance
• Cross-recovery numbers: circVDJ recovered nearly all top clones (19/20 top clones shared with 5'IPv2 in one dataset); overall clonotype recovery was 87% vs 97% for 5'IPv2 in PBMC comparisons (authors' reported figures) Recovery fractions
• Spatial and tissue: circVDJ applied to Visium cDNA from autopsy lung and lymph nodes recovered clonotypes with spatial addresses (after Dandelion processing) and revealed biologically plausible patterns (e.g., clonotypes enriched in T cell zones; differing clonality between COVID/non-COVID LNs) Spatial application
• Clinical example: neuroblastoma spatial cohorts showed striking differences: HR-NB had 116 identified T cell clones vs LR-NB 16 clones and HR clones were often excluded from tumor area, co-occurring with M2 macrophage signatures — interpretable biology consistent with immune exclusion phenotypes Neuroblastoma spatial results

3) Strengths

1. Platform-agnostic utility: converts widely used 3'-barcoded cDNA (single cell, nuclei, spatial Visium, Multiome) into TCR‑profiling compatible templates, expanding use of archived libraries and lowering barrier to entry compared to multiplex V gene PCR panels or long-read only solutions Platform compatibility
2. Cost and primer simplicity: fewer gene‑specific primers needed and no specialized long‑read sequencer required for most uses, making large scale retrospective analysis feasible Cost argument
3. Validation breadth: comparisons to 5'Immune profiling workflows and to MAS-ISO long‑read sequencing strengthen confidence in recovered clonotypes and quantitative concordance (R values reported) Validation evidence

4) Limitations, caveats, and blindspots (critical)

• Reduced alpha chain recovery in nuclei/Multiome: authors report lower TCRa contig recovery from nuclei (MO), reducing paired clonotype fraction (paired clonotypes fell to ~36% in MO assay), which limits ability to infer fine TCR pairing and alpha diversity from nuclei-based workflows Nuclei recovery limitation
• Imputation of missing chains can bias pairing: authors impute missing chains by assigning the most frequent partner in dataset; while pragmatic, imputation can artificially inflate apparent pairing consistency and conceal real many‑to‑many pairing biology (authors note <5% promiscuous pairing observed) Imputation note
• Index hopping and cross-contamination: authors describe bespoke filtering to remove spurious CBC+CDR3 contigs likely from index hopping; this raises the importance of rigorous experimental controls and reporting of demultiplexing issues for reproducibility Index hopping handling
• Dependence on RNA integrity and starting material: authors explicitly state degraded mRNA limits recovery (especially in autopsy material); spatial Visium spot-size (50 um) contains multiple cells and thus spot-level clonotype assignment has limited single-cell resolution unless paired with higher-resolution platforms Material quality caveat
• Limited donor/sample diversity: presented applications include PBMCs from a healthy donor, several autopsy samples and two neuroblastoma patients — promising but still a limited cohort; broader benchmarking across more donors, disease states, and sample preprocessing conditions would strengthen generalizability (authors note data availability pending) Cohort limits and availability

5) Reproducibility and computational pipeline critique

Authors provide an explicit wet lab protocol (Gibson circularization, nested PCR of constant regions, library prep) and describe computational steps: modified CellRanger whitelist substitution to accept non‑5' barcodes (3'GEX, MO, Visium), Dandelion for contig annotation, and downstream use of Squidpy for spatial co‑occurrence analyses. These are clear, but full reproducibility requires deposited code, exact primer sequences, PCR cycle conditions, and raw fastq data and sample barcodes: the manuscript states data will be deposited but gives no accession numbers at present — an important reproducibility shortcoming to be remedied Pipeline description and data availability

6) How convincing is the validation?

Validation uses three complementary approaches: comparison vs 5'Immune profiling (same donor PBMCs), technical triplicates, and comparison vs long‑read MAS-ISO-seq. Correlation coefficients are high (R from 0.78 to 0.93, and 0.84 vs long read). These quantitative agreements support the method's fidelity for clone frequency estimation and identification of abundant clones; however, rarer clonotypes and fine pairing (especially TCRa) can be under-recovered in nuclei and spatial spots, so claims should be tempered for low-frequency clones and single‑nucleus applications (authors acknowledge these limitations) Validation correlations

7) Recommendations for users and developers

• Use circVDJ for retrospective recovery from archived 3' libraries and for spatial TCR mapping when high-depth 5' data are unavailable.
• For experiments where accurate TCR pairing (especially alpha chain) is critical (e.g., TCR engineering, neoantigen specificity), prefer 5' targeted VDJ or orthogonal long‑read confirmation; circVDJ is strong for abundant clones and population-level clonality but less robust for low‑abundance/paired alpha recovery in nuclei-derived libraries 5
• Report and deposit: publish exact primer sequences, Gibson design details, PCR thermocycling conditions, and raw fastq + processed VDJ contigs (CellRanger/Dandelion outputs plus scripts) to maximize reproducibility.
• When applying to Visium/spot-level data, pair with higher resolution spatial assays (Visium HD, STOMICS, or single-cell resolution spatial platforms) or computational deconvolution (DOT, soFusion, or FUSION-like tools) to reduce ambiguity from multi-cell spots DOT integration reference

8) Specific experimental and computational improvements I would test

1. Systematic benchmarking across a larger donor panel (diverse ages, disease states) with matched 5' and 3' libraries to quantify sensitivity/specificity per-clone frequency spectrum.
2. Spike‑in synthetic TCR standards of defined frequency to estimate limit of detection and false positive rate (control for index hopping) — include UMIs targeting VDJ amplicons to estimate PCR duplicates.
3. Publish a reproducible nf-core/Snakemake pipeline wrapping CellRanger modifications and Dandelion steps; include exact whitelist files and commands used to swap barcode files.

9) How the paper could be falsified (clear tests)

The authors themselves list falsification criteria: if independent groups cannot reconstruct TCRa/TCRb contigs from 3'-barcoded libraries using the circVDJ protocol, or if clonotype frequencies show poor concordance with orthogonal 5' or long‑read methods across multiple datasets and tissues, then circVDJ claims would be undermined; similarly, inability to map clones spatially in Visium with concordant TCR gene expression and clone locations would challenge spatial claims Falsification criteria

10) Final balanced take

circVDJ-seq is a pragmatic, well-validated approach that meaningfully expands the ability to extract TCR repertoire information from commonly used 3'-barcoded single‑cell and spatial libraries. The method's strengths are accessibility, compatibility with archived cDNA, and strong quantitative concordance for abundant clones. The main limitations are lower alpha chain recovery in nuclei-derived libraries, reliance on imputation when chains are missing, and the need for careful demultiplexing controls. With broader benchmarking, public data deposition, and open pipelines, circVDJ-seq could become a standard tool for retrospective and spatial TCR repertoire studies Concluding claim

Need deeper computational reanalysis or reproducible code? Run an AI biology agent to iteratively reprocess raw sequencing data, run CellRanger/Dandelion pipelines, compute clonotype overlap metrics, and produce publication-quality figures.