The LD50, lethal dose 50 percent, is the amount of a substance required to kill 50 percent of a test population under specified experimental conditions. It is a standard measure of acute toxicity used to compare compounds and establish hazard classifications. For grayanotoxin, published LD50 data exist in rodent models across multiple isoforms and routes of administration.
The question is common and reasonable. The useful answer requires specifying four things: which isoform (GTX I, II, or III), which species (mouse or rat), which route of administration (intraperitoneal injection or oral), and what extrapolation caveats apply before that number means anything for a human consuming honey. All four matter, and this article addresses each.
It also addresses the separate question of what the human clinical case record tells us about lethality thresholds, which is, for practical purposes, the more relevant dataset.
The Published LD50 Data: What the Animal Studies Show
Grayanotoxin LD50 data come primarily from Japanese and European laboratory studies from the 1970s through the 2000s, using intraperitoneal (IP) injection as the primary route. The most frequently cited values come from Kimura (1983) for GTX I in mice, and Teramoto and colleagues for GTX III in rodent models.
| Compound | Species | Route | LD50 (mg/kg) | Notes |
| GTX I | Mouse | IP | ~1.3 | Primary toxic isoform. Most frequently cited acute toxicity reference. Source: Kimura 1983. |
| GTX I | Rat | IP | ~2.5-3.0 | Higher than the mouse IP value. Species differences significant. |
| GTX I | Rat | Oral | ~6.0 | Substantially higher than IP due to reduced bioavailability and first-pass metabolism. |
| GTX III | Rat | IP | ~1.58 | Primary arrhythmogenic isoform. Slightly higher potency than GTX I in some rat models. Source: Teramoto et al. |
| GTX III | Mouse | IP | ~1.0-1.2 | Consistent with rat data. Lower LD50 than GTX I in some studies. |
| GTX II | Mouse | IP | >200 | Markedly less toxic than GTX I and III. Not a primary clinical concern at honey matrix concentrations. |
| GTX mix (Andromedotoxin) | Mouse | IP | ~1.5-2.0 | Historical designation for crude GTX preparations. Reflects a mixed isoform profile. |
Reading the table correctly – two critical observations
First, the difference between the IP and oral LD50 for GTX I in rats, approximately 2.5 to 3.0 mg/kg for IP versus approximately 6.0 mg/kg for oral, reflects a fundamental pharmacokinetic reality. Direct injection into the peritoneal cavity bypasses gastrointestinal absorption, first-pass hepatic metabolism, and the diluting effect of stomach contents. IP injection delivers the compound directly to the systemic circulation at high bioavailability. Oral consumption from a honey matrix is a different exposure pathway entirely.
Second, GTX II at greater than 200 mg/kg illustrates that not all grayanotoxin isoforms carry equivalent toxicity. The three isoforms share the Site 2 voltage-gated sodium channel binding mechanism but differ substantially in binding affinity and potency. A COA reporting total GTX without distinguishing isoforms combines values with very different clinical weights.
Why IP Injection LD50 Values Cannot Be Applied to Oral Honey Consumption
The most common error in the discussion of grayanotoxin toxicity is applying IP injection LD50 values to scenarios involving oral consumption of honey. The two exposure routes are pharmacokinetically incomparable.
Bioavailability
Intraperitoneal injection delivers the compound to the highly vascularised peritoneal cavity, from which absorption into systemic circulation is rapid and near-complete. Oral ingestion requires the compound to survive the gastric environment, cross the intestinal mucosa, survive first-pass hepatic metabolism, and enter systemic circulation in whatever fraction survives those steps. For many lipophilic compounds, oral bioavailability is a fraction of IP bioavailability.
No formally published oral bioavailability study for GTX I or GTX III from a honey matrix in humans exists. Extrapolating the IP LD50 to an oral consumption scenario requires assuming near-complete bioavailability, an assumption that is almost certainly incorrect and that systematically overestimates the toxicity of oral consumption relative to IP injection.
The honey matrix effect
GTX in mad honey is not a purified compound in a neutral vehicle. It is embedded in a complex carbohydrate-rich viscous matrix containing water, sugars, organic acids, proteins, and other secondary metabolites. The absorption kinetics of GTX from this matrix differ from those of purified GTX in aqueous solution. The matrix may slow gastric emptying, affect intestinal absorption, and alter the time-to-peak-blood-concentration profile.
This is clinically relevant: the Aryal (2025) review cites blood concentration data for GTX I (2.52 to 4.55 ng/mL) and GTX III (17.5 to 27.3 ng/mL) associated with clinically significant hypotension in a six-patient dataset. These values are far below what would be predicted from IP LD50 data applied naively, consistent with substantially lower oral bioavailability than IP injection.
Species differences
Mouse and rat pharmacokinetics, receptor densities, cardiac conduction system properties, and metabolic enzyme expression differ from each other and from humans. Species extrapolation from rodent IP LD50 to human oral LD50 involves two separate and compounding uncertainties: species difference and route difference. Any figure attempting to state a human LD50 from these data is combining multiple unstated assumptions into a single number presented with more precision than the underlying evidence warrants.
Why No Human LD50 Exists or Can Be Established
No controlled human LD50 study for grayanotoxin exists or could ethically be conducted. Establishing an LD50 requires exposing a population to graduated doses and observing mortality at each level. This is acceptable in rodent models as part of standard toxicological risk assessment. It is not acceptable in humans.
What the human evidence base contains instead is a different kind of data: the clinical case record. Hundreds of documented cases of grayanotoxin poisoning from Turkey and Nepal, with intake amounts estimated retrospectively and clinical outcomes recorded, provide an upper-bound dataset for human acute toxicity. The record documents what amounts caused cardiovascular presentations requiring hospital treatment. It does not document what amounts were lethal in untreated cases, because almost all documented cases received emergency treatment.
What the Clinical Case Record Says About Human Lethality
Documented fatal outcomes from standard Rhododendron-derived grayanotoxin poisoning are rare in the modern case literature. The Salici and Atayoglu (2015) systematic review of 1,199 Turkish cases found a very low mortality rate in patients who received emergency treatment. The fatal cases documented in the literature are almost exclusively associated with delayed presentation to medical care, very high intake amounts, or severe pre-existing cardiac conditions that prevented adequate cardiovascular compensation.
The fatality cases that do appear in the literature do not cluster at a consistent intake amount that would suggest a pharmacological LD50 threshold. They involve patients whose presentations escalated to complete AV block or severe hemodynamic compromise before adequate treatment was available. Clinical lethality is therefore better characterized as a treatment access and response time variable than as a pharmacological dose threshold.
| The T. hypoglaucum 25.8% fatality figure does not apply here A 25.8% mortality figure sometimes cited in connection with honey poisoning refers to a specific cluster of deaths attributed to honey from Tripterygium hypoglaucum in China, a botanically distinct source with a different compound profile from standard Rhododendron-derived grayanotoxin. This figure must never be cited in the context of GTX I or GTX III poisoning from Rhododendron honey. The two datasets are not comparable and must never appear in the same epidemiological frame. |
Translating the LD50 Data to a Honey Consumption Context
A rough order-of-magnitude translation is useful, not as a precise risk threshold, but as a way of understanding the scale of the difference between experimental toxicology and real-world consumption.
A 70 kg human consuming the approximate oral LD50 dose of GTX I in rats (approximately 6 mg/kg) would consume approximately 420 mg of purified GTX I. At the high end of the Ahn et al. (2022) concentration range, 64.86 micrograms per gram, achieving 420 mg of GTX I would require consuming approximately 6,480 grams of that batch. That is not a realistic consumption amount.
At a more typical documented intoxication concentration of 10 to 20 micrograms per gram, reaching 420 mg would require 21,000 to 42,000 grams. The clinical case record documents intoxication at 15 to 30 grams from real-world batches. That discrepancy is consistent with: the oral-to-IP bioavailability ratio being well below 1; GTX producing clinically significant cardiovascular toxicity far below the LD50; and the rat oral LD50 not translating directly to humans. GTX produces life-threatening cardiovascular effects at blood concentrations far below those that would be needed to produce 50 percent lethality in a population.
The Isoform Question for Risk Assessment
Different batches of mad honey contain different ratios of GTX I, GTX II, and GTX III. A COA reporting only total grayanotoxin aggregate compounds with substantially different LD50 values (GTX I approximately 1.3 mg/kg versus GTX II greater than 200 mg/kg in mice via IP) into a single figure. For comprehensive risk characterization, isoform-specific quantification matters.
In practice, GTX I and GTX III are the primary clinical concern because they are both present in significant quantities in active mad honey and both have similar IP LD50 values in the 1 to 3 mg/kg range. GTX II, despite often being detected alongside them, is not a primary acute toxicity driver at the concentrations found in honey.
The Ahn et al. (2022) study reported all three isoforms simultaneously using LC-MS/MS, which is why that methodology is the gold standard for COA purposes. It distinguishes the clinically relevant isoforms from the less toxic ones and allows a more accurate risk characterization than total GTX detection alone.
What We Don’t Know Yet
No oral bioavailability study for GTX I or GTX III from a honey matrix in any species exists in the peer-reviewed literature. The oral LD50 values cited above come from aqueous or solvent vehicle preparations, not from honey. The actual bioavailability from honey may differ due to the matrix effects described above.
No human pharmacokinetic study exists. The absorption, distribution, metabolism, and elimination parameters for GTX in humans are not established. The interaction between GTX isoforms in mixed exposure, as occurs in real honey consumption, is not systematically studied. Most LD50 studies used single purified isoforms.
Summary
Grayanotoxin LD50 data come from rodent models using intraperitoneal injection. GTX I has an IP LD50 of approximately 1.3 mg/kg in mice. GTX III shows similar values. GTX II is markedly less toxic with an IP LD50 exceeding 200 mg/kg. These values cannot be directly applied to oral honey consumption: IP injection bypasses gastrointestinal absorption and first-pass metabolism; the honey matrix affects absorption kinetics; and species extrapolation introduces further uncertainty. The oral LD50 is substantially higher than the IP LD50 even in rodents.
No human LD50 data exists, and none can be ethically established. The clinical case record, showing life-threatening cardiovascular events at 15 to 30 grams of honey at real-world concentrations, is the more relevant reference point for human risk. Fatal outcomes from standard GTX poisoning with adequate treatment are rare. The clinical risk is AV block and hemodynamic compromise that becomes lethal without treatment, not acute compound toxicity at a pharmacological LD50 threshold.
