Repurposed antiparasitics in oncology: what the mechanism predicts

A recurring claim in the oncology drug-repurposing literature is that two approved antiparasitic agents — ivermectin, an avermectin, and mebendazole, a benzimidazole — produce clinically meaningful tumour control when taken orally by cancer patients. An observational report this spring put a specific number on it: a 197-patient cohort with a reported clinical benefit rate above 80% on a combined ivermectin–mebendazole protocol, described by its authors as hypothesis-generating and accompanied by an explicit call for randomised trials.

The biological question behind that report — do these drugs have real anti-cancer activity at orally achievable concentrations? — is a fair one, and a mechanism-led answer is possible without waiting for an RCT. Both drugs have well-characterised molecular targets, both have published clinical PK, and both have in vitro and xenograft work that defines the concentration thresholds at which each mechanism engages. The composed body model lets us assemble these and ask what the drug-attributable effect at 90 days should be — independently of what an observational cohort reports.

The short answer the model gives is: the mechanisms are real, the drug-attributable effect is much smaller than the reported clinical benefit rate, and the trial that would settle the gap has not been run.

Ivermectin: three thresholds, only two reachable

Ivermectin acts on tumour biology through at least three documented mechanisms, each with a distinct concentration threshold.

The lowest-threshold arm is immunomodulation. P2X4/P2X7 ATP-axis activation depletes myeloid-derived suppressor cells and shifts the CD8 T-effector to regulatory T-cell ratio, converting “cold” tumours toward an immunologically “hot” phenotype. The EC50 sits near 0.05 µM plasma, anchored on Draganov 2020 (npj Breast Cancer). At a standard 12 mg oral dose the plasma Cmax reaches roughly 0.05 µM (Schmith 2020, Clin Pharmacol Ther) — at threshold.

The second arm is PAK1 inhibition leading to cytostatic autophagy. PAK1 ubiquitination → AKT/mTOR inhibition → growth arrest. The threshold is approximately 1.5 µM, but in tissue, not plasma (Dou 2016, Cancer Research). Ivermectin has a log P near 5.8 and a published tumour-tissue partition coefficient of about 25× plasma (González Canga 2008, AAPS J; Lifschitz 2000 Vet Parasitol). A 25 mg/day regimen yields a plasma trough near 0.11 µM, corresponding to tumour-tissue concentration on the order of 2.8 µM — above the PAK1 threshold.

The third arm is direct cytotoxicity, with thresholds above 5 µM. This is unreachable in human plasma at any tolerable oral dose; ivermectin tolerability degrades sharply at higher exposures (Navarro 2020 meta-analysis; Muñoz 2018 fixed-dose PK/AE study), so the cytotoxic arm is not engaged at clinical exposures.

The picture this produces is concentration-driven, not “yes” or “no”: immunomodulation is at its EC50, PAK1 is above its tissue EC50, and the direct cytotoxic arm is not reached.

Mebendazole: tubulin, saturated

Mebendazole’s primary anti-mitotic mechanism is tubulin depolymerisation — the same class as the vinca alkaloids and colchicine, with an intracellular EC50 near 0.3 µM and cooperative binding (Hill coefficient ≈ 2). Oral bioavailability is around 20% and food-dependent (Dawson 1985, Br J Clin Pharmacol). At a 500 mg/day dose taken with food the steady-state plasma sits near 1 µM; with a lipophilic tumour-partition factor of about 5×, the effective tumour concentration reaches the low-single-digit micromolar range — well above the tubulin EC50. At that exposure, tubulin binding saturates. Tubulin depolymerisation in xenograft models has been documented since the early 2000s in lung (Mukhopadhyay 2002, Clin Cancer Res) and glioblastoma (Bai 2011, Neuro-Oncology).

This is a real cytostatic mechanism, exposure-saturated at the 500 mg dose. It is also slow: the arrested-cell fraction builds gradually over weeks of continuous exposure.

What the composed model says about the combination

In the composed body model, ivermectin (immune + PAK1 arms) and mebendazole (tubulin arm) are wired as separate kernels, each reading from its own PK module and coupling into a shared tumour state. Run forward for 90 days at the 25 mg / 500 mg combination described in the recent report, the model gives a cumulative drug-attributable tumour-control factor in the low single-digit range — direction positive, magnitude modest.

That is not nothing. It is several orders of magnitude below what a > 80% clinical benefit rate would require. The gap between the mechanism-derived prediction and the reported clinical benefit rate is the part the model cannot close from drug effect alone.

What would close the gap

The observational cohort report is explicit about what was happening alongside the protocol: roughly a quarter of patients on concurrent chemotherapy, a fifth on radiation, a fifth with surgery during follow-up, and around half on supplements with dietary changes. The cohort is self-selected and health-motivated, and the reported outcomes are patient-reported, not centrally adjudicated.

The composed model’s answer to this is not “the headline is wrong”; it is “the headline is computed on a sample where the drug effect and the standard-of-care effect cannot be separated.” Each of those concomitant interventions has its own documented anti-tumour effect size. The drug-attributable share of the observed clinical benefit rate cannot be recovered from the cohort as designed.

This is the same shape of problem as the one we wrote about for the anti-amyloid Alzheimer drugs — the data needed to distinguish the hypotheses does not yet exist, and the right next step is the trial that would produce it, not a stronger reading of the data we already have.

The trial that would settle it

A randomised design with two features would settle most of the question. First, randomisation against best supportive care, with concomitant chemotherapy and radiation balanced across arms and recorded as covariates rather than excluded — denying access to those would be neither feasible nor ethical in the populations the report describes. Second, centrally adjudicated radiologic response (RECIST 1.1) rather than patient-reported outcomes, with a pre-specified primary endpoint and an intention-to-treat analysis that includes the patients who would otherwise drop out.

The composed model is a useful instrument going into such a trial. It can be calibrated to the trial’s PK arm and used to pre-specify the magnitude of effect attributable to each mechanism — the expected immune-arm contribution at the trial dose, the expected mitotic-arrest contribution, the expected combined effect. A trial where the observed effect falls on the model’s central prediction is straightforwardly informative; one where it diverges is informative in a different way, and the model is a useful place to start asking why.

What this is not

Two clarifications, in the spirit of the methodology pieces this blog tries to write.

This is not a clinical recommendation. The model gives direction-of-effect and approximate magnitude; it does not select patients, time doses, or substitute for oncologic care. The “we are not clinicians” line in the disclaimer at the bottom of every post is real.

It is also not a verdict on drug-repurposing programmes in oncology in general. Mebendazole has run a multi-centre clinical study in glioma; ivermectin has open clinical work on breast cancer in combination with checkpoint blockade. The mechanism work in this space is serious, and the way to assess whether the combination is worth pursuing as a regimen is the trial that produces an interpretable answer — not the read of the data we have now.

Where the work lives

The mechanism kernels for both drugs (ivermectin: immune + PAK1 arms; mebendazole: tubulin arm) are wired into the composed body model with paper-anchored PK modules (Schmith 2020, Dawson 1985) and an in vitro–to–in vivo extrapolation chain that brings the xenograft EC50s onto plasma- and tissue-concentration axes the PK module produces. The dose-sweep that produced the mechanism-attribution numbers in this post is reproducible from the same kernels at the doses cited; the paper anchors are listed below and the audit lives alongside this post in our internal repository.

Sources. Schmith VD et al., Clin Pharmacol Ther 2020 (ivermectin PK). Draganov D et al., npj Breast Cancer 2020 (immune arm). Dou Q et al., Cancer Research 2016 (PAK1 / autophagy). González Canga A et al., AAPS J 2008 (log P, tissue partition). Lifschitz A et al., Vet Parasitol 2000 (tissue distribution). Navarro M et al., Clin Microbiol Infect 2020 (tolerability meta-analysis). Muñoz J et al., J Antimicrob Chemother 2018 (high-dose PK/AE). Dawson M & Watson TR, Br J Clin Pharmacol 1985 (mebendazole PK). Mukhopadhyay T et al., Clin Cancer Res 2002 (mebendazole xenograft). Bai R-Y et al., Neuro-Oncology 2011 (mebendazole glioblastoma). All anchors verifiable from public sources; the composed model and sweep are reproducible from our internal repository on request. The methodology page is at /our-method/.