Review Papers Against Their Own Results

One paragraph critique

Dosa introduces an elegant in vivo covalent RNA tagging strategy that links protein variants to short RNA barcodes via TrmA(E358Q), enabling direct biochemical selection read out by sequencing; the preprint demonstrates validation across binding, protease cleavage, and solubility assays with strong replicate correlations and external concordance, but important caveats remain about nonnative fusion context, barcode encoding separate from mRNA, and limited replicates for some assays (Dosa preprint describing method and assays)

Detailed paper review and critique

Executive summary

Dosa is a method that covalently links proteins to short RNA barcodes in vivo by fusing proteins of interest to TrmA(E358Q) and coexpressing a synthetic RNA containing the TrmA stem-loop plus a 10nt barcode; after lysis and biochemical selection, linked barcodes are recovered by reverse transcription and Illumina sequencing to compute enrichment scores that report biochemical phenotypes (binding, protease cleavage, solubility). The preprint provides multiple validations with high replicate correlations and concordance with orthogonal methods but also documents clear limitations related to nonnative fusion context and separate barcode transcripts (Dosa preprint describing method and assays)

What they did (concise technical)

• Engineered a TrmA(E358Q) catalytic mutant fused to proteins of interest with an N terminal 6X His tag to permit NiNTA input capture and optional epitope tags (eg FLAG) for output selection (Method plasmid architecture and barcode transcription)
• Produced prebarcoded plasmid libraries (≈5M barcoded plasmids, bottlenecked to give ~10–20 barcodes per variant) and associated variants by two-step cloning and PacBio subassembly to map barcodes to variant sequences (Library cloning and barcode mapping)
• Assays: FLAG epitope binding (152 variants, ~1900 barcodes, r=0.98 replicates, r=0.84 vs CasPlay), enteropeptidase cleavage (same 152 variants, r=0.96 replicates), HRV3C cleavage (178 variants, r≈0.99 replicates; r=0.87 vs YESS), and solubility profiling including Aβ1-42 (794 variants, replicate correlation r≈0.81) with data normalized as log2 ratios of output/input (Assays, library sizes, replicate correlations)

Strengths

1. Direct covalent linkage of barcode to protein in vivo allows downstream biochemical conditions incompatible with noncovalent display (eg 8M urea, heat) — the paper shows retention of barcode after solubilization in 8M urea and NiNTA enrichment under denaturing conditions (Barcode stability under denaturation)
2. Validated across multiple, biologically distinct assays with strong internal reproducibility and good concordance to orthogonal datasets (CasPlay, YESS), increasing external validity (Reproducibility and orthogonal agreement)
3. Demonstrated compatibility with mammalian cells (HEK293T experiments show enrichment on anti-FLAG beads), suggesting applicability beyond bacteria (Mammalian cell compatibility)
4. Robust library construction using PacBio long reads to map barcodes to sequences reduces misassignment risk and supports multiple barcodes per variant to quantify measurement variability (PacBio barcode mapping strategy)

Limitations, biases, and open questions

• Non native fusion context: proteins are hybrids containing TrmA(E358Q) and linked RNA; the authors acknowledge potential effects on folding or solubility and cite prior reporter studies but do not systematically quantify fusion-induced perturbations across diverse proteins — this could bias solubility and activity measurements (Non native fusion caveat)
• Barcode encoded separately from mRNA: this prevents recovery of the coding mRNA and makes iterative selection cycles (rounds of evolution/selection) nontrivial; authors propose possible fixes (engineering TrmA variants, encoding barcode into 3UTR) but have not implemented them (Separate barcode transcript limitation)
• Potential barcode crosstalk or linkage errors during lysis: they measure small amounts of noncognate linking and mitigate by adding excess stem-loop during lysis, but residual risk remains and depends on lysis conditions and expression levels — critical for very large libraries and low abundance variants (Barcode leakage mitigation and residual risk)
• Limited replicates for some assays and dependence on sequencing depth: many experiments report n=2 biological replicates; while correlations are strong, additional replicates and orthogonal biochemical validation across more proteins would strengthen claims about generalizability (Replicate counts and sequencing thresholds)
• Thermal stability and activity window of TrmA(E358Q): authors note TrmA is 42 kDa and smaller than dCas9 but suggest engineering thermostable TrmA variants to permit higher assay temperatures; current constraints may limit some biochemical conditions (TrmA thermostability and size considerations)

How convincing are their validations?

The validations are convincing for the specific systems tested: FLAG epitope binding recapitulates structural determinants and agrees with CasPlay (r=0.84) and shows high replicate reproducibility (r=0.98); protease substrate specificity (enteropeptidase and HRV3C) correlates strongly with biochemical datasets (r up to 0.96 and 0.87 vs YESS) and internal replicates; solubility screens for A beta variants match expected sequence pattern effects (hydrophobic increases aggregation; proline increases solubility) and structural interpretation consistent with known A beta filament models. These are solid validations for method feasibility in these assay classes but are limited to peptide epitopes, short protease substrates, and aggregation-prone peptides; performance on full-length, multidomain mammalian proteins with complex PTMs remains to be demonstrated at scale (Validation strengths across assays)

Reproducibility and data availability

Methods are described in detail including cloning, barcode insertion, PacBio subassembly, RT-PCR, sequencing, and data filtering (min counts thresholds). Raw barcode sequencing reads deposited under BioProject PRJNA1329341 improves reproducibility and reanalysis capability (Data deposition and methods detail)

Practical recommendations to users and adopters

• When applying Dosa to a new protein, run control comparisons to the untagged or minimally tagged protein where possible to quantify fusion-induced perturbations (e.g., expression level, activity, solubility by orthogonal assays).
• Use multiple barcodes per variant (authors used median ~10) and set read-count thresholds to limit sampling noise; verify barcode mapping with long reads.
• For solubility/aggregation workflows, validate selected hits by purified protein or orthogonal cell-based assays to confirm that fusion elements are not driving the phenotype.
• Consider engineering alternative TrmA stem-loops or TrmA variants (authors mention E358Q engineering and orthogonal stem-loops) to enable multiplexed orthogonal tagging or to reduce potential crosslinking in pooled contexts.

Potential extensions and research directions

1. Encode barcode into the 3UTR of the same mRNA as the protein to enable recovery of protein coding sequence and iterative selection rounds (authors propose this).
2. Engineer thermostable or smaller TrmA variants to expand assay temperature range and reduce fusion footprint.
3. Apply Dosa to proteome-scale libraries or targeted domain libraries to map biochemical landscapes of clinically relevant proteins and feed large datasets into machine learning models for prediction of solubility/stability/activity.
4. Combine Dosa with subcellular fractionation or proximity labelling to capture context-specific activities.

Conclusion and confidence

Dosa is a carefully executed demonstration of a covalent in vivo protein barcoding strategy that meaningfully extends the toolbox for massively parallel biochemical assays. The validations are internally consistent and concordant with independent datasets for the tested assays, and the method is likely to be broadly useful provided users control for fusion artifacts and sequencing/representation biases. Confidence in the method for peptide-level assays is high; confidence for general proteome-wide application is moderate until further demonstrations on larger, diverse sets of full-length proteins in mammalian contexts are available (Authors summary and limitations)

Author Reviews