Decision letter

VERDICT: minor

Summary of Evaluation

This manuscript documents the end-to-end translation of N332-GT5 gp140 — a germline-targeting HIV-1 Env trimer — from a stable CHO cell line through cGMP drug substance now being administered in HVTN 144. It is a careful, unusually complete technical record: cell line development with Leap-In transposition and two-stage clone ranking, Ambr®250 process optimisation, scale-up to XDR-50/XDR-200, a three-step purification train with documented robustness and worst-case excursions, intermediate hold-time stability, viral clearance against two model viruses, and product characterisation by SE-HPLC, RP-HPLC, BLI, two orthogonal glycoproteomics methods and negative-stain EM. Papers of this type are rarely published in this level of detail, and the field genuinely benefits: practitioners scaling other Env trimers or comparably complex glycoproteins will be able to use the parameter tables, hold-time limits and robustness envelopes directly. All five specialist reports agree the work is sound; scores clustered at 3–4.

The reviews and the debate converged on a single, consistent kind of problem, and it is important that the authors read it as such. The objections are not about whether the process works — the clinical material exists and has been dosed — but about the calibration of specific written claims to the evidence actually shown in this manuscript. Three items stand out.

First, and unanimously flagged by all five referees without prompting: the abstract asserts "genetic stability through 60 population doublings" as a headline result, but the Results section reports no outcome data whatsoever for that study. Sections 2.3.8 and 2.1.3 describe the protocol and state that sequences were confirmed, but no copy numbers, no productivity values at PD0/PD60±Gln, and no sequencing summary appear anywhere. The advocate in the debate conceded this outright. This is a claim with zero supporting data displayed, and it must be fixed — but it is fixed by reporting numbers the authors evidently already have, not by new experiments.

Second, the reference-standard change across runs (IAVI/Scripps 19Apr0088 → KBI S-20210314-0001 → KBI P65) confounds absolute titre comparisons between the RCB, MCB demonstration and GMP runs, and titre feeds into reported step yields. The authors disclose this themselves, which counts in their favour. The debate resolved this usefully by distinction: within-run harvest yields (91.4%, 90.8%, 91%) and the VCC/viability trajectories are unaffected, so the scalability argument survives at the level of recovery efficiency and growth behaviour; what does not survive as written is any cross-run statement about consistent absolute productivity, and the abstract's phrase "consistent product quality across multiple cGMP batches" — there is one cGMP batch. This is a wording and framing correction plus, ideally, a cross-calibration table if one exists.

Third, several quantitative claims are stated qualitatively or against unstated criteria: nsEM "nearly 100% native-like trimers" without reporting the fraction of the 6,086 particles so classified or the rejection criteria; ">99% trimeric purity" resting partly on a BLI assay that is a relative-potency measurement against an uncharacterised reference; the 0.7% HMW/LMW hold-time instability threshold and the 15% potency criterion without justification; several "<LOQ" impurity results without the numerical LOQ; and the "one particle per 250 million doses" viral clearance statement without the RVLP/mL, dose volume and harvest volume that would let a reader check it. Each of these is resolvable from existing records and analysis output.

I have considered whether any of this rises to major revision. The test I applied is whether fixing a central claim requires new experiments, new data, or a reanalysis whose outcome could change a conclusion. It does not. The reporting_reproducibility referee, who scored lowest, framed the confounds as breaks in the logical chain — a fair reading — but the remedies that referee itself proposes are: provide reference-lot characterisation data, clarify whether Table 10 compared split load material, and state acceptance criteria. Those are documents and text. The one item that would be a new experiment (a second independent cGMP batch) is not needed if the claim is scaled to the single batch that exists, which is the correct fix in any case. Similarly, requests for replicate clone-ranking runs, a second viral-clearance batch, or statistical tests on n=1 Ambr conditions are legitimate as observations but the appropriate response is to state the n and remove the implication of statistical superiority, not to redo the campaign. I have routed those to suggestions and questions rather than requirements.

On scope: the ethics referee raised whether an experimental bioprocess paper fits this venue. In Silico's scope is any original research manuscript whose claims a careful reader can evaluate from the manuscript and its cited materials, in any discipline; the "in silico" name refers to the referees, not the subject matter. The manuscript is in scope. It is also not clinical guidance — it reports manufacture and characterisation of a drug substance, not dosing or treatment recommendations — so the safety-related desk-rejection criterion does not apply.

The compliance audits add a substantial list of reporting gaps: antibody identifiers (PGT145, BG18_GL0, DEN3, 2G12 and the ELISA secondaries), cell line authentication and mycoplasma status, absent construct sequences, absent qPCR and RT-PCR primer sequences, incomplete MS acquisition parameters and no proteomics repository accession, incomplete EM processing parameters, no data availability statement, and four future-dated references with two malformed DOIs. None of these prevents evaluation of the central claims — the load-bearing citations (Dey 2018, Steichen 2019/2024, NCT05217641) all resolve and support what is attributed to them. I have folded the ones that materially affect reproducibility into the requirements and left the remainder as suggestions.

This is a useful paper that currently overstates itself in a handful of places and withholds data it appears to hold. Corrected, it is a resource I would point a colleague to.

Required Revisions

1. Report the genetic stability data. Add to Results the actual outcomes of the 60-PD study described in Section 2.3.8: transgene copy number, productivity, and mRNA/sequence identity for PD0, PD60+Gln and PD60−Gln, with the acceptance criteria applied. If the study was not completed for clone C235, say so and remove the claim from the abstract. As written, a headline abstract claim has no supporting data in the manuscript.

2. Correct the cGMP batch-number and reproducibility language. The abstract states "consistent product quality across multiple cGMP batches"; the manuscript describes one cGMP batch. Revise the abstract, Conclusions and Section 3.2.3/3.4 so that consistency claims are scaled to what was run: one GMP batch, one MCB demonstration run, one RCB supply run. Retain and make explicit the claims that are supported — comparable growth kinetics, comparable within-run harvest yields, and comparable BDS quality attributes between the demonstration and GMP batches (Table 11).

3. Address the reference-standard confound explicitly. Either (a) provide a cross-calibration between lots 19Apr0088, S-20210314-0001-SD2-E-M and P65 (e.g. BLI titre of a common sample measured against all three, with agreement stated), or (b) add an explicit statement in Section 3.2.3 that absolute titres across the three runs are not directly comparable, identify which downstream figures depend on titre (notably the step yields in Figure 16 and the >100% 2G12 yield), and restrict the productivity comparison accordingly. Also state which reference lot was used for the twelve Ambr®250 conditions in Figure 13, and confirm it was the same lot throughout; if not, the ranking of those conditions must be requalified.

4. Quantify the nsEM result. Report the total number of particles picked, the number entering and surviving each round of 2D classification, the fraction assigned to native-like trimer classes, and the morphological criteria used for that assignment. Replace "nearly 100% native-like trimers" with the measured fraction and its basis. If no particles were rejected as non-trimeric, state that as a number.

5. Reframe or substantiate the BLI contribution to the purity claim. BLI titre and BG18_GL0 binding are relative measurements against a reference standard that is not characterised in this manuscript. Either provide SE-HPLC/RP-HPLC characterisation of the reference lots, or state clearly that the >99% trimer claim rests on SE-HPLC and nsEM and that BLI is reported as relative potency/antigenicity rather than as a purity measure.

6. State the numerical LOQ for every assay reporting "<LOQ." This applies at minimum to residual Triton X-100 ("<0.002%" — give the corresponding ppm), residual Protein A, residual 2G12, residual HCP and residual DNA, in Tables 9, 10 and 11. Without the numerical limits, the reader cannot judge whether clearance meets specification or merely falls below detection.

7. Justify the hold-time and potency acceptance criteria. State whether the 0.7% HMW/LMW change threshold (Section 2.5.3/3.3.3) and the 15% relative-potency variation criterion were pre-specified before data collection, and give their basis (internal specification, regulatory guidance, or assay variability). Also state the numerical target potency range for the BDS referred to in Section 3.3.3.

8. Make the viral clearance claim checkable. Provide the estimated RVLP/mL in the bioreactor supernatant, the clinical dose volume and the harvest volume used to derive "less than one viral particle per 250 million doses." Also state, for each unit operation in Table 12, the number of independent spike experiments contributing to the reported mean ± SD, and note explicitly that the study was performed on one cGMP batch using a scale-down model.

9. Clarify the preparative SEC removal comparison. State whether the Control and SEC-Removal arms in Table 10 were run in parallel on a single split load, and label the compared intermediates so that the comparison is unambiguous (the current table compares SEC Load/Flowthrough against UF/DF Load/Final Retentate). State the pre-specified tolerance used to judge the two arms "comparable," and confirm unambiguously in Section 3.4 and Figure 16 that the cGMP batch was manufactured without the SEC step.

10. State the n and the selection basis for clone ranking and Ambr®250 optimisation. For Section 3.1.3, state that each of the 24 clones was assessed in a single 14-day run, give the quantitative ranking values for the top five clones, and state the thresholds used. For Section 3.2.2, state that each of the twelve conditions was a single vessel, and remove or requalify language implying that Control + 7a bolus was demonstrably superior to the next-best conditions (752 vs 721 vs 715 mg/L) where no replication or test supports the ordering.

11. Temper the novelty framing. Section 2.5 states the downstream process was developed from the published BG505 SOSIP.664 process (Dey et al., 2018), and the upstream and purification unit operations are established platform technologies. Revise "manufacturing paradigm" and "advancing the field toward rational vaccine design based on germline-targeting principles" in the abstract and Conclusions so that the manufacturing contribution — enabling a designed immunogen to reach the clinic, with a documented and transferable process — is distinguished from the immunogen design advance, which belongs to Steichen et al. (2019, 2024).

12. Add a data availability statement, and deposit the mass spectrometry raw data (PRIDE, MassIVE or equivalent) with an accession quoted in the manuscript. State where bioreactor logs, chromatograms, EM micrographs and analytical raw data can be obtained, or state plainly that they are available on request from a named contact.

13. Complete key reagent and method identifiers. Specifically: (i) antibody provenance for PGT145, BG18_GL0, DEN3 and 2G12 — expression source or supplier, and identifiers where they exist; vendor and catalogue number for the AP- and HRP-conjugated secondaries in Sections 2.2.4 and 2.2.7, with working dilutions; (ii) cell line provenance statement including whether the HD BIOP3-derived C235 clone and the MCB were tested for mycoplasma and adventitious agents (this is routine for a cGMP bank and presumably documented); (iii) primer and probe sequences for the residual CHO DNA qPCR and the RT-PCR transcript analysis, or a statement that they are proprietary and why; (iv) MS acquisition parameters for both glycoproteomics methods (resolution, AGC, precursor and fragment mass tolerances), enzyme sources and digestion/deglycosylation conditions, database version, and software versions for IP2/ProLuCID/Census2/GlycoMSQuant and Byos; (v) EM processing parameters — defocus range, exposure/dose, number of 2D classes — and the CryoSPARC version.

14. Fix the citation defects identified in the audit. Correct the malformed DOIs for Ramezani-Rad et al. 2025 (ref. 18) and Parks et al. 2025 (ref. 32); add a DOI or preprint identifier for Bale et al. 2025 (ref. 3), which is cited as the comparator for the nsEM claim and must be reachable; add the year to Sanders et al. (ref. 33); and confirm the publication status of all 2025-dated references. Also correct the CryoSPARC attribution — the reference given as "Pujani et al., 2017" is presumably Punjani et al., 2017.

Minor Suggestions

• Benchmark the final BDS attributes (%trimer, HCP, DNA, endotoxin, glycan profile) against the published BG505 SOSIP.664 cGMP batch (Dey et al., 2018) and, if reachable, Bale et al. (2025). This would let readers judge whether the achieved quality is state-of-the-art or routine for this molecule class, and would strengthen the platform argument at no experimental cost.
• Quantify furin cleavage by densitometry or MS rather than describing it as "close to complete" from gel appearance (Figures 8, 11).
• Discuss the DeGlyPHER/LC-MS occupancy discordance at N185e, N611 and N625 more substantively: state which method you treat as reference for occupancy and why, and whether the discordance has any bearing on the relative-potency result or on the epitope-relevant glycans (N332, N301, N137).
• Report error representation consistently across figures; several panels (Figure 14, Figure 10) currently carry none, while others report SEM.
• Consider stating briefly whether the achieved >99% trimer purity is expected to matter for the immunogenicity objectives of HVTN 144, or whether it is above the threshold where further purification would be consequential. A sentence would help readers outside bioprocessing calibrate the result.
• Sections 3.1.1 and the paragraph immediately following Figure 7 repeat the same two sentences about BG18/PGT145 comparability and SDS-PAGE loading volumes; delete the duplicate.
• Tables 1, 2, 6, 7, 8 and 9 have suffered layout corruption in the current PDF (column alignment lost, stray line numbers embedded in cells). These tables carry much of the paper's evidentiary weight and should be rebuilt before resubmission.

Questions for the Authors

1. Were the top-ranked clones re-tested in an independent fed-batch run before C235 was selected, or was selection based on the single run per clone shown in Figure 10?
2. Were the actual hold times used during the cGMP campaign within the limits established in the pilot-scale study, and were the ambient temperatures during those holds recorded and within 15–25 °C?
3. In the cGMP Capto adhere step, what were the actual loading factor, pH and conductivity, and how far were they from the worst-case-yield condition that gave ~24% step yield in development?
4. Why were XMuLV and MMV selected as the model viruses, and were they chosen as worst-case surrogates for the specific unit operations tested?
5. Is there any characterisation or stability data for the KBI reference lots (S-20210314-0001-SD2-E-M, P65) that could be summarised in a supplementary table?