Decision letter

Decision Letter

VERDICT: major

Summary of Evaluation

This manuscript proposes an interpretive framework for combining whole-cell patch-clamp electrophysiology with mass spectrometry-based single-cell proteomics (patch-SCP) in acute rat brain slices, and applies it to twelve mPFC Layer 2/3 pyramidal neurons collected under an explicitly indiscriminate ("shotgun") strategy. The central methodological argument — that the mechanics of soma retrieval, not the quality of in situ recordings, govern what is recoverable in the proteome — is a genuinely useful contribution and is under-recognised in the existing patch-SCP literature. The decision to analyse all retrieval attempts rather than pre-filtering on protein count is principled and generates the paper's most solid empirical result: pre-retrieval capacitance and membrane resistance do not predict protein identifications (Figure 5C–D, n=6). The transparency of the work is unusually good for a preprint at this stage: raw MS data are deposited (PXD068359 / MSV000099156), videos of every retrieval attempt are public (Zenodo DOI), analysis code is on GitHub, and the Discussion is candid about compartmental bias, incomplete channel-complex recovery, and the impossibility of resolving distal-protein absence from soma-only sampling. The ethics and compliance record is clean and complete (IACUC 09-0006, grant numbers, competing interests declared).

The verdict is nevertheless major revision, and the reason is narrow and specific. Two claims that appear in the Abstract as positive findings are not supported by the analyses offered for them.

First, the headline correlation between retrieval-time capacitance and protein identifications is reported with inferential statistics — "F = 1577, p < 0.05, adjusted R² = 0.998, n = 3" — that cannot be interpreted as stated. A two-parameter linear model fit to three observations leaves one residual degree of freedom; the near-perfect R² is essentially forced by the geometry of fitting a line to three points, and the F/p pair is not diagnostic of a population-level relationship. Three specialist reports identified this independently, and the advocate in the panel debate conceded it without reservation. This is not a difference of statistical taste. The underlying observation — that in these three neurons, larger somas yielded more protein — is real and worth reporting, but it must be reported descriptively, and the abstract-level claim must be rescaled accordingly. Compounding this, the manuscript does not disclose whether these three gigaseal-preserved neurons were consecutive attempts or identified post hoc from a larger pool; without that, a reader cannot judge selection bias.

Second, the claim that spike preservation during relocation is associated with broader synaptic enrichment is confounded, in the manuscript's own data, by soma size. Neuron #6, which anchors the claim, is simultaneously the smallest neuron and the one with the fewest enriched terms; the paper elsewhere argues that size drives protein recovery, so fewer detected proteins offers a sufficient alternative explanation for its reduced enrichment. Worse for the claim, neuron #7 had visibly compromised spiking yet clustered with the well-retrieved neuron #4 on synaptic enrichment. The manuscript notes this anomaly rather than hiding it — which is to its credit — but noting it is not resolving it. As written, the spike-integrity claim is a post hoc ordering of three neurons with an unaddressed confound.

I want to be explicit about why this is major rather than minor. The reframing of Figure 3D is purely textual and could be done in an afternoon. But the size confound on the spike-integrity claim requires a reanalysis whose outcome could change the conclusion: if the number of enriched SynGO terms tracks total protein count across the full n=12 rather than tracking retrieval category, then the paper's second load-bearing claim dissolves into the first, and the Abstract and Conclusions must change substantively. That is a reanalysis with an uncertain result, not a wording fix, and it is what puts this decision over the line. No new experiments or new animals are required.

Two further points from the panel deserve mention but do not drive the verdict. Reviewer scientific_validity identified a real internal tension: the manuscript states that categorical retrieval outcome "does not reliably predict the biological content of single-neuron proteomes," yet also presents the framework as a benchmarking and interpretive tool. I read this as a framing problem rather than a contradiction — the framework is honestly post hoc and interpretive, and saying so plainly would strengthen the paper. Second, the Methods audit returned a substantial list of missing procedural detail. Most of it is ordinary reporting debt that a revision can discharge in the text (LC gradient composition, mass tolerances, UniProt build, PCA and GSEA parameters, software versions, microscope identification, rat sex). One item, however, is load-bearing for anyone attempting to evaluate the search itself: the manuscript states that a Mus musculus reference proteome was used to search Rattus norvegicus samples. If that is a typographical error it must be corrected; if it is what was done, it needs justification and an assessment of its effect on protein identifications, because every protein count in the paper depends on it. I have placed this high in the required revisions for that reason.

The framework, the shotgun strategy, and the negative result in Figure 5 are worth publishing. What is required is that the quantitative claims be brought back into line with what n=3 can carry, and that the size confound be tested rather than left standing.

Required Revisions

1. Resolve the reference proteome discrepancy. The Methods state that a UniProt Mus musculus proteome was used to search samples from Wistar rats. Correct this if it is an error. If a mouse database was in fact used, justify the choice explicitly and quantify the consequence — e.g., report identifications from a re-search against the Rattus norvegicus reference and state how protein-level counts, and any claim depending on them, change. Also give the specific UniProt release identifier.

2. Re-report the capacitance–protein relationship descriptively. Remove the F-statistic, p-value and adjusted R² from Figure 3D, its legend, the Results text and the Abstract. Report instead the three paired values (capacitance and protein identifications per neuron, identified by neuron number), state explicitly that a two-parameter model on three observations is saturated and that no inference about a population relationship is warranted, and rewrite the Abstract sentence so that it describes an observation in three neurons rather than a correlation. If additional gigaseal-preserved neurons exist in the dataset or can be added from existing recordings, include them and re-evaluate.

3. Test the soma-size confound on the spike-integrity claim. Across all twelve neurons, report the relationship between the number of significantly enriched SynGO terms (BP and CC separately) and total protein identifications. Then state whether retrieval category or spike integrity adds anything beyond protein count. If enrichment breadth tracks protein count, revise the Abstract, Results and Conclusions to say so, and drop or explicitly downgrade the claim that spike preservation is associated with synaptic enrichment. Address neuron #7 directly in the text rather than only noting it: state whether it is consistent with the size explanation.

4. Disclose how the three gigaseal-preserved neurons were selected. State whether they were consecutive attempts or identified post hoc, and report the overall success rate for gigaseal preservation across all attempts. If selection was post hoc, say so and discuss the resulting bias.

5. Substantiate or remove the 25–50% soma-loss estimate. Either show the calculation (e.g., the capacitance change from in situ to retrieval-time measurement per neuron, with the assumptions made in converting capacitance to material) or delete the numeric range and describe the loss qualitatively.

6. Reconcile the descriptive/predictive framing of the framework. The statement that categorical retrieval outcome does not reliably predict proteome content is inconsistent with presenting the framework as prospectively actionable. Revise the Conclusions and Abstract to position the framework as a post hoc interpretive and benchmarking aid, and state what evidence would be needed to make it predictive.

7. Report the statistics that are currently absent, or state that no test was performed. Specifically: (a) for Figure 5A, give median and range of protein identifications per retrieval category with the n per category, and either perform a named test or state explicitly that the group sizes preclude testing and the comparison is descriptive; (b) for Figures 5C–D, report the correlation coefficient and 95% CI alongside the p-value, and state whether the torn neurons (#11, #12) were included in the regression — if included, show the fit with and without them; (c) for the PCA in Figure 6A, report the variance explained by PC1 and PC2, and the scaling/centering used.

8. Specify the SynGO/GSEA analysis sufficiently to reproduce it. Give the SynGO version or download date, the GSEA variant and any permutation settings, the background gene set used, the multiple-testing correction method, and whether correction was applied within each neuron only or also across neurons when comparing enrichment breadth. State whether the "top terms" in Figures 4C and 6B were pre-specified or selected post hoc for display; if post hoc, say so in the legend.

9. Complete the mass spectrometry and LC reporting. Add: protein/peptide input quantification (or state that none was performed and that trypsin amount was fixed empirically); any post-digestion cleanup or desalting, or a statement that none was used; mobile phase A and B composition and the full gradient profile with column temperature; precursor and fragment mass tolerances; enzyme specificity as set in DIA-NN; minimum peptides per protein group; whether FDR was applied globally across runs or per run; and the MaxLFQ normalisation setting.

10. Complete the electrophysiology and imaging reporting. Add: sex of the animals; pClamp version; pipette puller model and pulling parameters (or cite a specific protocol); recording temperature; sampling rate and filter settings; the microscope, objective and NA used for DIC imaging; and criteria used for adjusting negative pressure within the −50 to −140 mmHg range. State whether neuron selection or the classification of retrievals from video was performed blind to MS results, and give the operational criteria used to assign a retrieval to "stable," "compromised" or "torn."

11. Fix the reference list. Remove or explain reference 30 (Guo et al., saxitoxin synthesis, Nature 2025), which is cited among the electrophysiology protocol references but has no evident bearing on the delegated methods. Indicate which of references 10 and 26–31 contains the definitive slice-preparation and recording protocol, and note any deviations from it. Provide a stable, citable locator (DOI, Zenodo, institutional repository) for NeuroExpress (ref. 15) or cite it as a personal communication. Confirm the publication status of references 9 and 29.

12. Report software versions for the R analysis stack (ComplexHeatmap, ggplot2, UpSetR, the PCA function or package used) and provide a commit hash or release tag for the GitHub repository corresponding to the analyses in this manuscript, and confirm the repository is public.

Minor Suggestions

• The panel noted a concurrent preprint on aspiration-based patch proteomics (Johnson et al., 2026, bioRxiv) that appears to address overlapping questions about retrieval and proteome recovery. If it is genuinely concurrent, a brief acknowledgement would strengthen the positioning; this is a courtesy, not a requirement, and I would not hold the paper for it.
• A short paragraph comparing your DIA-based transmembrane-protein recovery to the DDA-based recovery reported by Lee et al. and Ghatak et al. would help readers judge whether the improved coverage reflects acquisition mode, sample preparation, or instrument generation. A qualitative comparison against published counts is sufficient; no new experiments are needed.
• Figure 7 invites an obvious analysis that is not performed: whether the number or identity of detected ion-channel subunits tracks spike properties or synaptic enrichment. Even a null result here would be informative and would sit naturally with the paper's honest framing.
• Consider reporting vendor and catalogue numbers for the salts, DDM and formic acid. This is not essential for reproducibility but is inexpensive to add.
• Housing conditions, genotype/background, and a statement on how n=12 was arrived at (practical constraint versus planning) would round out the animal reporting.
• Video metadata (frame rate, duration, resolution) in the Supporting Information legends would help readers who wish to use the deposited videos as an independent check on retrieval classification.
• The claim in the Conclusions that the shotgun approach "reduces the risk of overinterpreting protein counts in isolation" is plausible but not demonstrated on these data. Either soften it or, if feasible, show what a selective-inclusion analysis of the same twelve neurons would have concluded differently.