Research prompt — KIF18A: same-MoA clinical evidence for anti-aging / dual-purpose development

Role & Persona

You are an expert translational pharmacologist, mitosis and cell-cycle biologist, and clinical strategy consultant specialising in chromosomal instability (CIN) biology, geroscience, and oncology.

Background Context

We are evaluating the therapeutic potential of targeting Kinesin Family Member 18A (KIF18A) — a kinesin-8 family microtubule-depolymerising motor protein (UniProt Q8NI77, NCBI Gene 81930, motor domain residues 11–355) that suppresses kinetochore oscillations and drives chromosome congression at the metaphase plate. This is a mitotic motor, not a chromatin-modifying enzyme: do not import epigenetic, senotherapeutic or MYST-family framing from adjacent target classes, as none of it applies here.

Three internal reports are attached:

1. kif18a_anti_aging.md — biological rationale for KIF18A inhibition in anti-aging, argued via aneuploidy clearance, micronuclei formation, cGAS-STING activation and inflammaging.
2. kif18a_indication_prioritization_micheal_skill_v2.html — PandaOmics Agent indication prioritisation (2026-08-13), scoring KIF18A against 1,000 indications across 14 therapeutic areas on 23 omics, bibliometric and financial metrics, then analysing the top 50 in oncology and aging-relevant areas in detail. This version supersedes the earlier v1 report; use v2 wherever they differ.
3. kif18a_indication_prioritization_base_pandaclaw.md — a narrower 13-indication composite scoring with undisclosed weights.

Terminology. Write “selective KIF18A inhibitor” throughout; never “pan-KIF18A.” A pan inhibitor is meaningful for KAT6 because KAT6A and KAT6B are close paralogs requiring dual inhibition. KIF18A’s closest paralog KIF18B is only 36.4% identical (KIF25 27.6%, KIF3B 27.2%), so kinome selectivity is the design goal, not breadth.

Critical caveat — the reports disagree on the central question, two to one. Report 2 concludes that “no published evidence links KIF18A directly to aging biology, cellular senescence, or neurodegeneration” and that “aging-related indications lack direct evidence and are not recommended for pursuit,” flagging them Red and No-go. Report 3 concludes that “the dual-purpose oncology-aging narrative does not hold for a KIF18A inhibitor compound.” Only Report 1 grades the anti-aging case “moderate-to-strong.” Your first task is to adjudicate this, not to select the conclusion you prefer.

Core Objective

Thoroughly review the three background reports, then perform a rigorous clinical translation assessment to determine whether selective KIF18A inhibitors — or drugs with an identical or adjacent Mechanism of Action — have demonstrated human clinical evidence of safety, efficacy and robustness for anti-aging, age-related degenerative disease, or dual-purpose (oncology + geroscience) indications.

A well-evidenced negative answer is a fully acceptable and expected deliverable, and is more useful to us than a manufactured positive one.

1. Source Adjudication (complete before building on any report)

Rule on each conflict below, stating which reading survives and why.

• Direction of effect — the load-bearing conflict. Biswas et al. 2024 PNAS (PMID 39475646) is cited by Report 1 as supporting the aging thesis and by Report 3 as contraindicating it: maternal KIF18A motor-domain variants prematurely increase egg aneuploidy, so KIF18A function is protective, and Report 3 concludes an inhibitor “would likely worsen, not improve, oocyte quality” and that reproductive aging “would require KIF18A activation, not inhibition.” Report 2 independently points the same way — Kif18a-null mice are viable but show male infertility with germinal cell aplasia, testis is the highest-expressing normal tissue (12.8 nTPM), PandaOmics ranks infertility #8, and ClinVar pathogenic variants associate with instability and infertility phenotypes. Weigh this convergent loss-of-function-harms-reproduction evidence against a thesis in which pharmacological inhibition is beneficial, and reconcile the two or state plainly that they cannot be reconciled.
• Category coherence. Report 1 frames KIF18A inhibition as a “preventive senolytic.” Senolytics kill non-dividing senescent cells; a KIF18A inhibitor can only kill a cell that enters mitosis. Assess whether this is a category error — the anti-aging thesis rests on it.
• Aging clock evidence. Report 1 claims KIF18A is a top-10 feature (#9 of 227) in a mammalian life-history clock; Report 3 states it is “present in 4 clocks but NOT among top features.” Report 1 also reports PASTA and REG coefficients with opposite signs, and a ZhangBLUP section headed “11 CpGs” that tabulates 14 rows.
• Falsifiability. Report 1’s GTEx v10 analysis treats both rising and falling expression with age as supportive (10 tissues decreasing, 7 increasing). Identify what result would have falsified the hypothesis; if none would have, say so.
• Selectivity claim vs essentiality. Report 2 states both that KIF18A “is classified as an essential protein (HPA)” and that it “is dispensable for normal cell division,” resolving this by asserting that DepMap dependency is enriched in CIN-high cell lines while normal and CIN-low cells are not dependent. Determine whether the published CIN-selectivity data actually resolves the tension or merely restates it — this single claim carries the entire therapeutic window.
• Constraint, unresolved within Report 2. Section 7.2 asserts KIF18A “shows evidence of LoF intolerance,” while the Limitations section states “gnomAD constraint values were not directly accessible in this environment.” Retrieve the actual gnomAD LOEUF and pLI yourself and rule on it, since LoF tolerance versus intolerance changes the chronic-dosing safety argument materially.
• Aging rank discrepancy across versions. The superseded v1 report placed the classical aging indications far down the 1,000-item list (Alzheimer’s #432, skin aging #513, Parkinson’s #546, osteoporosis #619, age-related macular degeneration #856); v2 reports neurodegenerative disease at #40 and diabetes mellitus at #37 within its filtered top-50 view, both flagged No-go. Establish whether this is an ontology-term difference or a scoring change, and state which ranking you are relying on.

2. Human Clinical Trial Landscape Assessment

Report 2 supplies a materially complete census of six active Phase 1/2 KIF18A programs; treat it as a starting point and verify every record against the ClinicalTrials.gov v2 API individually, confirming that sponsor and intervention match the claim. An NCT ID that resolves can still be the wrong ID, and one in these reports does not resolve at all.

Also resolve HS387 (Zhejiang Hisun) and the Shanghai Apeiron cyclohexenyl series, and correct the ATX020 attribution: Report 3 assigns it to “Athenex / NCI,” but its own reference list cites Sparling et al. 2025 ACS Med Chem Lett (PMID 41257005) and Nair et al. 2025 Cells (PMID 41369352), both Accent Therapeutics work. Adjudicate the conflicting VLS-1488 ASCO 2025 readouts against the primary abstract — Report 2 reports no dose-limiting toxicities up to 800 mg in 52 patients; Report 3 reports 52 enrolled with 7 of 17 evaluable ovarian patients showing tumour reduction and 3 confirmed partial responses.

For each asset, summarise the human data actually available: pharmacokinetics and pharmacodynamics, target-engagement biomarkers, dose-limiting toxicities, maximum tolerated dose and recommended Phase 2 dose, and primary efficacy endpoints. No Phase 2 or Phase 3 efficacy data exists anywhere in this class — report the Phase 1/1b evidence as what it is rather than against a Phase 1–3 template that would imply maturity the field does not have. Cross-check every asset against the local Citeline Pharmaprojects export at ~/citeline_drugs_export/merged/combined_all.csv, which takes precedence over other sources.

3. Mechanism-of-Action Mapping (concentric rings)

Work outwards and report what human evidence each ring actually carries:

1. Other KIF18A inhibitors — direct MoA identity.
2. CIN / aneuploidy-selective lethality — MPS1/TTK inhibitors (empesertib, CFI-402257), CENP-E inhibitor GSK923295, and any whole-genome-doubling-selective agent.
3. Antimitotics broadly — Eg5/KSP inhibitors (ispinesib, filanesib, litronesib), PLK1 inhibitors (volasertib, onvansertib), Aurora A/B inhibitors (alisertib, barasertib), with taxanes as the tolerability yardstick.
4. Replication-stress adjacency — WEE1 and PKMYT1 inhibitors.

Test the differentiation claim rather than restating it. Report 2 documents that ispinesib failed Phase 2 across melanoma, HNSCC, HCC, RCC and prostate cancer with no objective responses in any trial, attributing the failure to absent patient-selection biomarkers, a narrow window driven by bone marrow toxicity, and allosteric resistance (Talapatra et al. 2013) — then asserts KIF18A inhibitors are “fundamentally differentiated by CIN-selective synthetic lethality.” That assertion is the entire investment case and is currently supported by preclinical data only. Establish what human evidence, if any, supports it, and note that Report 2 lists competitive intensity (five-plus companies, same mechanism) as the primary risk rather than mechanism failure.

4. Indication Cross-Validation (prioritisation vs clinical reality)

Compare the internally prioritised shortlist against active clinical pipelines and real-world data, then sort surviving indications into three strategic buckets: (a) pure oncology, (b) pure aging/degenerative pathology (fibrosis, neurodegeneration, metabolic and vascular aging), and (c) genuine dual-purpose.

Bucket (b) is where the internal analysis is weakest and deserves the most scrutiny. Report 2 finds oncology occupying 29 of the top 50 filtered indications with breast cancer #3, HCC #4, CRC #5, GBM #10, ovarian carcinoma #22 and NSCLC #24, while the neurologic, cardiovascular and endocrine areas contribute only 7 low-scoring indications whose signal is “driven by network co-embedding rather than direct evidence.” The Alzheimer’s differential-expression signal is logFC 0.12 and judged “likely driven by glial proliferation rather than a direct KIF18A-disease mechanism,” against 0.52 in HCC and 0.49 in gastric carcinoma. State plainly if bucket (c) turns out to be empty.

One trap to avoid: Liu M et al. 2023 on KIF18A and PTEN signalling in colorectal cancer was published in the journal Aging. Journal of publication is not evidence of aging biology — do not let venue names inflate the aging evidence count.

5. Safety, Therapeutic Window & Translational Risk

Assess the safety profile and therapeutic window explicitly comparing acute, intermittent, high-dose oncology schedules against the chronic low-dose regimens a geroscience intervention would require. This gap is the central translational risk and is wider here than for a reversible epigenetic agent: KIF18A inhibition drives irreversible mitotic catastrophe in any cell that enters mitosis.

Cover the proliferative-tissue liabilities that follow from baseline expression — testis 12.8 nTPM, lymphoid tissue 8.9, bone marrow 7.8 — namely bone marrow toxicity (the acknowledged theoretical risk, and the specific liability that killed the Eg5/KSP class), gastrointestinal mucosal injury, and male fertility, for which the Kif18a-null germinal cell aplasia phenotype is direct in vivo evidence rather than inference. Note that “no dose-limiting toxicities up to 800 mg” in a 52-patient Phase 1 of advanced cancer patients establishes acute tolerability only and says nothing about months-to-years dosing in healthy people.

Address the species mismatch: mouse (76.2%) and rat (71.9%) identity fall below the 80% high-confidence threshold, and Report 2 recommends macaque (96.8%) and dog (85.4%) as preferred translational models. Any chronic-dosing toxicology package for an aging indication inherits that constraint, at substantially higher cost.

Then answer the decisive regulatory question. Aging is not an approvable indication at FDA or EMA, so set out the actual path and the standard workaround of a specific age-related disease indication. Benchmark against the real geroscience state of the art with hard numbers and NCT IDs — dasatinib plus quercetin, fisetin, navitoclax, UBX0101 (which failed in osteoarthritis), UBX1325 in retinal disease, rapamycin/PEARL, metformin/MILES — and state which biomarkers (epigenetic clocks, IL-6/CRP, p16INK4a burden, gait speed, SPPB) have ever supported a real go/no-go decision rather than merely being reported. Finally: what safety threshold and dosing paradigm would a DSMB require before a cell-cycle-targeting cytotoxic agent is given to a non-cancer or healthy aging population?

6. Evidence Standards & Verification Protocol

• Grade every reported result by source class: peer-reviewed publication, conference abstract, or company press release. Keep these separated in the reference list.
• Never invent an NCT number, PMID, date or result. Mark anything you cannot verify as UNVERIFIED and state what you tried.
• Avoid composite 0–10 scores with undisclosed weights of the kind Report 3 uses. Give evidence tables with named sources and explicit grades.
• Positive control for negative findings. If you conclude that no drug in this class has ever been dosed in a human trial for a non-oncology, aging or healthspan indication, first prove the search could have found one — run a control query that does return non-oncology trials for a cytotoxic or cell-cycle agent (low-dose methotrexate in cardiovascular inflammation is a suitable control) and report it alongside the negative result.
• Report 2 explicitly did not run PheWAS on LoF variants, Mendelian randomisation with colocalisation, disease scRNA atlas expression, or IMPC knockout review. Where any of these would change the aging verdict, say so and price it.
• Tooling. Do not use the WebSearch tool; it fails in this environment. Use the Tavily API for web search, gh for code and repository search, and cfget or the web-scrape skill for Cloudflare/JS/403-blocked pages. ClinicalTrials.gov v2 is the authoritative trial source; verify against it and independent press, never company press releases alone.
• Audience. This report is internal, so real compound codes including ISM9682 are acceptable; flag which facts are public in case we later cut an external version. Write in full sentences, spell out jargon on first use, and write “cell line” in full.

Deliverable Format & Structure

A comprehensive, executive-level technical report:

1. Executive Summary — the direct answer, its confidence level, the single biggest reason it could be wrong, and a high-level Go / No-Go verdict on dual-purpose development.
2. Source Adjudication — resolution of the seven conflicts in §1, with the surviving reading for each.
3. Mechanism & Pathway Synthesis — mapping of KIF18A in aging versus cancer pathways, and whether the two converge mechanistically or only rhetorically.
4. Clinical Benchmark Table — a structured Markdown table of every verified KIF18A and same-MoA clinical asset: asset name, sponsor, target/MoA, phase and status, NCT ID, indication, key clinical findings, safety profile, and evidence grade.
5. MoA Ring Map — human evidence carried by each concentric ring, with the ispinesib Phase 2 failures analysed as precedent rather than footnoted, and an explicit verdict on whether CIN-selectivity has yet been shown in humans to supply the window that class lacked.
6. Prioritisation Validation & Dual-Purpose Mapping — internal DEG and pathway findings matched against human trial evidence, sorted into the three strategic buckets.
7. Strategic Recommendations & Translational Gaps — what would have to be true to move this from speculative to credible, in cost order, each item paired with the experiment that would falsify it, plus recommended preclinical and clinical designs for an anti-aging proof of concept.