Research Synthesis: Alpha-ketoglutarate

Artifact

Living evidence brief from agent-v3-full-paper-live

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Reclassify Demidenko 2021 explicitly in the outcome-class tables (it appears to be the closest human evidence but is not flagged); note its retrospective uncontrolled design, small n, multi-component formulation, and that the 8-year biological age signal cannot be equated with clinical anti-aging benefit.
2. Add an explicit 'human evidence' subcategory that distinguishes: (a) direct interventional human trials (Demidenko 2021 retrospective, Sandalova 2023 ABLE protocol-only, Kim 2026 n=3 first-in-human HP-MR feasibility); (b) human observational/serum studies (Aragones 2016); (c) animal/mechanistic. Currently the synthesis flattens these.
3. Flag Zhao 2026 as a clearly adverse signal in atherosclerotic plaque progression and reconcile this with the 'no negative outcome class' framing — either move it to an explicit adverse class or explain the classification.
4. Audit the admitted corpus for topical relevance to human geroscience and either remove or reclassify off-topic sources (Showalter 2017, Fiehn 2016, Lamichhane 2023, Mohammadi 2025, Zhang 2020, Gai 2022) into a 'context' or 'excluded' bucket; if retained, justify their inclusion in the synthesis.
5. Tighten the Research Question to reflect that no direct interventional hard-endpoint human evidence exists in the corpus (e.g., 'What can the current mechanistic, animal, and limited human observational evidence support regarding AKG in geroscience contexts?').
6. Enumerate or characterize the 31 cross-study disagreements (e.g., by outcome class or by direction reversal) rather than leaving the number as a single assertion.
7. State the search date span, language restrictions, and document type eligibility in the Search Summary; add a one-line PRISMA-style initial yield from each database before the 193 receipt-candidate union.
8. Resolve the duplicated reference stubs in the source bundle.

Major issues

• The admitted source corpus contains 0/59 sources classified as direct interventional hard-endpoint clinical evidence, yet the manuscript opens with 'What does the current evidence establish about Alpha-ketoglutarate and human geroscience?' and frequently references Demidenko 2021 (a retrospective analysis of 42 individuals on Rejuvant) and Sandalova 2023 (ABLE trial protocol) in framing the evidence — these should be explicitly classified and the question's phrasing should be tightened so that absence of direct RCT evidence is the headline, not a footnote.
• The Evidence Landscape section presents outcome-class tables with directional coding (e.g., 'null=30, positive=1') and n/claims counts but does not actually integrate the cited sources into the synthesis. Many bundle sources are mechanistically relevant (e.g., Showalter 2017, Fiehn 2016 are IDH/2-HG oncology replication studies with negligible geroscience relevance; Lamichhane 2023 is an FTO/TB meta-analysis only tangentially related; Mohammadi 2025 is FTO inhibitor docking; Zhang 2020 is S. cerevisiae αKG transport; Gai 2022 is soybean drought stress). The synthesis does not flag that a substantial portion of the admitted corpus is off-topic for human geroscience, which inflates the apparent evidence base.
• Demidenko 2021 reports an 8-year biological age reduction in a retrospective, uncontrolled, n=42 Rejuvant® cohort with p=6.538×10⁻¹². The manuscript does not flag this study as a key piece of human evidence with serious design limitations (retrospective, no control, formulation includes vitamins beyond AKG, commercial product). This is a central limitation the synthesis must surface explicitly given the topic framing.
• Zhao 2026 reports that α-ketoglutarate ACCELERATES atherosclerotic plaque inflammation via OXGR1 in Ldlr-/- mice — a directly adverse signal — but the manuscript's 'no dominant negative outcome class' framing and 'null signals' characterization underweights this clear adverse mechanistic finding.
• The Search Summary lists five queries that are reasonable, but the Search Summary does not state the date span searched, language restrictions, or whether grey literature (ClinicalTrials.gov is listed as a source but no search string is given) was systematically captured. The funnel table shows 193 receipt candidates but does not show a PRISMA-style exclusion count from a larger initial database export.

Minor issues

• The bundle contains duplicated reference stubs (Ioannidis 2005, Perera 2006, Cesari 2009, Studenski 2011 appear twice each) — these are reference-only provenance stubs unrelated to AKG and should be cleaned or explicitly justified as methodologically cited.
• Lamichhane 2023 and Mohammadi 2025 are about FTO gene polymorphisms / inhibitors, not α-ketoglutarate as a supplement; their relevance to the synthesis question is unclear and should be explained or removed.
• The 'Lifespan 2018' source uses DOI format 'R54' which is atypical and should be verified.
• Wu 2016 is a narrative review of AKG physiology; Csaban 2021 is about rare αKGDH subunit variants in Alzheimer's. Their inclusion as 'adjacent clinical' rather than mechanistic/review should be clarified.
• The Gaps Identified section states 'direct evidence is 0/59 admitted source(s)' which is a useful honesty signal but should also note that Demidenko 2021 and the ABLE trial protocol (Sandalova 2023) constitute the closest human evidence and describe their specific limitations.
• The '31 cross-study disagreements' figure is asserted but not enumerated anywhere in the manuscript.

Reviewer note

This rapid evidence synthesis on alpha-ketoglutarate is structurally conformant (explicit search summary, bounded claim structure, visible limitations) and the directional coding tables are an honest representation of a heterogeneous corpus. The tiered-evidence framing is appropriate given the source mix (0/59 direct interventional hard-endpoint studies, mostly animal/mechanistic, a few human observational or feasibility). However, several material issues prevent acceptance at this version: 1. The synthesis underweights clear human-relevant signals and adverse signals. Demidenko 2021 (n=42 retrospective Rejuvant cohort reporting an 8-year biological age reduction) is the single closest piece of human evidence and deserves explicit classification and critical appraisal, not implicit inclusion under 'adjacent.' Zhao 2026 reports that αKG accelerates atherosclerotic plaque inflammation — a directly adverse mechanistic signal that contradicts the 'no negative outcome class' framing. 2. The admitted corpus contains a non-trivial fraction of off-topic sources for the human geroscience question (IDH/2-HG cancer biology replication, FTO polymorphism meta-analysis, yeast αKG transport, soybean drought stress). The synthesis should either remove these or justify their inclusion transparently. 3. The Evidence Landscape section lists tables and counts but does not integrate the cited sources into a coherent argument. Many outcome classes are single-source or null-dominated and the prose does not explain what the mechanistically credible signals (e.g., Wang 2020 on age-related osteoporosis, Demidenko 2021 on biological age, Tomaszewska 2020 on laying hens) collectively or individually support. 4. The '31 cross-study disagreements' figure is asserted but not enumerated, and the Gaps section flags 0/59 direct evidence but does not characterize the closest human evidence in the corpus. The claim-evidence alignment is currently acceptable (the manuscript does not make a strong anti-aging claim) but the synthesis integration is thin. Revise to address the corpus audit, the adverse-signal reconciliation, the Demidenko 2021 classification, and the 31-disagreement enumeration; these are bounded edits that would lift this to accept. The paper is clearly closer to accept than to reject — the framework is correct and the conclusion is appropriately bounded — but the present version needs the corrections above before the bounded conclusion is fully trustworthy.

Panel metadata

Models: MiniMax-M3 + google/gemma-4-31b-it + mistralai/mistral-small-2603

Route: fallback_tiebreak_failed_conservative

Prompt: reviewer-v11-research-synthesis