Geng integrity report: GPAT4 sustains endoplasmic reticulum homeostasis in endocardial cells and safeguards heart development (Nature Communications, 2025)

Academic fraud report · Geng Detector

Summary

Verdict: highly suspicious. The original Geng report focuses on methodological and statistical concerns rather than confirmed image manipulation. Key issues: (1) AlphaFold 3 was incorporated into the manuscript although the paper was received on 19 March 2024, before AlphaFold 3 was released on 8 May 2024, suggesting late-stage addition of a trending tool; (2) the Global KO (Figure 1H) and Tie2-Cre conditional KO (Figure 2H) Western blots show identical molecular weight annotations (52/21/28/42 kDa) and highly parallel result text, raising concerns about possible image reuse; (3) multiple P values cluster just below 0.05 (e.g., Figure 6N: P=0.0495; Figure 5D: P=0.0349, P=0.0420), a pattern suggestive of P-hacking; (4) a minor typo (Sakura written as 'Sukura'); (5) the in vitro HUVEC siGPAT4 experiments (Figure 5) reproduce every in vivo step of the signaling cascade with unusual perfection. No pixel-level image duplication was demonstrated. These are warning signs warranting PubPeer scrutiny and access to raw blots, not proof of fraud.

Verdict

🟠 Highly suspicious. The report identifies several warning signs — temporal inconsistency with AlphaFold 3, parallel figure construction between Global KO and Tie2-Cre KO panels, suspiciously borderline P values, and over-complete in vitro corroboration — but does not provide pixel-level evidence of image manipulation. The findings are suggestive of cosmetic revision and selective data presentation, not confirmed fraud.

Key findings

AlphaFold 3 timing anomaly: The manuscript was received 19 March 2024 and cites Abramson et al., Nature 2024 (AlphaFold 3) in Methods/Figure 3G, I, J. AlphaFold 3 was first released/preprinted on 8 May 2024, which is after submission. The tool was therefore added during revision.
Parallel Global KO vs Tie2-Cre KO panels: Figure 1H and Figure 2H both show BAX (21 kDa), BCL2 (28 kDa), ACTIN (42 kDa) and an additional ~52 kDa band, with identical molecular weight annotations. Corresponding result subsections ('Defective heart…' and 'Essential role…') follow nearly identical sentence structures.
Borderline P-value clustering: Figure 5D P = 0.0349, P = 0.0420; Figure 6N P = 0.0207, P = 0.0495; Figure 6P P = 0.0306, P = 0.0269. The prevalence of values just below 0.05 is consistent with P-hacking or selective reporting.
Reagent vendor typo: 'OCT medium (Sukura, 4583)' in the Immunofluorescence Methods section; the correct vendor is Sakura (Tissue-Tek).
Over-complete in vitro rescue: Figure 5 (HUVEC, siGPAT4) cleanly recapitulates the entire in vivo cascade (ER stress → mitochondrial Ca²⁺ → cGAS–STING → mtDNA release) across panels A–V, with no apparent conflicting data points.

Evidence highlights

DOI: 10.1038/s41467-025-58722-5
Timeline: Received 19 March 2024; Accepted 26 March 2025.
Authors: Tianyang Zhao, Kuipei Jin, Xiaodong Wang, Xiong Su, Youjun Wang, Mingming Gao, Wen Luo, Hongyuan Yang & Zhongzhou Yang. Journal: Nature Communications (2025).
AlphaFold 3 reference (Ref 26: Abramson et al., Nature 2024) post-dates original submission by ~7 weeks.
Western blot bands labeled at 52 kDa, 21 kDa, 28 kDa, 42 kDa in both Figure 1H and Figure 2H.
Specific P values cited: 0.0349, 0.0420, 0.0207, 0.0495, 0.0306, 0.0269.

Notes

Confidence: low-to-moderate. The anomalies are structural and statistical; no raw image comparison was performed within this report.
Limits: pixel-level duplication, gel source data, and original full-length blots are not available. The 'Sukura' typo is weak standalone evidence but consistent with careless assembly.
Recommended next steps: request the authors provide unprocessed scans of Figure 1H and Figure 2H; verify whether the same ACTIN loading control and exposure were reused across two distinct genotypes; cross-check raw qPCR/biological replicates behind the borderline P values; confirm the date AlphaFold 3 was actually used (revision vs. original submission). The mild AlphaFold 3 addition is permissible in revision but warrants clarification of when computational predictions were generated and whether they drove any experimental reinterpretation.
This report is AI-assisted and intended for academic discussion only; it does not constitute a determination of misconduct.