Mad honey is often discussed as if it were one chemically stable product. This paper shows why that is misleading.
In 2022, researchers developed and validated a liquid chromatography–tandem mass spectrometry (LC-MS/MS) method to quantify grayanotoxin I (GTX I) and grayanotoxin III (GTX III) in honey, then applied that method to 60 mad honey samples from Nepal.
The result was not just a technical validation paper. It was also a clear demonstration that grayanotoxin content can vary sharply from sample to sample. GTX I and GTX III were detected together in 33 of the 60 samples, with GTX I ranging from 0.75 to 64.86 µg/g and GTX III ranging from 0.25 to 63.99 µg/g. The authors also reported no correlation between the two toxins.
That combination of findings matters. It means mad honey can be measured with a robust analytical method, but it cannot be assumed to have a uniform toxin profile from one batch to the next. For a research-focused knowledge base, that is the central point of the study.
Study at a Glance
- Paper: Determination of Grayanotoxin I and Grayanotoxin III in mad honey from Nepal using liquid chromatography-tandem mass spectrometry
- Authors: Su Youn Ahn, Suncheun Kim, Hwangeui Cho
- Journal: Analytical Science & Technology
- Year: 2022
- Study type: analytical method development, validation, and applied testing of confiscated Nepal mad honey samples.
Why This Study Matters
This paper is important for two separate reasons.
First, it addresses a laboratory problem. Honey is a difficult matrix to analyze, especially when the compounds of interest are present in small amounts, and the sample itself is dominated by sugars and other interfering components. The authors argue that older methods were either too narrow, too cumbersome, or not well adapted to the matrix.
Some earlier workflows focused mainly on GTX III, required relatively large sample amounts and solvent volumes, or caused practical issues such as ion suppression and repeated instrument cleaning. The authors therefore set out to build a matrix-specific method for honey rather than rely on methods designed for other sample types.
Second, it addresses a product-level problem. Mad honey is frequently treated in public discussion as though it has one recognizable potency profile. This paper undermines that assumption. Once the method was applied to real samples, the results showed large variability across the honey analyzed.
That makes the study useful not only to toxicologists and analytical chemists, but also to anyone interested in batch testing, safety evaluation, or the basic question of whether “mad honey” is chemically consistent.
Why Honey Is a Difficult Matrix for Grayanotoxin Testing
One of the strongest sections of the paper is the explanation of why honey is analytically troublesome. The authors identify the very high sugar content of honey as the main challenge. Those sugars can reduce selectivity and sensitivity during analysis, and if they are not adequately removed during sample preparation, they can contaminate the LC column and the mass spectrometry interface.
In other words, the problem is not only detecting grayanotoxins; it is detecting them cleanly in a sticky, interference-heavy sample.
This is why the sample preparation step became central to the method. The researchers tested different extraction strategies to improve both selectivity and sensitivity. They evaluated liquid-liquid extraction using several organic solvents, but found that this approach required frequent instrument cleaning.
They also examined a salting-out method using ammonium sulfate, but this caused peak tailing in the chromatograms. These were not minor inconveniences. They were practical signs that the extraction process still allowed too much matrix interference or poor chromatographic behavior.
The method they settled on was solid-phase extraction (SPE). According to the paper, SPE produced clean chromatograms, effective removal of interferences, and adequate extraction recoveries for the analytes. The authors also note a practical detail that makes the method feel less abstract: diluting honey with methanol increased viscosity and made cartridge loading difficult, so the samples were instead diluted with water and mixed thoroughly until the honey fully dissolved.
Shaking worked better than sonication or magnetic stirring for producing a well-mixed sample. After loading, the cartridge was washed with water to remove polar substances and then eluted with methanol. The overall result was a cleaner analytical sample with less introduction of highly polar sugars into the LC-MS system.
That section matters because it shows this was not just an academic protocol. It was a method built around the actual physical behavior of honey.
Method Overview: How the Researchers Measured GTX I and GTX III
The final analytical setup used LC-MS/MS in multiple-reaction monitoring mode with positive electrospray ionization. The compounds were separated on a Phenomenex Kinetex Biphenyl column, and the mobile phases were based on 0.5% acetic acid in water and 0.5% acetic acid in methanol.
The authors report that methanol performed better than acetonitrile as the organic phase, giving higher signal response, better resolution, and finer peak shape. They also tested several additives and found that 0.5% acetic acid improved peak intensity and symmetry.
The chromatographic choices were not arbitrary. The authors compared several columns and found that both Kinetex Biphenyl and Kinetex C18 gave similar separation efficiency, but the biphenyl column produced slightly better peak intensities, peak shape, and resolution, so it was selected for the final method.
For quantification, the method used specific precursor-to-product ion transitions for GTX I and GTX III, along with clindamycin as the internal standard. Calibration standards were prepared in blank honey across a range of 0.25, 0.5, 1, 5, 10, and 100 µg/g, with QC samples at 0.25, 1, 10, and 80 µg/g. Honey aliquots of 0.2 g were diluted, extracted via SPE, evaporated, reconstituted, and injected into the LC-MS/MS system.
The important thing for a non-specialist reader is that this was a matrix-matched quantitative assay. The standards and QC samples were built in honey, not in a simplified solvent-only environment, which is important when the matrix itself is known to create analytical difficulties.
Validation Results: What the Method Could Reliably Do
The paper validated the method for selectivity, lower limit of quantification, linearity, accuracy, precision, matrix effects, dilution effect, and stability, using International Council for Harmonisation guidance as the framework.
The selectivity testing used six independent blank honey samples from different manufacturers. The authors report that no significant interference peaks were observed at the elution times of the analytes or internal standard. That matters because a method can only be trusted if the signal being measured is not being confused with background compounds naturally present in honey.
The calibration range for both GTX I and GTX III was 0.25 to 100 µg/g, with very strong linearity across that range. The reported regression coefficients were 0.9995 for GTX I and 0.9997 for GTX III. The lower limit of quantification for both toxins was 0.25 µg/g, with accuracy ranging from 90.96 to 114.23%.
Precision and accuracy were also strong across the validation levels. Intra-day and inter-day accuracy ranged from 94.33 to 109.30% for GTX I and from 95.43 to 114.01% for GTX III. Intra-day and inter-day precision remained below 10.59% and 8.19%, respectively. The authors interpret these results as evidence that the method had sufficient precision and accuracy for quantitative measurement in honey.
Extraction recovery was consistently high. GTX I recovery ranged from 87.44 to 92.58%, while GTX III recovery ranged from 86.98 to 90.74% across low, medium, and high QC concentrations. Matrix effects ranged from 86.14 to 91.81% for GTX I and 85.57 to 92.53% for GTX III, which the authors describe as negligible on the ionization of the analytes and internal standard in this method.
The paper also reports acceptable dilution performance and stability. Samples diluted twofold and eightfold with blank honey remained within acceptable accuracy and precision ranges, and the analytes were stable during short-term storage, long-term storage, and post-preparation storage in the autosampler. In practical terms, the authors conclude that the analytes could be processed and stored under routine laboratory conditions.
All of this matters because the batch-variation findings carry more weight when they come from a method that was carefully validated rather than a loose screening approach.
What the 60 Nepal Samples Actually Showed
Once the method was developed and validated, the authors applied it to 60 confiscated mad honey samples brought from Nepal. This is where the paper shifts from method validation into a direct statement about real-world variability.
GTX I and GTX III were detected together in 33 out of the 60 samples. Among those 33 positives, GTX I concentrations ranged from 0.75 to 64.86 µg/g, with a mean content of 25.07 µg/g. GTX III concentrations ranged from 0.25 to 63.99 µg/g, with a mean content of 17.05 µg/g. The authors note that this corresponds to up to an 86-fold difference in GTX I and a 255-fold difference in GTX III among the mad honey analyzed.
That is not a small degree of drift around a common baseline. It is a wide chemical spread.
The sample table makes that concrete. Some samples cluster in the high range for GTX I: H10 reached 64.86 µg/g, H8 measured 57.53 µg/g, H2 measured 54.53 µg/g, and H9 measured 49.58 µg/g. Other samples sat much lower: H21 measured 1.64 µg/g, H22 1.32 µg/g, H25 1.20 µg/g, and H32 just 0.75 µg/g.
GTX III had a different pattern. Some samples that were only moderate in GTX I were very high in GTX III. H28, for example, had 36.05 µg/g of GTX I but 63.99 µg/g of GTX III. H11 had 26.64 µg/g of GTX I but 42.97 µg/g of GTX III. H7 had 19.18 µg/g of GTX I and 42.09 µg/g of GTX III. At the low end, H21 through H27 included several samples clustered close to 0.25 to 0.42 µg/g for GTX III.
This is one of the reasons the paper is so useful. It does not merely say that variability exists. It shows that variability in a sample-by-sample table, where the spread is impossible to miss.
What “No Correlation” Means in Practice
The authors report that there was no correlation between the concentrations of GTX I and GTX III.
That line matters more than it may seem at first glance. It means that the two toxins do not rise and fall together in a predictable way across the samples analyzed. A jar high in GTX I is not automatically high in GTX III, and a jar low in one is not automatically low in the other.
This finding complicates any oversimplified approach to testing or regulation. The paper explicitly notes that only GTX III had been regulated under the relevant framework discussed by the authors, but their results suggest that the nonexistence of both GTX I and GTX III should be evaluated for food safety. In other words, focusing on just one toxin may leave part of the picture unmeasured.
For an indexed research site, that is one of the strongest takeaways in the entire paper: batch variation is not just about “more” or “less” toxin overall. It is also about differing toxin profiles within the same product category.
Why the Authors Think This Variability Exists
The paper is cautious about causes, but it does not ignore them.
The authors suggest that the huge variability in grayanotoxin data among mad honey samples might occur due to production season, region, or processing procedure. They also note that mad honey produced in the spring is known to be more toxic and to sometimes contain higher grayanotoxin contents than honey made in other seasons.
That point is important because it prevents the paper from being read too narrowly. The study does not claim to prove exactly why each sample differed. But it does frame the variation as plausible in a natural product shaped by floral source, time of harvest, geography, and handling. In that sense, the data fit a larger pattern: the chemistry of mad honey is not fixed at the category level.
What This Study Does Not Show
As useful as this paper is, its scope should be kept clear.
It does not provide a universal dosing rule for consumers. It does not prove how any specific jar will affect any specific person. It does not show that toxin concentration alone determines every clinical outcome.
The authors note that the amount of mad honey causing poisoning has been reported as between 5 and 30 g, and that poisoning severity generally depends on the amount ingested. At the same time, they argue that grayanotoxin content in the honey could be an important determinant of the cause and severity of poisoning.
So the paper supports a careful middle position. Amount consumed matters. Toxin concentration matters. But this is not a consumer guide. It is an analytical paper showing that toxin measurement and batch variability are central to understanding the risk.
Why This Study Matters for Safety and Future Research
The closing sections of the paper move beyond laboratory performance and into practical relevance.
The authors argue that more careful examination and controls on mad honey are needed because of the possibility of grayanotoxin poisoning. They also state that the determination of grayanotoxins in mad honey through qualitative and quantitative analysis is meaningful for preventing and proving grayanotoxin poisoning.
In the conclusion, they summarize the significance of the assay in equally direct terms. The method provided good linearity, precision, accuracy, and sensitivity for quantitative measurement in honey. SPE sample treatment produced clean samples with enhanced selectivity and high extraction recovery. Most importantly, the validated method can be used for the identification and quantification of grayanotoxins in honey suspected of containing them.
The authors then make the regulatory implication explicit: since only GTX III had been regulated under the related framework, their result suggests that the absence of both GTX I and GTX III should be considered in food safety evaluation. They also say the assay may help prevent intake of mad honey as a potential poison and help confirm grayanotoxin poisoning for forensic and clinical studies.
That gives the paper a wider relevance than a narrow method note. It is a paper about measurement, but also about why measurement changes the conversation.
Bottom Line
This 2022 study does more than show that grayanotoxins can be detected in mad honey. It provides a validated LC-MS/MS method designed specifically for a difficult honey matrix, and it demonstrates that mad honey from Nepal can vary substantially in GTX I and GTX III content from one sample to another.
The method achieved strong linearity, acceptable precision and accuracy, high recovery, negligible matrix effects, and practical stability under routine laboratory conditions. When applied to 60 real samples, it found GTX I and GTX III together in 33 of them, with broad ranges and no correlation between the two toxins.
The core conclusion is straightforward: Mad honey is measurable, but it is not uniform. Any serious discussion of testing, safety, or batch comparison has to begin there.
