Review of "Novel MECP2 gene therapy is effective in a multicenter study using two mouse models of Rett syndrome and is safe in non-human primates" (Molecular Therapy, 2023)

Academic fraud report · Geng Detector

Summary

This assessment, generated from a Chinese-language 'Geng' academic-integrity review, reaches a verdict of "questionable" rather than confirmed fraud. The paper reports a large multinational preclinical study of MECP2 gene therapy for Rett syndrome, spanning two mouse models and an 18-month non-human primate (NHP) toxicology arm, and is therefore positioned as a candidate for clinical translation. The review's most concrete red flag is text-level: Figure 2E and Figure 2F legends report p-values of 'p > 9999' and 'p = 9976', which are mathematically impossible for a probability and strongly suggest a copy-paste or data-export column-shift error (consistent with mis-handled Prism/Excel output). A separate Figure 4L legend duplicates the word 'velocity' where a paired 'distance' metric is expected. Methods text unusually records Hamilton syringe lot numbers and 2005 dates, implying raw SOP copy-paste. The NHP safety arm uses only n=3 animals (split n=2/n=1) compared against non-contemporaneous historical saline controls, which is methodologically weak but not inherently fabricated. Image-based reuse/splicing checks (Western blots, IHC) were not feasible from the text-only extract. Confidence is moderate; findings should be treated as prompts for raw-data inspection, not as proof of misconduct.

Verdict

Questionable / Requires raw-data verification. No direct evidence of fabricated data or image manipulation was established from the textual record alone. Several surface anomalies — most notably impossible p-values, a duplicated caption term, and unusual inclusion of consumable lot metadata in Methods — indicate sloppy data handling, weak experimental design (historical NHP controls, n=2/n=1), and insufficient editorial proofreading. Image-level analysis of Figures 3, 5, 6, and 7 was not possible from this review and is required before any fraud determination.

Key findings

Impossible p-values in figure legends. Figure 2E legend reports p > 9999 for the WT $3.00\times 10^{10}$ dose group; Figure 2F legend reports p = 9976 for the same dose group. A probability cannot exceed 1, so these strings almost certainly reflect a column-shift / format error during export from statistics software or a spreadsheet paste operation. The error pattern is consistent with the Figure 4L duplication of the word velocity (where a paired distance term is expected), reinforcing a copy-paste artifact interpretation.
Methods text appears copied from raw laboratory records. Hamilton syringe lot numbers and 2005 dates (Lot #324603 11/08/2005, 02/17/2005) are listed in the NHP handling section. Academic methods sections rarely include consumable lot IDs unless text was mechanically lifted from an SOP or bench notebook.
NHP toxicology design is statistically and methodologically weak. The 18-month NHP safety study used only 3 treated animals (n=2 at $1.6\times 10^{13}$ vg; n=1 at $9.2\times 10^{12}$ vg) with no concurrent control group. Toxicity, body weight, and IHC endpoints (Figures 6, 7) were instead compared against historical saline-injected control tissues drawn from the same colony but different batches/time points. Because the antibody cannot distinguish endogenous from exogenous MECP2, batch effects cannot be excluded at this sample size.
Cross-site reproducibility discrepancy is honestly reported. Median survival of untreated KO mice differed between NCH (USA; 68 days) and University of Edinburgh (UK; 86.8 days). The authors attribute this to husbandry differences and/or phenotypic drift between geographically separated sub-lines. Honest disclosure of inter-site variability is more consistent with genuine biology than with fabrication.
Image-based checks were not performed. Western blots and IHC panels in Figures 3, 5, 6, and 7 could not be examined for lane splicing, background anomalies, or reuse because only text was available.

Evidence highlights

DOI: 10.1016/j.ymthe.2023.07.013 — *Molecular Therapy* Vol. 31, No. 9, September 2023. Received 8 October 2022; accepted 19 July 2023.
Figure 2E legend: p > 9999 for WT $3.00\times 10^{10}$ dose group (mathematically invalid).
Figure 2F legend: p = 9976 for the same WT $3.00\times 10^{10}$ dose group (mathematically invalid).
Figure 4L legend: duplicated term velocity where a distance metric is expected.
Methods, NHP handling: explicit Hamilton syringe lot numbers and 2005 dates retained verbatim.
NHP safety arm: n=3 treated animals, split n=2 ($1.6\times 10^{13}$ vg) and n=1 ($9.2\times 10^{12}$ vg); compared against historical saline controls.
Survival of untreated KO mice: 68 days (NCH) vs. 86.8 days (University of Edinburgh).

Notes

The textual anomalies are more parsimoniously explained by careless data export / proofreading than by intentional fabrication, but they raise legitimate concerns about data-management hygiene for a paper intended to support an IND-enabling toxicology package.
The NHP design limitation is a methodological weakness rather than evidence of misconduct; it should be raised in peer review of any related clinical-translation proposal.
A PubPeer comment focused on the p-value artifacts and the historical-control design is reasonable; a formal journal complaint would be premature without raw-image and raw-data inspection.
Reviewer confidence: moderate for the textual findings (direct quotations from the legends/Methods); low for any fraud inference, because pixel-level image analysis has not been performed.